Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Empirical modelling of windthrow risk and hazard mapping using Geographic Information Systems Lanquaye, Clayfield Odarkor 2003

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

Item Metadata

Download

Media
831-ubc_2003-0147.pdf [ 4.72MB ]
Metadata
JSON: 831-1.0075094.json
JSON-LD: 831-1.0075094-ld.json
RDF/XML (Pretty): 831-1.0075094-rdf.xml
RDF/JSON: 831-1.0075094-rdf.json
Turtle: 831-1.0075094-turtle.txt
N-Triples: 831-1.0075094-rdf-ntriples.txt
Original Record: 831-1.0075094-source.json
Full Text
831-1.0075094-fulltext.txt
Citation
831-1.0075094.ris

Full Text

EMPIRICAL MODELLING OF WINDTHROW RISK AND HAZARD MAPPING USING GEOGRAPHIC INFORMATION SYSTEMS CLAYFIELD ODARKOR LANQUAYE B. Sc. Natural Resources Management, The University of Science and Technology, Kumasi, Ghana, 1999 A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENT FOR THE D E G R E E OF MASTER OF SCIENCE in THE FACULTY OF GRADUATE STUDIES THE FACULTY OF FORESTRY We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA 2 0 0 3 - 0 3 - 0 3 © Clayfield Odarkor Lanquaye, 2003 UBC Rare Books and Special Collections - Thesis Authorisation Form Page 1 of 1 In presenting t h i s thesis i n p a r t i a l f u l f i l m e n t of the requirements for an advanced degree at the Un i v e r s i t y of B r i t i s h Columbia, I agree that the Library s h a l l make i t f r e e l y a v a i l a b l e for reference and study. I further agree that permission for extensive copying of t h i s thesis for sch o l a r l y purposes may be granted by the head of my department or by his or her representatives. I t i s understood that copying or p u b l i c a t i o n of t h i s t h e s i s f o r f i n a n c i a l gain s h a l l not be allowed without my written permission. Department of The U n i v e r s i t y of B r i t i s h Columbia Vancouver, Canada http://www.library.ubc.ca/spcoll/thesauth.html 3/4/2003 ABSTRACT There has been a history of wind damage (windthrow) in Weyerhaeuser's North Island Timberlands (NIT) north of Campbell River on Vancouver Island, British Columbia. Identification of the location and type of stands most at risk is needed to target mitigative measures. This study investigated the relationship between windthrow damage along cutblock edges and stand, ecosystem, management, wind and topographic variables. Windthrow occurrence along cutblock boundaries was mapped using aerial photographs. Using ArcView Geographic Information System (GIS), a total of 22,304 forested segments were obtained and used to study the relationship between cutblock edge windthrow and other stand level variables within NIT. Damage declined from 13% of segments within the first 25m of cutblock boundaries to 1 % of segments in the band between 50 and 75m from the cutblock boundary. Western redcedar (Thuja plicata Donn) dominated stands suffered the least loss from windthrow compared to those dominated by other species. The proportion of segments damaged increased with topographic exposure, mean annual windspeeds, boundary exposure to the south, soil fertility, block size and stocking. Logistic regression models were fit using stand, site, ecosystem, wind and management variables to predict probability of damage. The best-fit models had a predictive accuracy of 68-72%. One of these models was selected to produce a windthrow hazard map for the NIT. A comparison between the resultant model and one developed in an earlier study 75 km to the north revealed that empirical models are reasonably portable within this area of Vancouver Island. TABLE OF CONTENTS Page ABSTRACT ii TABLE OF CONTENTS iii LIST OF TABLES vi LIST OF FIGURES viii LIST OF APPENDICES xi ACKNOWLEDGEMENTS xiii 1 INTRODUCTION 1 1.1 WHY ASSESS WINDTHROW RISK? 1 1.2 SCOPE AND OBJECTIVES 3 1.3 APPROACH 4 1.4 THESIS OUTLINE 4 2 LITERATURE REVIEW 6 2.1 MECHANICS OF WINDTHROW 6 2.2 FACTORS AFFECTING WINDTHROW DAMAGE 9 2.2.1 Individual Tree and Stand Level Characteristics 10 2.2.2 . Soil Characteristics 16 2.2.3 Topographic Characteristics 18 2.2.4 Management Variables 20 2.2.5 Climatic Conditions 21 iii 2.2.6 British Columbia Coastal Wind Regime 21 2.2.7 Windthrow Risk Factors Reported in Coastal BC and Alaska 23 2.3 WINDTHROW RISK ASSESSMENT APPROACHES 29 2.3.1 Observational Approach 29 2.3.2 The Mechanistic Approach.: 30 2.3.3 The Empirical Approach 31 2.4 RESERVATIONS ABOUT ASSESSMENT APPROACHES 32 2.5 USE OF GEOGRAPHIC INFORMATION SYSTEMS IN EMPIRICAL STUDIES 33 2.6 GIS-BASED WINDTHROW STUDIES 34 2.6.1 European Studies 34 2.6.2 Coastal North American Studies 35 2.7 SUMMARY, RESEARCH QUESTIONS, AND HYPOTHESES 36 3 MATERIALS AND METHODS 39 3.1 DESCRIPTION OF STUDY AREA.. 39 3.2 INFORMATION SOURCES AND DATASET CREATION 42 3.2.1 Information Assembly and Data Conversion 42 3.2.2 Correcting Mapping Inconsistencies 43 3.2.3 Creation of Sample Units 45 3.2.4 Windthrow Detection and Mapping 45 3.2.5 Determination of Independent Variables 47 3.2.6 Construction of Segment Database 53 3.2.7 Correlation 53 3.2.8 Spatial Correlation 54 3.2.9 Creation of Datasets With Spatially Independent Observations 55 3.2.10 Determination of Dependent Variables 56 3.2.11 Contingency Tables 57 3.2.12 Model Fitting Procedures 58 iv 3.2.13 Model Comparison: Robustness and Portability 63 3.2.14 Windthrow Hazard Map Creation 64 4 RESULTS 65 4.1 Exploration of Independent Variables: Ranges and Means of x-Variables 65 4.2 CORRELATION BETWEEN SELECTED VARIABLES 66 4.3 CONTINGENCY TABLES 67 4.4 MODELS 73 4.4.1 Comparing Port McNeill and NIT Models 79 4.4.2 Windthrow Hazard Map Creation 82 5 DISCUSSION 83 6 CONCLUSIONS 91 7 RECOMMENDATIONS 94 8 REFERENCES 96 APPENDIX 1 110 APPENDIX II 112 APPENDIX III 113 APPENDIX IV 115 APPENDIX V 117 APPENDIX VI 118 APPENDIX Vll 120 APPENDIX Vlll 122 APPENDIX IX 123 APPENDIX X 125 v LIST OF TABLES Table Page 1. Peak hourly windspeeds (two minute mean) and their directions for Chatham Point station near study area for the years 1964-1992 24 2. Peak hourly windspeeds and gust return periods in years for selected stations 25 3. Summary of risk factors in Coastal BC and Alaska by various authors 27 4. A summary of the climate for each of the 4 major subzones 41 5. Number of segments (n) used for statistical analysis 56 6. Summary of response variables and procedures used to create them 57 7. Values of key variables for the study segments (n = 22,304) 65 8. Correlations between selected independent variables (n = 6715) 66 9. Most frequent wind directions for the 6715 segments according to BC Hydro wind data...67 10. Variables and coefficients in initial logistic models (dependent variable is WTCN ) 74 11. Variables and coefficients in the best-fit logistic regression models for combined datasets - overall models 75 12. Percent of correct predictions for individual segments using test dataset #2, n = 1425... .76 13. Percent of actual damaged segments and predicted probability of damage using test dataset #2 sorted by predicted probability of damage and divided into 10 groups, n = 1425 76 14. Effect of changing area damaged for classifying a segment as 'damaged' on variable coefficients for all independent variables, n = 6715 78 v i Percent of actual damaged segments and the probability of damage predicted by the Port McNeill model sorted by predicted probability of damage and divided into 10 groups (simple model), n = 6715 LIST OF FIGURES Figure Page 1. Forces acting on a tree (redrawn from Peltola and Kellomaki 1993) 6 2. (a) Map of coastal British Columbia showing study site; (b) Detailed map of North Island Timberlands (NIT) 40 3. (a) Example of discrepancies in forest cover and logging history boundaries. (b)Example of logging history and forest cover layers corrected to align with forest boundaries in ortho-rectified aerial photographs 44 4. Example of cutblock edge windthrow on aerial photograph along with ArcView table showing percent canopy lost by windthrow polygon 46 5. Measures of edge exposure due to harvesting. DIREX, sum of number of openings in eight cardinal directions. SCORE, sum of distance within opening for eight cardinal directions; 49 6. Representation of buffer distances around a cutblock within which sample points are located. B^ B2 and B3 = buffer distance 25, 50 and 75 m respectively 51 7. Site index classes and the distribution of species-specific site index data. Class I represents the highest productivity sites and class IV the lowest. Box plots contain 84% of sample. Ba = amabilis fir, Cw = western redcedar, Fd = Douglas fir, Hw = western hemlock and Ss = Sitka spruce. (From Greene and Klinka 1994) 52 8. Semivariance versus the number of 25 m segments or distance for: (a) crown closure lost; (b) profile curvature; (c) elevation; and (d) aspect 55 Vll l 9. Proportion of damaged segments for (a) Buffer, 25 m, 50 m and 75m, and (b) BRG, bearing at right angles to boundary inward towards block for buffer 25 m (°), n = 6,715. # = number of observations; (bar) = proportion; (•) = number of segments 68 10. Proportion of damaged segments for classes of major independent variables, n = 6,715. # = number of observations; (bar) = proportion; (•) = number of segments, (a)- (c) for 25m segments and (a1) - (c1) for 50 m segments. (a/a1) Species; B = amabilis fir; C = western redcedar; CY = yellow cypress; F = Douglas fir; H = hemlock. (b/b1) Direx = number of exposed directions, 0 - 7 . (c/c1) = block size; S, small = 15 - 50 ha; M, medium =150 - 500ha; L, large = 500 ha+ 69 11. Percent crown canopy lost by species in damaged segments in (a) 25 m buffer and (b) 50 m buffer. B = true, amabilis and grand fir; C = western redcedar; CY = yellow cypress; F = Douglas fir; H = western and mountain hemlock... 70 12. Proportion of damaged segments for classes of major independent variables, n = 6,715. # = number of observations; (bar) = proportion; (•) = number of segments, (a) - (c) for 25m segments and (a1) - (c1) for 50m segments. (a/a1) = BEC_subzone_variant; CWHvml, CWHvm2, CWHxm2, MH mm1. (b/b1) Stock; I = immature conifers; M>350 = mature conifers with volume >350 m3; M<350 = mature conifers with volume < 50 m3. (c/c1) Nutrients; poor and rich 71 13. Proportion of damaged segments for classes of major independent variables, n = 6,715. # = number of observations; (bar) = proportion; (•)= number of segments, (a)- (c) for 25m segments and (a1) - (c1) for 50m segments. (a/a1) moisture; dry, fair and wet. (b/b1) = surface material, C&F, colluvial and fluvial; M.morainal; O, organic. (c/c1)mwspeed, mean annual windspeed (m/s) 72 14. Predicted probability of damage versus percentage of segments actually damaged for dataset #2 sorted by predicted probability of damage and divided into 10 groups, with 1:1 line for a) WTCN b) WTT20 and c) WTP60 77 15. Port McNeill model Predicted probability of damage (simple model) versus percentage of segments actually damaged for combined datasets sorted by predicted probability of damage and divided into 10 groups, with 1:1 line for (a) WTCIOand (b) WTC50 81 16. Number of 100 m by 100 m cells for forested portions of NIT occupied by each damage probability range 82 x LIST OF APPENDICES APPENDIX I. SUMMARY OF PROCEDURES USED TO BUILD THE WINDTHROW RISK MODEL 109 Table I. A. Windthrow risk model procedures 109 APPENDIX II. DEFINITION OF INDEPENDENT VARIABLES 111 Table II.A. Summary of independent variables 111 APPENDIX III. CREATING CLASS VARIABLES 113 APPENDIX IV. TOPEX CALCULATIONS 115 APPENDIX V. NUMBER OF DAMAGED SEGMENTS FOR COMBINATIONS OF SEGMENT AREA AND CANOPY LOSS 117 Table V. A. Number of damaged segments for combinations of segment area and canopy loss (total n = 6715 with damage on 874 segments) 117 APPENDIX VI. I NITIAL MODEL VARIABLES AND COEFFICIENTS (Table VI. A and VI. B) 118 Table VI. A. Variables and coefficients in initial logistic models - WTT20 118 Table VI. B. Variables and coefficients in initial logistic models -WTP60 119 APPENDIX Vll. MAPPING FORMULAE FOR THE NORTH ISLAND TIMBERLANDS 120 Table Vll. A. Map formulae used to calculate windthrow risk for WTCN model 120 Table Vll. B. Map formulae used to calculate windthrow risk for WTT20 model 121 Table Vll. C. Map formulae used to calculate windthrow risk for WTP60 model 121 APPENDIX VIII. P ORT MCNEILL EQUATIONS USED IN MODEL COMPARISON (Table VIII. A and VIII. B) 122 Table VIII. A. Port McNeill (PM) simple model #3 122 x i Table V l l l . B. Port McNeill (PM) full model #1 122 A P P E N D I X IX. C O M P A R I S O N OF P O R T MCNEILL AND NIT M O D E L S (Table IX. A and IX. B) 123 Table IX. A. Percent of actual damage segments and probability of damage predicted by the Port McNeill model sorted by predicted probability of damage and divided into 10 groups (full model) 123 Table IX. B. Variables and coefficients for P M model refit with NIT combined dataset 124 A P P E N D I X X. WINDTHROW HAZARD MAP A N D L E G E N D 125 Table X. A. Explanation of Map Legend 126 xii ACKNOWLEDGEMENTS My foremost thanks go to Dr. Steve Mitchell, my supervisor and a member of my graduate committee, who played a big role in my admission to the University of British Columbia as a foreign student. I will also like to thank him for his clear vision, support and focus, which led to the successful completion of my graduate work. Dr. Val Lemay, who apart from being a committee member helped improve my statistical analysis skills in general and especially for this study, also deserves special mentioning. I would also like to thank Dr. Mike Meitner and Mr. Bill Beese for their input and advice in all aspects of my graduate work. Ms. Yolanta Kulis, a GIS technician with the Forest Sciences department, provided technical advice throughout my study. Thanks to Mr. Gary Purpur, Weyerhaeuser Limited, Nanaimo, for providing the GIS data sources for the study. Many thanks to Mr. Frank Schreiner, Weyerhaeuser Limited, Campbell River, for his efficiency and promptness in providing additional data and explaining the contents of the datasets. Many thanks to Weyerhaeuser Limited, the International Tropical Timber Organization (ITTO), the P.E.O Sisterhood of North America and the University of British Columbia who provided financial support for my graduate study and research. xiii 1 INTRODUCTION Strong wind events impact forests over the globe in both temperate and tropical regions (Everham III 1995). Windthrow, a natural phenomenon in forests, can be defined as the breakage or uprooting of trees by wind. This results from the interaction between climate, stand, tree, soil and topographic factors. The mode of failure can be in a form of stem failure, root failure and uprooting (Mergen 1954; Somerville 1979; Stathers etal. 1994; Moore 2000). Windthrow can be termed endemic or catastrophic. The former results from routine peak winds with return intervals less than five years and is influenced strongly by site conditions and silvicultural practice and can therefore be predicted. Catastrophic windthrow results from winds with longer return periods and unlike endemic windthrow is influenced strongly by windspeed, wind direction and local topographic features. It is therefore less predictable. While catastrophic windthrow occurs in both stable and unstable stands, endemic windthrow occurs mostly in less stable stands (Miller 1985). 1.1 WHY A S S E S S WINDTHROW RISK? In 1990, 100 million cubic meters of forest was blown in the course of one night as a storm swept over Europe (Peltola et al. 2000). In British Columbia, it has been calculated that wind blew down timber equivalent to 4% of the annual 1 allowable cut in 1991 (Mitchell etal. 2001). The December 14 , 2001 windstorm on Vancouver Island had a peak windspeed of 89 km/h at Chatham Point lighthouse and lead to 586 ha of windthrow in the Weyerhaeuser North Island Timberlands operation (NIT) (B. Beese, Weyerhaeuser Limited, Nanaimo, personal communication, 2002). Slodicak (1995) reported that in the Czech Republic, between 1981 and 1990, more than 50% of total yield was cut because of some kind of injury (mostly windthrow or snowbreak). In some years (1984, 1985 and 1990), the salvage cutting represented nearly 90% of the allowable cut for conifers. Slodicak pointed out that the situation had reached the point where normal forest management was being gradually changed to coping with the consequences of catastrophes. In British Columbia, five-year development plans are drawn with interagency and public consultation to enable integration of timber, wildlife, fisheries, hydrology and aesthetic concerns. Mitchell (1995) stated that subsequent loss of designated forested streamside buffers, wildlife corridors and visual quality to windthrow seriously disrupts the intent and implementation of integrated resource planning. In addition to obvious economic losses (i.e., a devaluation of timber losses through breakage and increased harvesting costs), ecological damage and loss of recreational opportunities can occur (Lekes and Dandul 2000). The design of harvesting influences windthrow risk. Moore and Somerville (1998) reported that, in New Zealand, a large proportion of the wind damage was 2 directly associated with management activities, particularly recent clearfelling, late thinning, and the creation of new non-windfirm stand boundaries. Mitchell et al. (2001) reported that 24% of clearcut boundary (40 m from edge) segments were wind damaged near Port McNeill, northern Vancouver Island. Windthrow assessment to identify high-risk areas and the subsequent prescription of proper silvicultural strategies by forest managers is necessary to mitigate the windthrow impacts (Fridman and Valinger 1998). 1.2 S C O P E AND OBJECTIVES This thesis reports on an investigation into windthrow risk and management in North Island Timberlands (NIT), Northern Vancouver Island. The windthrow considered in this study resulted primarily from endemic winds. These winds are produced by the large-scale Pacific low-pressure systems that cross coastal British Columbia each winter. This research extends the work of Mitchell et al. (2001) who fit empirical windthrow risk models for Western Forest Products (WFP) Tree Farm Licence near Port McNeill (PM) to the north of the location of the current study. The specific objectives of this project were as follows: • To map windthrow occurrence along cutblock boundaries using aerial photos. • To develop new methods for characterizing topographic exposure and cutblock design. • To test hypotheses abut windthrow risk factors. 3 • To build windthrow risk prediction models. • To test the robustness and portability of empirical models. • To produce windthrow hazard maps using variables available in geographic information system databases. 1.3 A P P R O A C H Wind, damage and stand level information was compiled within a geographic information system (GIS) to create a dataset that was exported for analysis. Contingency tables gave an overview of the association of damage with individual variables, and logistic regression models were developed for prediction. These empirical models were then imported into the GIS to create windthrow hazard maps for the study area. The NIT model and the Port McNeill models were compared to test model portability. 1.4 THESIS OUTLINE Chapter 2, the literature review, introduces the mechanics and biology of windthrow, windthrow risk assessment methods and the role GIS can play. Important variables in windthrow study, research questions and hypotheses are also considered in this chapter. Chapter 3 describes the study area and the GIS and statistical methods and definitions used for the research. Chapter 4 presents the results of the study. In Chapter 5, the general association of wind damage 4 with individual variables is discussed and the predictive models are compared. The issue of model portability and robustness is also discussed in this chapter. Chapters 6 and 7 contain the conclusions and recommendations, respectively. 5 2 LITERATURE REVIEW 2.1 MECHANICS OF WINDTHROW Windthrow is a dynamic process. Mechanical stresses generated in tree stems and roots are influenced by windspeed and its variation above the canopy, the way in which tree crowns act as drag elements transmitting wind forces to individual trees, and the mechanical properties of individual trees (Milne 1995). The roots and soil are subjected to oscillating forces transmitted by the stem. Wind exerts a horizontal force, or drag (Equation 1). This, in combination with a gravitational moment due to displaced crown and stem mass, produces a turning moment (Figure 1; Equations 2 and 3). The critical turning moment is defined as the maximum turning moment that a tree stem or root system can withstand (Smith etal. 1987). • Wind (U) Canopy top Figure 1. Forces acting on a tree (redrawn from Peltola and Kellomaki 1993). 6 The total horizontal force acting on the tree is the sum of the drag forces due to the stem and crown. The wind induced sectional force at height z (m) is: (1) F ( 1 ) ( z ) = 0 . 5 * C d * p * A ( z ) * U ( z ) 2 Where: F(i) (z) = wind force (N) U(Z) =windspeed (ms"1) A ( z ) = projected area of the tree against the wind (m2) C d = drag coefficient (dimensionless) of the crown p = air density (kgm"3). An additional sectional force of gravity comes into play once a substantial swaying or bending of the tree occurs and that is: (2) F(2)(z) = M ( z ) . g Where: F(2) (z) = force due to gravity (N) for section centered at height z M(Z) = sectional mass of the stem and crown (kg) g = gravitational constant (9.81 ms"2) To simplify the calculation of turning moment (T), the horizontal force due to wind (Fi) can be considered to concentrate at the tree centre of pressure (wz), and force due to gravity (F2) at the centre of gravity, yielding: (3) T = F(i)*w2 +F(2)*xc where: T = total turning moment (Nm) xc = horizontal displacement of the stem center of gravity from the upright position w z = tree center of pressure (m) Drag coefficients differ among trees, depending upon branch and foliage properties. This arises from the fact that the drag force is proportional to the area 7 of exposed branches and stems to the wind and the drag coefficient of foliage (i.e., a function of shape). An increase in windspeed leads to a more streamlined tree shape and decreased projected canopy area, meaning that for trees drag coefficients change with windspeed. Mayhead et al. (1975) therefore derived an equation for drag (given in kg) as a function of live branch mass (which is fixed) and windspeed. For Sitka spruce (Picea sitchensis (Bong.) Carr.) the equation is: ( 4 ) D = 0.04436U2 * m 2 / 3 * e"° 0 0 0 9 7 7 9 * u * u D = drag (kg) m = live branch mass (kg) u = windspeed (ms"1) e=exponent The amplitude of sway, and hence the effect of gravitational force, depends on the height, stiffness and shape of the stem, the stiffness of the anchorage of the root system, the effect of adjacent tree crowns (damping) and the speed and turbulence of the wind (Peltola and Kellomaki 1993). Swaying of the bole also damps energy depending on the diameter, elasticity and shape of the bole (Stathers et al. 1994). The back and forth sway of trees in response to the direction of wind applies tensile, compressive and shearing stresses to all sides of the root system which normally starts with small diameter roots. Repeated swaying causes a progressive weakening of the root-soil system. Large root systems are stronger and provide more resistance to swaying, and small increases in rooting depth and area can significantly increase the resistance to overturning (Dietrich et al. 1982). Various studies (e.g., Putz et al. 1983; Smith et al. 1987) have shown that windsnap happens mainly in trees with strong roots, 8 while trees with weak or decayed roots tend to uproot. However, the bending moment at which a stem would break is often close to that at which it would uproot (Coutts 1983). 2.2 FACTORS AFFECTING WINDTHROW DAMAGE Any factor that influences effective root anchorage, the strength properties of the tree, and the direction and characteristics of the wind within and above the stand can be said to affect windthrow risk. Fridman and Valinger (1998) listed variables needed for risk modelling, including: (i) single tree data, e.g., species, biomechanical resistance, tree form, and root structure, (ii) stand data, e.g., species composition, stand age, mean height, and stand density, and (iii) site data, e.g., climatic conditions, geographical position, soil water, aspect and slope information. It is worth noting that the mechanics of windthrow and biology of windfirmness are universal; therefore, the role of environmental factors is consistent in widely varying forest types. One of the challenges in evaluating the contribution of individual risk factors is that many environmental factors (e.g. age and height; height and site quality, rooting and topographic exposure) are correlated. 9 2.2.1 Individual Tree and Stand Level Characteristics A stand is a community of trees sufficiently uniform in species composition, age, arrangement, and condition to be distinguishable as a group from the forest or other growth on the adjoining area, and thus forming a silviculture or management entity (Smith et al. 1997). Stand height and density or stocking and species composition are important factors in windthrow (Foster and Boose 1995), as are silvicultural treatments such as clearfelling, thinning, pruning, and edge feathering. Some variables, which make a stand unique and may help in windthrow assessment, are discussed in this section. That plants respond to mechanical stimuli is common knowledge. This is vividly demonstrated by the Albizia julibrissin (mimosa) plant, which folds up its leaves when touched (Hall et al. 1971). When a tree undergoes higher strains as a result of wind loads, acclimative growth is initiated (Ennos 1995). This developmental acclimation prevents the tree from buckling under the existing wind loading conditions by reducing drag and increasing mechanical strength (Telewski and Jaffe 1986; Matheck 1989; Telewski 1995; Wood 1995). Stokes et al. (1995a) also observed a significant difference in root growth between wind-stressed and non-wind-stressed trees. Hierarchically, resources are allocated for a plant's growth, in the order of descending priorities, as follows: buds/foliar formation, new root formation, 10 storage, radial or diameter growth and finally, protective chemical formation (Waring and Schlesinger 1985). But these priorities vary in response to environmental stresses. For example, trees in dense stands use most of their available growth resources for crown and fine root formation. 2.2.1.1 Species The different characteristics of different species may produce differences in their resistance to windthrow. For example, species with flexible crowns that can be easily shaped by the wind may present a reduced sail area (e.g. Mayhead et al. 1975). Papesch (1974) reported that rooting depth and area, size, number of roots, and whether roots interlock all affect stability. The inherent mechanical properties of a particular species should, therefore, determine its susceptibility to windthrow. While many authors have noted differences in species vulnerability, Brokaw and Walker (1991), found that there was no clear relationship between damage of conifers versus broadleaf plants in their study of hurricanes in fourteen tropical and temperate forests. This discrepancy likely results from the plasticity of trees and the confounding effects of site and stand characteristics on tree form. Trees of any species with larger or denser crowns, poor root anchorage, root and bole rots and high height-to-diameter ratios are normally more vulnerable to windthrow as opposed to trees with smaller and open crowns, good root anchorage, non-diseased roots and low height-to-diameter ratios (Stathers etal. 1994). 11 Between species differences in vulnerability have been found in mixed species stands and in studies where site and stand conditions are uniform. Studholme (1995) found that New Zealand Douglas fir (Pseudotsuga menziesii (Mirb.) Franco) trees have better windthrow resistance than radiata pine {Pinus radiata D. Don.) trees. Studies of exposed stands on well-drained soils showed conifers (pines and spruce) to be preferentially windthrown compared to hardwoods (Foster and Boose 1995). A study in New England, USA, suggested that fast-growing species are susceptible to damage as a result of their tall stature, the concentration of foliage in the upper canopy, and low-density wood (Jensen 1941). 2.2.1.2 Stand Height Foster (1988), investigating hurricane damage in eastern hardwoods, found that the percentage of windthrown trees increased with stand height. Smith et al. (1987), in their study of mechanical stability of black spruce (Picea mariana (Mill) B.S.P.) in northern Ontario, found that critical turning moment increased with tree height to a certain height. Beyond this height trees become more vulnerable to wind damage. Peltola and Kellomaki (1993) found that critical windspeed for uprooting trees decreased with increasing height. They also commented that tall and slender trees with large crowns are more susceptible to wind-induced damage than similar trees with small crowns. 12 2.2.1.3 Stocking/Stand Density Stand density is a measure of the area occupied by trees, usually given in terms of well-spaced trees per hectare or basal area per hectare. Stocking, on the other hand, describes how stand density relates to someone's notion of what ought to be. Hence, for sawlog production, young stands with a density of 1200 stems/ha or more can be termed as fully stocked (Smith et al. 1997). For mature stands full crown closure is normally expected. Stocking class at a reference year is sometimes included in the forest cover label and is based on leading commercial species, stand age and/or the size (diameter), and number of stems per hectare (British Columbia Ministry of Forests 2001a). Laiho (1987) found stand density to be important in explaining damage caused by snow or wind. The stocking of a stand can reflect management practices (e.g., thinning), as well as the fertility of the site. Low productivity sites may not carry full crown closure. While some researchers have found that damage increases with stand density (e.g. Cremer et al. 1982), others have argued that shelter and damping in dense stands should increase overall stand stability (e.g., Smith et al. 1987; Foster 1988). Mayhead et al. (1975) observed that more stable trees have lower damping values. His damping experiment showed that trees growing at the widest spacings were the most efficient in dissipating energy absorbed from the wind and the least dependent on the support of neighbors; hence, they should be less affected by damage to neighbors. Gardiner et al. (1997) considered the 13 same question. They concurred with Mayhead's predictions for individual trees, but suggested that trees in widely spaced stands are more likely to overturn than trees at normal spacing because of the increased wind loading. They concluded that seeking to improve total stand stability by increasing spacing requires a more complete knowledge of the variation of vulnerability and a better understanding of the spread of wind damage in forests. Persson (1972) observed that wind and snow often damages tall and slender trees. Researchers have shown that trees in wider spaced stands have higher taper with a resultant superior stability (e.g. Peltola and Kellomaki 1993 [theoretical]; Valinger et al. 1993 [empirical]). Wider initial spacing is recommended as a silvicultural option to minimize damage by snow or wind (Fleming and Crossfield, 1983; Slodicak 1995; Mitchell 2000). 2.2.1.4 Stand Age Some studies have found a positive relationship between damage and stand age (e.g., Foster 1988). Foster also reiterated that damage was largely confined to overstory trees in young stands, whereas in older stands, an increasing percentage of co-dominant, intermediate and understory trees were damaged. In his research, Moore (2000) found that many of the failures in trees were associated with stem defects; in his case, mature stands with root rots and other stem defects may be more vulnerable than younger stands. Laiho (1987) also 14 found stand age to be an important variable in explaining damage caused by wind and snow. An increasing amount of damage with increasing age was observed in balsam fir (Abies balsamea) stands in a 1994 windthrow and this was related to an increasing occurrence of root and butt rot due to age (Whitney 1989). It should be noted however, that age is often confounded with height. 2.2.1.5 Site Index The productivity of a site largely determines how quickly trees will grow and, therefore, affects the volume of timber that will grow in regenerated stands and the age at which those stands will reach merchantable size (British Columbia Ministry of Forests 2001a). Site quality was historically classified in forest cover typing as good, medium, poor, or low. Site index (SI) is an indication of the productivity or growth potential of sites for particular species and is now commonly reported in forest cover mapping. Site index, in British Columbia, is expressed as the potential tree height at 50 years breast height age. Trees sampled for SI determination must be dominants with no period of suppression and no stem damage. Site index provides standardized comparisons of productive potential between sites across a broad range of existing stand conditions (British Columbia Ministry of Forests 1996). Site index reflects the inherent physical site factors, such as climate (solar input, precipitation, temperature, etc) and soil (nutrients, moisture, etc.) that are known 15 to affect windthrow risk. Harris (1989) found higher damage on richer sites in coastal Alaska. Since height at a given age increases with increasing site quality for vigorous stands, it is difficult to separate the effects of site quality from those of height. Mitchell et al. (2001) found a strong positive relationship between windthrow and site quality and a weaker relationship between windthrow and height. They suggested that in addition to height, there was some other property of high quality sites that contributed to windthrow risk. 2.2.2 Soil Characteristics When a tree is moved laterally by the force of the wind, the leeward and windward sides behave differently. The root-soil system is subjected to bending and compressive forces against the bearing surfaces of the soil on the leeward side, whereas on the windward side, where the root-soil is lifted, the system is subjected to tensile and possibly shearing forces. The tensile strength of soil is about three to five orders of magnitude weaker than that of roots under tension (Coutts 1983). The roots in the soil thus provide a stiffening effect much like reinforcement in a beam. Branching of main roots causes a substantial reduction in this stiffening effect. The shear strength of soils decreases with increased moisture content and windthrow often occurs during winter gales when the soil is wet. 16 Different soils offer different root-soil resistances (R rs). Coutts (1983) found that plants rooting through loam into compacted gravel had the largest root reinforcement, whereas, in clay soils, roots were pulled out instead of breaking. Decreased amount of root material in soils gives R r s to be less than soil strength leading to a fracture in the soil, due to low elasticity. The amount of root material in the soil required to increase the R r s to exceed the soil strength, termed the critical rooting density, is modified by soil factors, which influence the root-soil bond. Dean and Ford (1983) found the occurrence of lateral roots, the angle at which they subtend, their origins, change in direction, and other characteristics to be very systematic in roots. However, these systematic processes were disrupted by changes in the soil environment. In particular, angles of descent and of forking changed when the roots encounter denser substrates, impenetrable layers, and/or other obstructions. Deep, well-drained soils provide better root anchorage and, hence, more windfirm trees than shallow and saturated soils (Alexander 1964; Rizzo and Harrington 1998). However, the richness of a soil will determine how tall stands become and the degree of site occupancy, which, in turn, will affect risk of windthrow. Mitchell et al. (2001) found less damage on organic soils compared to colluvial, fluvial and morainal soils, likely because organic soils were occupied by short, poorly stocked, redcedar dominated stands. 17 2.2.3 Topographic Characteristics Topography can modify the direction, speed and turbulence of prevailing winds and so influence the potential for wind damage to a stand (Quine 1995; Everham and Brokaw 1996). Hannah et al. (1995) found strong correlations between windspeed, geographic and topographic characteristics of a site such as altitude and topex-to-distance. Foster and Boose (1995) observed that in hilly or mountainous areas within hurricane paths, topographic exposure may make the difference between little damage and complete destruction for the same forest type, and also that the pattern and relative extent of protected and exposed areas varied with topography. They mentioned the acceleration of the wind over ridges and summits and channelling of the wind up valleys and around protuberances as other complex topographic effects influencing wind damage. The interaction of topography and climate to determine high wind speeds (wind exposure) was given a weight of 77% of the total score in the British windthrow hazard rating system. Topography alone makes up 30% of this weight (Miller 1985). Properly characterizing wind exposure, therefore, is a key issue in windthrow risk assessment. In the absence of direct local measurements of wind, it is necessary to develop indicators. 18 2.2.3.1 Aspect, Slope and Elevation The strength of wind experienced by a tree and, hence, the probability of the critical windspeed being exceeded will depend upon the tree's location in the landscape in relation to the passage of weather systems. Aspect, slope, and elevation are basic topographic variables, which have been used by various researchers in windthrow prediction (e.g., Quine 1995; Mitchell et al. 2001; Kramer et al. 2001). 2.2.3.2 Terrain Analysis/Curvature The routing of water and nutrients over the surface of a landscape represents a fundamental geomorphological process that is intimately tied to landscape form. The subdivision of the continuous surface into discrete hydrological units provides an important step in the geomorphological treatment of an elevation model. Profile and planform curvature are terrain variables that can be calculated using a grid theme within ArcView 3.2. The profile curvature affects the acceleration and deceleration of flow, and therefore influences erosion and deposition. The planform curvature influences convergence and divergence of flow. A positive curvature indicates that the surface is upwardly convex at that cell. A negative curvature indicates that the surface is upwardly concave at that cell. A value of zero indicates that the surface is flat (ESRI 2000). The profile curvature may be useful for building a windthrow model, since it relates to 19 deposition of water and nutrient flows. This may indicate freely versus poorly drained soils. 2.2.3.3 Other Topographic Variables Topex is a measure of topographic exposure at a given location. It is defined as the summation of the measured skyline angles at eight points on the compass with no directional value being less than zero (Wilson 1984). Hannah et al. (1995) used a modification of this measure, topex-to-distance, which is the sum of the angles to the eight cardinal directions to a fixed distance, with negative angles allowed. Hannah et al. (1995) found that all three variants of this measure (topex-to-1 km, topex-to-2 km, and topex-to-3 km) correlated well with annual windspeeds for twenty-one sites in Scotland. Topex-to-2 km was used by Mitchell et al. (2001) for his Port McNeill study. 2.2.4 Management Variables Harvesting timber exposes trees along the edges of cutblocks to increased wind loading. The increase depends upon the initial density and structure of the stand, the orientation of the edge relative to the direction of strong winds, the width of the opening, time since harvest, and the shape of the boundary. Rollerson and McGourlick (2001) found an increasing amount of windthrow, as well as an increasing depth of windthrow penetration from leeward > parallel > windward 20 boundary exposure classes. Gardiner et al. (1997) found that a rapid increase in wind loading with increasing gap and recommended that openings should be reduced in the direction of damaging winds to reduce vulnerability. 2.2.5 Climatic Conditions Climatic conditions affect windspeed, gustiness, storm duration, soil moisture conditions, and snow and rain loading on the crown (Stathers et al. 1994). In the United Kingdom, for example, the wind climate is the most limiting factor to forest growth (Cannell and Courts 1988). Wind speeds as low as 54 km/h can cause considerable windthrow. The longer the storm lasts, the more the tree sways and, hence, the more roots are broken and anchorage loosened. Snow or ice loading increases crown mass and drag (Mayhead et al. 1975), which, in turn, increases windthrow susceptibility. Also, a wet soil following snowmelt or rain has low root-soil adhesion and soil shear strength hence poor anchorage (Stathers et al. 1994). A good predictive model is dependent on accurate assessment of windspeed, but many forest locations lack such records. 2.2.6 British Columbia Coastal Wind Regime Northern Vancouver Island is geographically complex with smaller islands and channels along the edge and a mountainous interior. This, coupled with the rarity of climate stations, makes it impossible to give an accurate weather or wind 21 statistics for particular areas. However, the Marine Weather Hazards manual compiled by Environment Canada provides a general overview of the wind regime in this area of Vancouver Island (note: it is a convention in meteorology that winds are referred to by the direction from which they blow). Southerly winds are the strongest and peak during the winter months, while westerlies are common during summer in Johnson Straits (Environment Canada 1992). The winds are strongly funnelled between the Vancouver Island mountains and the Coast Range and can be as much as 27 km/h stronger, in some situations, than stations in the Queen Charlotte Strait. In the afternoons, westerly winds do develop in Queen Charlotte Strait as light sea breezes in the summer and these are funnelled down Johnson Strait. Such wavelike winds can reach between 55 km/h and 65 km/h by the evening and this eases by around 2 am (Environment Canada 1992). The wind regime at coastal stations is more severe than at stations in the British Columbia interior (Mitchell 1999). 2.2.6.1 Mean Windspeeds, Gusts and Return Periods Windspeed is the speed at which wind passes a given point. A gust is a positive instantaneous departure from the mean windspeed. An hourly windspeed is usually a one or two minute mean windspeed taken on or near the hour. Windspeed averaged over a whole year is the mean annual windspeed. The highest sustained one or two-minute wind within a given period is the peak hourly windspeed. A return period is the average time within which a given windspeed 22 will be exceeded just once (Environment Canada 1989 and 1993; Murphy and Jackson 1997). Hourly wind speeds of 40 km/h with gusts to 100 km/h were documented in a Scots pine (Pinus sylvestris (L)) plantation during a destructive gale in England (Oliver and Mayhead 1974). Coatta (cited in Mitchell 1999) reported that a peak hourly mean wind speed of 54km/h was recorded during the May 5 t h , 1990 windthrow event in Quesnel. Table 1 shows the maximum annual values for hourly windspeeds with directions for Chatham Point, a station close to the study site on Johnstone Strait. Winds of 55 km/h have return periods of between one to three years at Vancouver Island stations (Table 2). 2.2.7 Windthrow Risk Factors Reported in Coastal BC and Alaska Mitchell et al. (2001) found damage to be lower in redcedar stands than in hemlock (Tsuga heterophylla (Raf.) Sarg.) dominated stands although species itself was not a significant variable in any of their models. Other coastal studies have similarly found redcedar to be more windthrow resistant than other species such as hemlock, amabilis fir (Abies amabilis (Dougl.) Forbes), and Sitka spruce (e.g., Harris 1989; Beese 2001; Rollerson and McGourlick 2001; Rowan et al. 2001). Mitchell et al. also found that more damage occurred in immature compared to mature stands, on good quality soils compared to low quality soils, and on topographically exposed segments with southern orientation. 2 3 Table 1. Peak hourly windspeeds (two minute mean) and their directions for Chatham Point station near study area for the years 1964-1992. Year Mean wind speed, km/h Direction 1964 61 NW 1965 63 W 1966 56 W 1967 58 NE 1968 53 NE 1969 58 SW 1970 63 SW 1971 66 E 1972 68 SW 1973 58 NW 1974 60 W 1975 68 W 1976 64 NW 1977 64 S E 1978 50 S E 1979 56 E 1980 60 NW 1981 53 S E 1982 56 S E 1983 58 S E 1984 66 S E 1985 69 S E 1986 63 W 1987 63 S E 1988 108 E 1989 58 NW 1990 58 NW 1991 60 S E 1992 63 S E (Environment Canada 1994) 2 4 Table 2. Peak hourly windspeeds and gust return periods in years for selected stations. Station Hourly Mean Windspeed (km/h) 50 60 70 80 90 100 (Return period - years) Cape Northern tip of 1 1.2 3.2 13.3 62.5 Scott Vancouver Island Comox 70km south of 1 1 1.9 6.7 28.7 128.2 Campbell River Vancouver 1 2 13 Peak gust Speed Years Comox 70km south of 130 14.5 Campbell River 150 74.3 E. Coatta (Cited in Mitchell 1999) In his study of clearcut edges on the Queen Charlotte Islands, Rollerson (1979), found taller, larger diameter trees on higher elevation sites along south facing boundaries to be more frequently damaged. Trees of height greater than 20 m in less dense and low volume stands on shallow soils were found by Holmes (1985) to be more frequently damaged in his study of clearcut edges in the Tsitika watershed in NIT. Rollerson and McGourlick (2001) found more damage in well-drained soils on highly exposed stands located in the windward direction in their study of riparian 25 reserves near Port McNeill. Beese (2001) found more damage on shelterwood plots (30% basal area maintained) and more damage in dominant crown classes and on boundaries with southeasterly orientation in his study of the Montane Alternate Systems Study (MASS) in central Vancouver Island. Rowan et al. (2001) found less damage in plots with modified (topping, pruning or a combination of the two) stands compared to unmodified stands. Table 3 gives a more complete summary of the results of these studies. 26 3 o — i m fl) ro n 0) _ 3 - cu S (Ti-ll o o fD % Q_ CD =• cn ZJ CD » 8 3 o. o 5} co -o ^ P " II II ^ < Z 3 O CD Z l Q . E 0_3 II T a 5' CQ ICQ CD Q. 5' CD o S. 3 rr QJ cn CD cn °> o 2 . o o c ~ CQ n cn CO 11 II <§ cn o O CL rn » II § cn a> o cn 5. CD =• 3 cn - r a f P " r-I " CD _ '"*• o 3 € CD 9- ^  C=' || 3 3 :>£ TJ c n 3 •o o o II co 2. C£> w O D 5 tt) CD Q_ — CD CD O" 0) | § 0) O. 5-CD m cn - i s 2 CD ~ CD O cn CD o CD —| cn a) W 3 00 a i 5 0) =J CD CD CD CD — i tt) CD ~ -S °' CD rf 0 •< 9: $ LT "S 5' » S ? S t • 2 zr CD 3 -+ O ^ < CD < O cn a o o 5' ro cn a o ~X o w 0) o cn 3 ^ " CD OJ (i) s 'S CD Q) CD CD O O _ Is s i o ' cn cn 3 5' w cn B r i 5' CD 0) CL |co CD tt) ICQ 0 CL r s o T T UJ cu « W <S tt) ^ a, CL Ei 0 tt) c r v 00 rs - h ~ Q_ fl) =r fl) O - h =2 5' tt) fDCQ 0. O 3 C CT * c l 2 . II CO o c o rt zr ET Q . 0' cn c cn ri' (Q o C L <• V 0 D CO CD ^ Q. O c o 9> 00 ^ co cn' > o 0) o cn tt) T T cn tt) bj 0 a e> ?, 8 S-cu Q - CQ f 0 V O 0 CL tt) cn T3 —\ c o 0 tt) ^ Q g NJ o °- ^ < 5 2 CD cn 0 cn 0 QJ cn 2. CL o fl) cn 3 V >< 2. a) 9- =CQ 1 f 0 0 T3 cn - i cn z:. g 0 ~n ? r n —• - i — 0 ft> g - O V r 1 tt) O 0) O Ti CQ t 3 Q v . w S o j 0 -a >< 0 fl) = T J . 0_ V ro V o Q)_ O O tt) to | O CQ O 0 fl) _ . cn n «-*" o fl) M — 0 3 0 1 CQ Q. 5' cn C L o CO cn ? r < cu tt) < < 0 o ? ^ CD CL S 3 0 -« CD < cn cn TJ O cn < cn a , t- ^ Q. rr 0 =h o 8 1 ^ CD ^ 7 o o 0 3 3 a) i—t-c - i 0 V 3 a ET 0 0 o o o 0 CL tt) c o g v M | g II c ^ < - i Q)_ ^ 73 V V TJ V CD V V TJ V cn 0 m 5 X x CD O o 0 0 C L CL C L 0 0) 3 cu CQ 0 0 cn cn 0 V CO Z O ° a m a) cn 0 cn O C L a flT 3 ca o Q) n QJ u' a> 3 - a> cr Go' || rt^ 3 J - l ' o O c/>' $ i—f-3 n CD C L CD GO 3 CD 7] ed II o CD 3 a. o CO -* a —1 o' p- n ~Z_ C L II Z 3 o CD 3- Q . c ' o 3 II T l o C L — i CO II ani Do o ca II 00 po O CO II II CO o C/) o o a. c • 00 II m II 00 CD o CO c rt-CD —i east hem o ck, II t— II CD o 3 edi n 3 3 AA edi c II 3 we rt-"a II o o o o 33 o OJ CD O o CO 00 co_ 03 O ro o o 00 0 0 oo CD I N) O S " QJ < I (fl II < 00 7 3 O <• » 0 3 T3 - 7 C T CD —I = Q_ o • 3 a NJ ^ § £ 2 ? » O - i c £ Z L O o 3 "0 O 3. o CD 03 0J 0J 3 Q. V £>CD < OJ V O C L OJ o C L 0J 0J 3 C ° 3 & 3 Z * o CD C L ° - Eii O CD C L o o 00 CO CD CD CD < 0J =: o W CO 8 o =S 3 0 0J 9- 3 Si' v 0 =;• o OJ O V ° 8 o o V 0 o § C L - i 0J_ Q_ 0J v 3 0 C L 9-3 OJ -o 3 ' 0 C L 0 v a 0 0 o 3 , C T 0 0 O X X 3 o o 00 00 0 0 _ C L Q 0 V 00 v C L 0J o KJ 00 o 3 ° Z 3" O 0 3 - i 3" 9: II CB Z 00 ^ 0 00 0 </> T l O 3 00 r- 0 rt- >TL < 3 . 7 I. ^ ' W i . <° 3 C L oo 3 I O ( Q S" ^ 3 >< a 0 0 >< CP" T3 ^ O 00 0 C L T3 NJ 3 . NJ QJ ^ T 3 ' <r QJ — • 51 < 3 . C L 0 5' o> 0 C L 3 V ^ 00 — DJ ^ 0 " ^ 0 Q-T5 C L V 0J C L V 00 ^ 00 C L 0 C L C L CO-3" cn o oo CO A < U V c 3 0 C L V ro 0) v O o < o oo 0 Q-0 oo 0 C L ro 3 -0 ' oo —to C L <° rtT 3 ro v CQ 00 00 0 00 ST 3 C L oo 9r W S ^ 0 CO T3 = oo o O < o a < o < 0 5 00 ~ 1 « rt-oo C L CD 3 3 0 00 o (7) 0 ?2 - 1 CQ 00 Z7 OJ cr "S ° 0 c 00 3 C L DJ V 3 ? w o 9 o oo < c: « ° 5 C L QJ DJ 3 DJ CQ 0 rt?' C o o DJ 5' 3 OJ 3 Q-l o O (Q 0J TS 0J 3 DJ CQ 0 3 0 3 3 Q-l 2.3 WINDTHROW RISK A S S E S S M E N T A P P R O A C H E S Risk, in this thesis, is the probability of a tree or stand being blown down by an endemic wind. Identification of the location and type of stands most at risk for windthrow could be done objectively through a formal classification or informally through personal observation. Perceived risk may differ significantly from a formal objective assessment (Smith 1992). Risk assessments are done in several other fields, for example, in agricultural climatic risk assessment. In engineering, the magnitude of an event, such as flood and/or storm that will damage a structure can be calculated (Cook 1985). Climatic statistics are then used to calculate the probability of this event occurring and structures are designed to withstand that event expected to occur a specific period of time (e.g., 50 years) depending upon the expected lifespan and use of the structure. There are three main ways of assessing windthrow risk: observational, mechanistic, and empirical. The resulting models relate damage magnitude or probability to one or more of the following: tree, stand, site, management, and topographic and climatic variables. 2.3.1 Observational Approach The observational approach tallies the presence of factors known to be associated with higher incidence of damage. The relative risk of windthrow is 29 considered to increase with increased number of risk indicators observed (e.g., Stathers et al. 1994). Managers look out for trees with characteristics like asymmetric or stilt roots, disproportionately large crowns, root rot, growing on unstable substrates, etc. Trees with a combination of these characteristics in a stand or along a proposed cutblock boundary give them an indication of the likelihood of windthrow and the form of treatment needed to forestall this event (Stathers etal. 1994). 2.3.2 The Mechanistic Approach In the mechanistic modelling approach, the likelihood of damage is based on an evaluation of the critical windspeed for tree failure, and the probability of a wind of that speed occurring at a given location. These models enable managers to identify the risk of damage for different parts of a forested area and to evaluate the strategies required for optimum stability at a particular site (Talkkari et al. 2000). Such models also allow researchers to study how wind conditions, site and tree properties affect the mechanisms of damage in various locations and under various forms of stand management. Smith et al. (1987) used a mechanistic approach to determine the optimal height of black spruce, in the clay belt region of northern Ontario, before losses due to windthrow become excessive. In their experiment, the stability of black spruce sample trees was measured by winching them over and determining their critical 3 0 turning moment. This was expressed as a function of height, dominant stand height, and stand stocking using regression analysis. Using crown and height measurements, calculations were made to estimate the windspeed necessary to exceed the measured turning moments. Wind return periods were derived from climate station data for the clay belt region. Several other researchers have used this approach (e.g., Peltola and Kellomaki 1993; Gardiner et al. 1997). These mechanical models are based on static wind loading. In their wind tunnel experiments, Mayhead et al. (1975) found it impossible to conduct dynamic tests on trees, because of the extreme difficulty of providing a firm unchanging anchorage for the base of stems. There have been subsequent experiments on wind or sway components (e.g., Chen et al. 1995; Guitard and Castra 1995; Milne 1995), but no fully developed dynamic model was found in literature. Static models have been successfully developed for single species uniform canopied stands of healthy small-medium (<35cm dbh) trees. Considerable work is needed to extend this approach to multi-species, non-uniform stands, and stands of larger and older trees. 2.3.3 The Empirical Approach Empirical models relate the presence or magnitude of wind damage in sampling units to the, attributes of these units. For example, Fridman and Valinger (1988) used logistic regression to predict the risk of damage from snow and wind based 31 on tree, stand and site characteristics of Scots pine in Sweden. Their model was able to predict damage in both damaged and undamaged plots with good accuracy. Empirical models establish relationships between response and independent variables with no indication of it being causal or associative. The empirical method approach is suitable for stands with complex and variable structure and composition, and where geography and soils are heterogeneous. 2.4 RESERVATIONS ABOUT A S S E S S M E N T A P P R O A C H E S Static mechanical models ignore the dynamic processes. For example, assessment of windspeed required to blow over trees of different forms and anchorage in natural conditions are simplified in wind tunnel experiments. Applying a mechanistic model requires knowledge of local windspeeds. Many forest locations, however, have sparse windspeed records due to difficulties in maintaining stations in remote areas (Hannah et al. 1995). Also mechanistic models have not yet developed drag and critical turning moment relationships for many species, large trees, complex stand structures, trees with stem or root decay and mixed species stands. Effects of cutblock shape and boundary orientation on windspeed are not well quantified. Simple linear models may be flexible to extrapolations over a range outside the one that a study was conducted. This may not be the case for complex multi-linear relationships. Predictions from such empirical models may hold only within 32 the range of data used in fitting the models. Hence, they should be applied with caution in sites with conditions that are different from where a study took place. Forest planning models generally take a deterministic view of the world in which temporal variation in natural forces is ignored. Similarly, observational assessments and mechanistic models for predicting windthrow risk in plantation forests rank different sites and/or silvicultural treatments, but cannot successfully assign a probability value to the occurrence of damage (Moore and Somerville 1998). It is recommended that models need to be more probabilistic and less deterministic (SAF 1993). 2.5 USE OF GEOGRAPHIC INFORMATION SYSTEMS IN EMPIRICAL STUDIES Geographic information systems (GIS) are an organized collection of computer hardware, software, geographic data and personnel designed to efficiently capture, store, update, manipulate, analyze and display all forms of geographically referenced information (Burrough and McDonnell 1998). These systems are useful in constructing empirical models, which require information about large numbers of sample units. Geographic information systems and aerial photograph interpretation provide the opportunity to gather stand or landscape level information for large areas, generate spatial variables, conduct analyses, and map the results. With this approach, laborious ground surveys are avoided. Hence, larger areas can be used to generate larger sample sizes giving greater 33 flexibility over sampling procedures and sample sizes for statistical analysis (Wright and Quine 1993). 2.6 GIS-BASED WINDTHROW STUDIES 2.6.1 European Studies Wright and Quine (1993) investigated storm damage to trees at Wykeham Forest in North Yorkshire. Local forest staff mapped the extent of wind damage in ground surveys and divided the forest into polygons with none, partial and total windthrow. Survey results were digitized. Windthrow and individual stand, soil and topographic layers were gridded at 50 * 50 m. The association between windthrow and individual variables was examined, but no predictive model was built. In New Zealand, Moore and Somerville (1998) assessed the risk of wind damage to plantation forests by combining mechanistic stand/tree failure and airflow models within a GIS system. The resulting model performed poorly in complex steep terrain, but gave reasonable predictions in simple terrain. Lekes and Dandul (2000) developed a theoretical regional wind damage risk classification (WINDARC) for Czech Republic. Damage incidence within different soil and vegetation types was analyzed to produce risk classes from 1 (lowest 34 risk) to 9 (highest risk). A numerical airflow model was then used to calculate terrain exposure and this was integrated with the risk classes to produce the WINDARC (lowest risk of 1, highest risk of 9) representing the current threat of forest stands from wind. In comparing calculated WINDARC scores with present damage, a high correspondence with actual wind damage occurrence was observed. They commented that the forest stands classified by WINDARC scores 8 and 9 are routinely being damaged by wind. 2.6.2 Coastal North American Studies Kramer et al. (2001) investigated the role of abiotic factors (slope, elevation, soil stability, and exposure to prevailing windstorm) in controlling patterns of stand-replacing windthrow in the pristine coastal temperate rain forests of southeast Alaska. They divided the forest into 0.8 ha cells and classified these as being initiated by stand-replacing windthrow, or not, based on aerial photographs and ground-truthing data. The slope, elevation, soil stability, and exposure categories where then extracted for each sample cell. Relationships between individual variables and windthrow were examined and a multiple logistic regression model was built with a 72% correct classification. Mitchell et al. (2001) modeled cutblock edge windthrow risk on the north end of Vancouver Island near Port McNeill using stand level information. Segments 35 50 m long by 40 m deep were created around the edge of sample cutblocks and this was overlaid, in a GIS, with forest cover, logging history, ecosystem, management, topographic and windthrow layers. Logistic models were then fit using 60% of the data and tested with the remaining 40% of the data. The model yielded correct outcome predictions of 71-76% and was exported back into a GIS to generate windthrow hazard maps for the area with predicted probabilities of various percentages of forest stand boundaries getting damaged. Mitchell et al. (2001) did not address a number of issues. Spatial correlation, a phenomenon of importance in dealing with spatial data of this kind, was not accounted for. No wind data were incorporated and there was no variable accounting for a relationship between peak wind directions and bearing of segments in relation to damage. The study did not investigate damage patterns for different distances from the edge. Lastly, segments were classified as damaged only in terms of segment area lost. 2.7 SUMMARY, RESEARCH QUESTIONS, AND HYPOTHESES Windthrow results from the interaction between climate, stand, tree, soil, and topographic factors. The empirical modelling approach is suitable for areas such as NIT where stands are variable in structure and composition and topography and soils are heterogeneous. Due to the large numbers of sample units required and the need to derive variables with a spatial component such as topographic 36 exposure and boundary condition, GIS is a useful analytical tool. Unresolved questions that this research sought to answer include: a) Is wind a driving force in windthrow damage prediction? b) Is there a clear directionality to the damage? c) How does the damage pattern compare with trends discussed in the literature? d) Are other variables apart from those documented to-date useful in predicting damage? e) Are the models developed in Port McNeill robust or portable enough to work in NIT? Based on literature, it is expected that damage will: 1) be greater at the immediate edge of cutblocks than further into standing trees; 2) increase with mean wind speed; 3) increase as exposure of edge segments within cutblocks increases; 4) increase with topographic exposure; 5) increase with increasing stocking; 6) increase with stand height; 7) increase with soil fertility; 8) be lower on organic soils; and 9) be less frequent in cedar stands compared to hemlock stands. 37 It is also expected that: 10) the new variable profile curvature will contribute to damage prediction; 11) the models will be robust within the NIT operating area; and 12) that the Port McNeill models will be portable to NIT since conditions are broadly similar. 38 3 MATERIALS AND METHODS 3.1 DESCRIPTION OF STUDY AREA The study area is Weyerhaeuser Limited's North Island Timberlands operating area, located northwest of Campbell River in the middle section of Vancouver Island. It is bounded on the north and east by Johnson Strait. This operating area is 203,065 ha, and is located within Tree Farm Licence (TFL) 39, which is the largest TFL in British Columbia (British Columbia MOF 2001b). The communities of Sayward, Campbell River and Kelsey Bay are situated in or close to NIT (Figure 2), and the operating area includes the watersheds of the Tsitika and Eve Rivers. Chatham Point lighthouse is the nearest Meteorological Service of Canada (formerly Atmospheric Environment Service) weather station. Coastal Western Hemlock (CWH) and Mountain Hemlock (MH) are the two forested biogeoclimatic zones found within NIT (Table 4). Western hemlock and Pacific silver fir are the dominant tree species. Varying amounts of yellow cedar (Chamecyparis nootkatensis (D. Don) Spach.), Douglas-fir (Pseudotsuga menziesii (Mirb.) Franco), western redcedar and Sitka spruce are also present (Greene and Klinka 1994). Timber harvesting in the area started in 1910. Forest cover is dominated by multi-storied old growth hemlock stands to the north and west, and uniform second growth stands dominated by hemlock in the major valley bottoms in the east, and Douglas fir in the Sayward forest to the south. 39 Figure 2. (a) Map of coastal British Columbia showing study site; (b) Detailed map of North Island Timberlands (NIT) 4 0 Table 4. A summary of the climate for each of the 4 major subzones Biogeoclimatic unit CWHvml CWHvm2 CWHxm MHmml Name of reference station Reference elevation (m) Haney Loon Lake 354 Tunnel Camp 671 Cumberland 159 Grouse Mt. Resort 1128 Mean annual Range precipitation (mm) ref. Stn. 1555 to 4387 2682 2760 to 2850 2850 1100 to2721 1570 2565 to 2954 2565 Mean annual Range temperature (°C) ref. Stn. 7 to 10.1 8.3 No data 7.8 to 10.7 8.7 4.5 to 5.0 4.6 Extreme minimum Range temperature (°C) ref. Stn. - 8.9 to -22.8 -19.4 No data -13.5 to -25.6 -20.6 -18.5 to -26.7 -18.5 Extreme maximum Range temperature (°C) ref. Stn. 27.8 to 41.4 34.4 No data 29.4 to 43.9 43.9 29.0 to 33.3 29.0 (From Green and Klinka, 1994). *Mt = mountain, ref. stn. = reference station, CWHvml = Coastal Western Hemlock submontane very wet maritime, CWHvm2 = Coastal Western Hemlock montane very wet maritime, CWHxm = Coastal Western Hemlock very dry maritime, MHmml = Mountain Hemlock moist maritime. The area has varied terrain, ranging from rugged mountains to wet lowlands, and the forests are highly productive. The U-shaped valleys between mountains of over 1500 m elevation are indications of glacial action. Soils are derived primarily from morainal material followed by colluvial and fluvial deposits (Howes 1981; British Columbia Ministry of Forests 2001b). The climate is humid with cool summers, mild winters and little to substantial snow. Mean annual precipitation ranges between 1555 and 4387 mm and 80% of this falls between October and March (Howes 1981; Greene and Klinka 1994). Winter winds are largely from the east-southeast and are the strongest yearly winds. Westerlies are more predominant in the summer (Environment Canada 1992). 4 1 3.2 INFORMATION S O U R C E S AND DATASET CREATION 3.2.1 Information Assembly and Data Conversion Ecosystem, forest cover, logging history, roads and hydrology layers were obtained as Arclnfo interchange coverages (*.e00 extension) from the Nanaimo Woodlands and NIT Operations staff. These were converted to ArcView shape files using the ArcView import wizard function to enable data analysis in ArcView. Other information obtained from the above source included 1:5,000 salvage maps and 1:15,000 colour aerial photographs. Terrain Resource Information Management (TRIM) point elevation maps, created from aerial photos, were obtained from the Forest Information Resources Management Systems laboratory in the Faculty of Forestry, University of British Columbia with the permission of the Base Mapping and Geomatic Services Branch, Ministry of Sustainable Resource Management. Simulated wind speed data on a 1km scale grid and directional data on a 5km scale grid was also obtained from British Columbia Hydro (British Columbia Hydro 2001). 3.2.1.1 British Columbia Wind Resource Data Province wide wind resource maps at 5km and 1km scale, provided by British Columbia Hydro, are numerical simulations obtained using the Mesoscale 42 Atmospheric Simulation System (MASS). The model was initialized using gridded historical weather data (including wind speed and direction, temperature, pressure, and other parameters) at multiple levels from the ground to the top of the atmosphere and predicts wind flow over local topography. The simulation was run for 365 randomly selected days drawn from the period 1984 to 1998. The resulting database included mean wind speeds and frequency distributions by direction at 5km scale for the whole province. This was later refined to a 1 km scale for mean windspeeds at 65m above the ground. The data were validated using 29 Environment Canada stations across British Columbia. It was found that predicted speeds were slightly lower than measured/extrapolated speeds and the r2 correlation between the measured/extrapolated and predicted mean was 0.62 (British Columbia Hydro 2001). This simulated wind data was used as a substitute for true wind data in this study. 3.2.2 Correcting Mapping inconsistencies Harvested polygons on the forest cover and logging history layers were spatially inconsistent by as much as 50m in some cases. This is not surprising with GIS data since the two layers were prepared at different times using different source data. To correct mapping inconsistencies both the forest cover and logging history layers were edited in ArcView to match the boundaries of sample blocks as they appeared on ortho-rectified aerial photographs (Figure 3). 43 a) fj j Log hist_8m EZ1 K22fo rests h CD j K22correctec m _J Wt_all.shp ED _| K22rbsht.tif Figure 3. (a) Example of discrepancies in forest cover and logging history boundaries. (b)Example of logging history and forest cover layers corrected to align with forest boundaries in ortho-rectified aerial photographs. 3.2.3 Creation of Sample Units The sampling frame was all boundaries of cutblocks harvested between 1987 and 1993 on eight 1:20,000 mapsheets. The cut-off dates reflected the year in which silviculture prescriptions were first required (1987) and one year before the aerial photographs were taken (1994 photos). The mapsheets were systematically selected to represent the range of conditions across the operating area. The selected mapsheets accounted for approximately half the area in NIT. In order to evaluate windthrow penetration beyond the cutblock boundaries, three outside buffers of 25, 50 and 75 m were created from the corrected logging history layer. Each buffer was subdivided into sampling units (segments) 25 m deep by 25 m long. 3.2.4 Windthrow Detection and Mapping A magnifying stereoscope (3x magnification) was used to detect windthrow on 1:15,000 nominal scale color aerial photographs taken in September 1994. These photographs were then scanned with a resolution of 2400 dots per inch. This produced high quality images, which were ortho-rectified using the roads layer in ArcView for geographic control. These images were magnified on screen within ArcView and the windthrow was digitized on screen (Figure 4). Percent canopy loss was estimated to the nearest 10% for each windthrow polygon and added to the data-table for the windthrow layer. Large windthrow areas with 45 varying levels of canopy loss were split into multiple polygons. Windthrow salvage maps provided by the company were digitized and formed a distinct windthrow salvage layer. O Attributes of Wt_all.shrj H - r Id Cutbfock ZA>sf SSHEL 1 BB(2] 70 _ i Polygon 3 88(2) !•'70 Polygon 2 BC 60 . . . . . . . . . —i Polygon . . . 11(2) .... Polygon BB(2) **»»> 5 na 20 Polygon 1 BC 20 Polygon 6 BB(2) 50 **s"» 1 5020 40 Polygon 1 5019 60 Polygon 3 5020 30 Po^gon 2 5020" jjj Polygon 13 5001 50 Polygon 5 5006 60 11 5022 70 Polygon 10 5022 70 Polygon 12 5022 60 Polygon 2 5022 60 Poljigon 4 5022 60 Polygon 7 5022 60 Po^ igon 0 4045 70 Polygon 12 4045 60 Figure 4. Example of cutblock edge windthrow on aerial photograph along with ArcView table showing percent canopy lost by windthrow polygon 3.2.4.1 Helicopter Inspection Helicopter inspections of approximately 5% of the damaged cutblock edges were conducted on July 31, 2001 and April 25, 2002. The purpose of these inspections was to verify the status of boundary segments whose condition was unclear on the photographs, and to obtain images of variable retention cutblocks. A video camera and digital camera were used to capture wind damage information. The inspections focused on areas on the aerial photos that were in deep shadow 46 because of steep terrain and hence not very visible, and where canopy gaps were suspected to result from edaphic conditions (wetlands, rocks), or slash burning. Further details on procedures used to assemble data and construct models are contained in Appendix I 3.2.5 Determination of Independent Variables Independent variables are defined in Appendix II. For comparison with an earlier model fit with data from Western Forest Products Port McNeill operations, it was necessary to create some class variables using the NIT dataset (see Appendix III for decisions rules used in creating class variables). Creating digital elevation models (DEM) for three different resolutions (50*50, 100*100 and 200*200 m), from TRIM elevation points, was a direct process in ArcView. The topographic variables aspect, elevation, slope, topographic exposure and profile curvature were then derived from the resulting DEM for each resolution. 3.2.5.1 Topex-to-Distance Calculations Classical topex is the sum of the angle to skyline in degrees for the eight cardinal directions. The minimum value is zero, meaning that the lowest topex score is 0 47 In topex-to-distance, classical topex is modified in the following ways in order to facilitate map-based calculations: 1. the maximum distance in each direction is limited - in Topex_2K to 2km, in Topex-1Kto 1km, etc.; 2. the greatest angle to ground within this distance is used; and 3. negative values are permitted. A low score still indicates higher exposure - a mountaintop would have a very negative score (see Appendix IV for procedures for calculating Topex-to-distance for map building). Limiting topex distances is necessary for working within a GIS (Quine and White 1998), and customized avenue scripts were used for the calculation (Ruel et al. 2002). To test the effect of distance and resolution, topex-to-1 km, topex-to-2 km, and topex-to-3 km from 50 and 100 m resolution DEMs were calculated. 3.2.5.2 Exposure of Edges by Harvesting In the variable DIREX, segments were assigned a value from zero to eight based on how many of the eight cardinal directions from the segment centroid fell within a harvested opening for a distance of at least 100 m. An avenue script was used to generate a table of segment exposure scores in the eight cardinal directions. For example, Figure 5 shows a segment with only the SE, S and SW sides exposed to a distance greater than 100 m and hence will be given a DIREX value 4 8 of three (low directional exposure). Another segment however is exposed in the NE, E, SE , S, SW, and W directions to a distance greater than 100 m and hence will have a DIREX value of six (high directional exposure). The minimum opening distance of 100 m was chosen because it represents the distance at which the sheltering effect of a stand to windward begins to rapidly decline (e.g., Gardiner et al. 1997). The variable S C O R E is the sum of the distances across the opening in meters for each of the eight cardinal directions. Figure 5. Measures of edge exposure due to harvesting. DIREX, sum of number of openings in eight cardinal directions. SCORE , sum of distance within opening for eight cardinal directions. 3.2.5.3 Bearing and Bearing to Centroid Boundary orientation and position variables were calculated in SAS 8.2 (SAS 2001) as follows. The x and y coordinates for the nodes at the beginning and end of each segment were used to calculate the bearing perpendicular to the edge of the segment towards the harvested opening (BRG). The x and y coordinates of the centroid of the segment and the centroid of the harvested opening were used 49 to calculate the bearing to opening centroid (BRGCN). In both cases, the values of these variables ranged from 0 to 360°. Based on the assumption that winds originated from the south, and that east and west facing boundaries experience the same wind loading, these variables were transformed using equations 5 and 6: (5) cosBRG = cosine (BRG) C B R G = arcos (cosBRG) (6) cosBRGCN = cosine (BRGCN) CBRGCN = arcos (cosBRGCN) The net result of these transformations is that boundaries with the greatest exposure to the south take the value of 180°, those exposed to the north take the value 0° and those exposed to the east or west takes the value 90°. 3.2.5.4 Exposure of Edges to Wind The variable ATTACK is the angle between the bearing of a segment boundary and peak wind direction for that location. It was generated from the British Columbia Hydro wind resource directional data and boundary orientation calculated in SAS. The original values ranged from 0 to 360°, but as for BRG and BRGCN this was transformed to values from 0 to180° by means of a cosine-50 arcos conversion to produce CATTACK. For this variable, a boundary facing directly into the wind (highest wind exposure) has a value of 0°, and a boundary facing directly away from the wind (lowest wind exposure) has a value of 180°. Boundaries that are parallel to the direction of wind on either side of the opening have a value of 90°. 3.2.5.5 Buffer Distance Buffer distance is a variable that indicates how far sample points are located into the forest from the edge of a cutblock (Figure 6). This allowed an investigation into the pattern of damage from the immediate edge to any distance worth investigating. Distances of 25, 50 and 75 m were investigated in this study. Cutblock opening < Figure 6. Representation of buffer distances around a cutblock within which sample points are located. Bi, B 2 and B 3 = buffer distance 25, 50 and 75 m respectively. 51 3.2.5.6 SITEBEC The variable SITEBEC, median site index for a given site series, was obtained from Figure 7 (from Greene and Klinka 1994). This variable gives a site index (SI) for the leading species in a given forest cover polygon using the site series from the ecosystem layer. It was included as an alternate variable to the site index estimate included in the forest cover label. Class I Class II 5 0 4 0 CO O 3 0 L O . \ t 2 0 CO 1 0 0 5 0 4 0 oi O 3 0 L O \ 2 0 1 0 • O • • • i - F P A h -Ba Cw Fd Hw S s 0 20 181 66 14 Class III 1 1 1 1 - „ I . . A I ± J ? $ a E ' • 1 » 1 _i • Ba C w Fd Hw S s n - 20 19 157 46 15 5 0 4 0 3 0 2 0 1 0 0 5 0 4 0 3 0 2 0 TO O Ba Cw F d Hw S s n - 73 38 287 322 74 Class IV 1 1 1 1 1 + 1 1 B a C w Fd Hw S s n - 0 24 0 23 0 Figure 7. Site index classes and the distribution of species-specific site index data. Class I represents the highest productivity sites and class IV the lowest. Box plots contain 84% of sample. Ba = amabilis fir, Cw = western redcedar, Fd = Douglas fir, Hw = western hemlock and Ss = Sitka spruce. (From Greene and Klinka 1994) 52 3.2.6 Construction of Segment Database With all necessary layers obtained or created, the database was extracted by overlaying one layer onto the other using the ArcView geoprocessing wizard's "assign data by location" option. A point situated in the center, called centroid, represented each segment. Each centroid representing a segment from the 25, 50, and 75 m buffers then obtained the various unique attributes from the layers. The edge exposure scores from which the variable DIREX was obtained were in a form of a table, not a GIS layer, and this was also joined to the dataset through ArcView's "table-join" function. 3.2.7 Correlation Correlations between continuous variables, except site series, were determined using Pearson's simple correlation coefficients to identify highly correlated independent variables. Site series, which was not continuous as coded in the ecosystem layer, was rearranged in a logical order to make it continuous. Where two variables were highly correlated (>0.7), only one was included in model fitting. 53 3.2.8 Spatial Correlation Neighboring sample units tend to have similar properties, therefore environmental and spatial effects will be related. This condition may cause a model fitted with such data to exhibit spatial autocorrelation (Manly 1991). Controlling for the effect of geographical proximity is necessary in these situations to avoid overfitting. Spatial autocorrelation was determined using semivariance, which expresses the degree of relationship between points on a surface (Carr 1995). Semivariance is half the variance of the differences between all possible points spaced a constant distance apart. For stand variables that vary over small distances, semivariance values increase quickly as distance between points increases. For variables that vary over larger distances such as elevation, semivariance values increase slowly. The semivariance values for crown loss and profile curvature increased rapidly until the fifth cell (125 m distance) and then increased only slightly afterwards (Figure 8,a and b); for aspect, semivariance increased until after the seventh cell while it continued to increase for elevation in all twelve cells (Figure 8,c &d). To reduce the spatial autocorrelation among observations while maintaining reasonable numbers of sample units, the maximum neighborhood over which sample points were selected to estimate a grid node was held at 125m (5 segments). Hence, only one segment was selected for analysis from within a 125 by 125 m panel. 54 <u 450 n •S 300 J 150 X x x x x x x x x (b) X 0 2 4 6 8 10 12 Number of 25m cells o 0.04 0.03 0.02 0.01 0 X X X X x x X X X X X X 2 4 6 8 10 Number of 25m cells 12 (c) S 0.90 4 0.60 = 0.30 0.00 X X X X X X X x X x 10 12 Number of 25m cells (d) 0.9 0.6 0.3 0 X X x x X X 0 2 4 6 8 10 12 Number of 25m cells Figure 8. Semivariance versus the number of 25 m segments or distance for: (a) crown closure lost; (b) profile curvature; (c) elevation; and (d) aspect. 3.2.9 Creation of Datasets With Spatially Independent Observations The entire dataset contained 37,717 points (each point representing the centroid of a 25*25 m segment). After non-forested segments (height less than 10 m) were deleted, 22,304 sample points remained for the 25 m, 50 m and 75 m buffers. From the total dataset of 6715 forested segments within the 25 m buffer, five spatially independent datasets were generated (Table 5). This was done in SAS by assigning points to 125 m by 125 m panels over the entire area. The lowest numbered segment in each panel was selected to create the first dataset. 55 Selected segments were then deleted from each panel. A second dataset was then built using the next lowest numbered segment. This was repeated five times leading to five datasets, where data in each dataset was assumed to be spatially uncorrelated. Table 5. Number of segments (n) used for statistical analysis Process Number of segments Overlay of all coverages in ArcView 37,717 After deleting segments with height less than 10m 22,304 25m buffer segments for contingency tables and 6715 modeling Dataset #1 1512 Dataset #2 1425 Dataset #3 955 Dataset #4 960 Dataset #5 595 3.2.10 Determination of Dependent Variables In order to create dependent variables with two values, damaged or undamaged, variables with different threshold combinations of segment area damaged and percent canopy lost were created (Table 6). These can then be used to generate 56 windthrow risk hazard maps representing the probability of damage to a given severity threshold. Table 6. Summary of response variables and procedures used to create them. Num-ber Procedure Variable Remarks 1 Segments were classified as being windthrown if the centroid of a segment area falls within a windthrow polygon. 2 Segments were classified as being windthrown if the percent of segment area within windthrow polygons exceeded a chosen threshold (e.g. 10%, or 60%). These are segments with greater than 30% of segment area within windthrow polygons and with percentage crown loss to windthrow set at different thresholds (e.g. 20, 50 and 70%). WTCN WTP10 WTP 60 WTT20 WTT50 WTT70 The simplest way of classifying windthrow. Low cut points (e.g. 10%) are used where any level of damage is of interest. Higher cut points (e.g. 60%) are used where only severe damage is of interest. This variable combines area of damage with severity (Appendix V). Cut points chosen reflect the level of windthrow that is significant to management. 3.2.11 Contingency Tables The 25, 50 and 75 m buffers each had a set of segments. These sets of segments were analyzed separately to evaluate how the windthrow pattern changes as one goes further away from the boundary. However, the 50 and 75 m 57 buffers were considered only in the initial data analysis stages (contingency tables), because of the low numbers of segments with penetrating damage. Contingency tables were built using the simplest windthrow event classification (segment centroid falls within windthrow polygon, WTCN, Table 6) against independent variables. For this analysis, continuous independent variables were converted to class variables. 3.2.12 Model Fitting Procedures 3.2.12.1 Multiple Discriminant Analysis Versus Logistic Regression Analysis Two methods of evaluating and analyzing the results were tried and tested. Multiple discriminant analysis (MDA), a statistical technique for classifying observations into mutually exclusive and exhaustive groups on the basis of a set of independent variables (Dillon and Goldstein 1984), was used on a pilot dataset. Principal components analysis (PCA) was used to reduce original number of variables based on the Kaisser-Guttman rule for selecting only significant principal components (Tabachnick and Fidell 2001) and based on correlation matrix. These variables were then used in computing two discriminant classification functions for a segment that will be windthrown or otherwise. The pilot dataset was divided into two: building versus testing datasets. 58 The MDA procedure gave satisfactory results producing a model with 66% correct classification of damage and 80% classification of non-damage when validated with the test dataset. This procedure, however, had some drawbacks. MDA is based on two assumptions, that all predictor variables should have multivariate normality of distribution and that there should be homogeneity of the variance-covariance matrices for all groups. According to Manly (2000), without these assumptions satisfied, the tests of significance to determine if groups (damage/non-damage) are different are not reliable. None of the assumptions could be met by the NIT dataset. Also MDA yields a multivariate equation with no probabilities. Unlike MDA, logistic regression does not necessarily require the above assumptions about the distributions of the predictor variables. In logistic regression, the predictors do not have to be normally distributed, related, or of equal variance within each group, though these may enhance the power of the logistic regression (Tabachnick & Fidell 2001). Not only was this perfect for the NIT dataset but also with logistic regression, the response variable (damage/non-damage) could be generated by several criteria (Table 6) to yield different equations. This in turn allowed for the production of different windthrow risk hazard maps with probabilities for different threshold levels of damage severity. Logistic regression analysis was therefore chosen to analyze the NIT data. 5 9 3.2.12.2 Logistic Regression Analysis A logistic regression model, which allows one to predict a discrete outcome, was used to estimate model coefficients and generate a probability of windthrow occurrence for segments in the 25 m buffer. No model was developed for the 50 and 75 m buffers because of the low number of windthrow segments present. Dataset #1 (1512 observations) was used to fit three different regression models (termed initial models). The three different equations were created by including or excluding important variables based on their significance in a forward selection regression. Datasets #2 to #5 were reserved for testing the accuracy of the models fit with dataset #1. The datasets were kept separate because of spatial correlation considerations. To make best use of the data available, the models were then re-fit with all 6715 segments (combined dataset) using only the variables included in the best performing model from dataset #1 (termed overall models). Variables that occurred repeatedly in the best-fit models were used in fitting equations for four levels of the dependent variable WTP (WTP20, WTP40, WTP60 and WTP80), using the combined dataset. This allowed evaluation of any systematic changes in the coefficients of model parameters as threshold percentage area damaged increased. In other words, this procedure helped to 60 identify which variables were more important for predicting higher severity damage. 3.2.12.3 Goodness of Fit Test The Hosmer-Lemeshow goodness of fit test (H-L test) was used to test goodness of model fit. The H-L test sorts test data segments by the predicted probability of damage, and then groups them into 10 equal sized groups. The segments with the lowest probability values go into the first group (Group 1), and so on, up to the last group (Group 10) which is made up of segments with the highest probability values. A comparison is then made between the actual proportion of segments in each group that were damaged and the average predicted probability of damage for the group. A good model is expected to produce a non-significant Chi-square test result indicating that there is no significant difference between the observed proportions and expected probabilities for both damaged and un-damaged cases (Tabachnick and Fidell 2001). For ten groups as used in this study the critical Chi-square value is 15.507(a = 0.5). A test value below this number indicates good fit. 61 3.2.12.4 C-Value or c-Statistic The c-statistic is comparable to but not quite like an R2 (coefficient of determination) value. This test was used as a measure of the correct prediction strength of a model. It measures the discriminatory power of a logistic equation. It may be interpreted as the probability of a correct classification of a randomly selected pair of cases from each outcome category. It varies from 0.5 (the model's predictions are no better than chance) to 1.0 (the model always assigns higher probabilities to correct cases than to incorrect cases). Thus c is the percent of all possible pairs of cases in which the model assigns a higher probability to a correct case than to an incorrect case (Tabachnick and Fidell 2001). 3.2.12.5 Sensitivity, Specificity, Predictive Value and Cut-off Point The sensitivity of a test in this thesis is the proportion of segments with windthrow which are correctly classified as such. Windthrow segments incorrectly classified as non-windthrow segments are false negatives. The higher the sensitivity, the greater the detection rate and the lower the false negative rate. The specificity of a test is the proportion of non-windthrow segments which are predicted as such. Non-windthrow incorrectly classified as windthrow events are false positives. The higher the specificity, the lower will be the false positive rate and the lower the 62 proportion of non-windthrow segments that will be unnecessarily considered as windthrow segments (Bandolier 1994; Stokes et al. 1995b; Tabachnick and Fidell 2001). The positive predictive value of a model is the probability of a windthrow segment actually being classified as such. While the sensitivity and specificity of a model are constant within the populations under test - and generally wherever the test is performed - the predictive value of a model result depends not only on the sensitivity of the test but also on the prevalence of the condition (windthrow) within the population being tested (Bandolier 1994). The higher the prevalence of the condition the higher the predictive value. A particular cut-off point ranging from 0 to 1 (from classification tables in SAS logistic regression outputs) gives particular percentages for sensitivity and specificity values. Ideally, where it is very important not to miss a windthrow-prone stand, a cut-off point leading to high sensitivity is used. In this study a compromise was made such that sensitivity and overall correct prediction are optimized. 3.2.13 Model Comparison: Robustness and Portability Robustness refers to whether models accurately predict for test datasets not used in model fitting. This quality is, therefore, intrinsic and, in this thesis, answers the question whether the models produce reliable results for test segments from within the NIT operating area not used in model fitting. Portability on the other hand refers to whether models can accurately predict damage for segments in different operating areas. Robustness was tested by means of 63 different internal datasets used in fitting and testing models. That of portability was tested as follows. Two Port McNeill models WTP10 and WTP50 by Mitchell et al. (2001) were used to predict damage on the NIT data and these predictions were compared with the actual outcomes in the NIT data. In a second test, a model was fit for the NIT data using the same variables as for two PM models. The coefficients were then compared for the two sets of models. 3.2.14 Windthrow Hazard Map Creation Adequately fitting, local windthrow risk prediction regression models were developed using stand, site, ecosystem, topographic, wind and management variables. Areas above certain elevations, not represented in the cutblock edge dataset, were masked out. The models predict the probability of damage of a given level of severity to cutblock edges for various combinations of environmental and management conditions. Since the purpose of mapping is to identify those locations on the landscape with higher intrinsic vulnerability, management variables in the models were held constant at values that represent full wind exposure to boundaries across harvested openings. These formulas were exported back to ArcView for map creation. 64 4 RESULTS 4.1 EXPLORATION OF INDEPENDENT VARIABLES: RANGES AND MEANS OF X-VARIABLES Empirical models work best for locations with attributes that fall within the limits of the model fitting data. Summary statistics for key variables for the 25, 50 and 75 m buffers are given in Table 7. Table 7. Values of key variables for the study segments (n = 22,304). Variable Label Mean Std Dev Minimum Maximum A G E (yrs) Aspect (°) A G E 267 102 18 534 ASP100 175 111 0 360 Elevation (m) ELV100 509 219 43 1188 Height (m) HEIGHT 44 14 10 78 Mean wind speed (m/s) MWSPEED 4 0.8 2 8 Profile curvature (°) PRCURV100 0.04 0.16 -1 0.87 Total wedge score (m) S C O R E 1538 740 0 5002 Site index SI 22 5 8 47 Site series SITE S1 3 2 0 14 Sitebec SITEBEC 25 3 8 36 Slope (°) SLP100 17 10 0 52 T o p e x 1 0 0 _ 1 K ( ° ) TPX100 1K 57 40 -100 238 Volume conifers (m3) VOLC 769 397 0 1951 Volume deciduous (m3) VOLD 1 8.5 0 212 * Std Dev = standard deviation 65 4.2 CORRELATION BETWEEN S E L E C T E D VARIABLES There were moderate to strong correlations between several of the variables. Notable amongst them were the positive correlations between height and age, height and site index, volume and age, and topex and profile curvature, as well as the negative correlation between site series (site_s1) and slope (Table 8). Each of these correlations is expected. The poor correlation between the two estimates of site productivity SITEBEC (derived from ecosystem layer) and site index (from forest cover layer) was unexpected. Similarly, the poor correlation between SITEBEC and HEIGHT was unexpected. The non-correlation in these two estimates of site productivity reflects differences in the way they are obtained. Table 8. Correlations between selected independent variables (n = 6715) Variable 1 Variable 2 Correlation coefficient TOPEX 1K PROFILE CURVATURE 0.70 SITE S1 S L O P E -0.4 M W S P E E D ELEVATION 0.46 HEIGHT A G E 0.40 HEIGHT SITE INDEX 0.80 VOLC HEIGHT 0.50 VOLC A G E 0.30 ELEVATION SLOPE 0.50 SITEBEC SITE INDEX 0.13 SITEBEC HEIGHT 0.13 66 Table 9. Most frequent wind directions for the 6715 segments according to British Columbia Hydro wind data Direction Frequency Percent Number of damaged segments (% of segments) SSE 1813 27 380 (21) s s w 1404 21 181 (13) s w 2812 42 270 (10) NW 627 9 42 (7) NNW 59 1 1 (2) Table 9 shows that about 95% of the damaged segments had peak winds originating from southerly directions. This is consistent with the general wind climate for coastal British Columbia. 4.3 CONTINGENCY TABLES Damage was more frequent within the first 25 m of the cutblock (13%) but decreased successively in the 50 and 75 m buffers (Figure 9a). Southeastern facing boundaries were more frequently damaged than northwestern-facing segments (Figure 9b). In the 25 m buffer, stands dominated by cedar were less frequently damaged, and amabilis-fir leading stands were more frequently damaged, than stands dominated by other species. In the 50 m buffer, hemlock stands were damaged more frequently than other species (Figure 10a). Most of the damaged cedar was in the lowest canopy loss category compared to amabilis fir leading stands, which had the highest percent canopy loss. Hemlock stands had medium canopy loss (Figure 11). The proportion of segments damaged increased with increasing number of directions that segments are exposed to 67 (Figure 10 b), increasing block size (Figure 10c), increasing soil fertility (Figure 12c) and increasing soil moisture (Figure 13a). Figure 11a shows segments in the C W H very dry maritime subzone to be the least frequently damaged. Damage within the first 25 m buffer was more frequent in organic soils than other soils. Beyond the 25 m buffer however, morainal soils had a higher proportion of segments damaged compared to organic (Figure 13b). Mature stands with more than 350m 3 volume had a higher proportion of segments damaged than less well stocked stands (Figure 12b). The proportion of segments damaged increased with increasing mean annual windspeed (Figure 13c). (a) 02 T — o 10.1 o 6000 3000 25m 50m 75m Buffer/m (b) 10.16 'tr o 0 O . O 8 Q_ 20OO 1500 * 1000 N E S W BRG Figure 9. Proportion of damaged segments for (a) Buffer, 25 m, 50 m and 75m, and (b) BRG, bearing at right angles to boundary inward towards block for buffer 25 m (°), n = 6,715. # = number of observations; (bar) = proportion; (•) = number of segments. 68 (10a) (10a1) 0.06 B C CY F Spedes (10b) (10b1) 3500 1500% -500 0 1 2 3 4 5 6 7 •rex 3500 1500 =* -500 0 1 2 3 4 5 6 7 •rex (10c) (10c1) 0.15 c: .2 0.1 tz o o. 2 0.05 + Q_ 4500 2500 500 0.06 S M L blk size class • 4000 2000* S M L blk size class Figure 10. Proportion of damaged segments for classes of major independent variables, n = 6,715. # = number of observations; (bar) = proportion; (•) = number of segments, (a) - (c) for 25m segments and (a1) - (c1) for 50 m segments. (a/a1) Species; B = amabilis fir; C = western redcedar; CY = yellow cypress; F = Douglas fir; H = hemlock. (b/b1) Direx = number of exposed directions, 0 - 7 . (c/c1) = block size; S, small = 15 - 50 ha; M, medium =150 - 500ha; L, large = 500 ha+. 69 (11a) 100% 80% 60% 40% 20% 0% 191 19 77 I 12 C Y F Species 572 • 60+ canopy lost • 30-59 canopy lost n 10-29 canopy lost Figure 11. Percent crown canopy lost by species in damaged segments in (a) 25 m buffer and (b) 50 m buffer. B = true, amabilis and grand fir; C = western redcedar; CY = yellow cypress; F = Douglas fir; H = western and mountain hemlock 70 (12a) (12a1) 4500 2000 * -500 CWH CWH C W H M H m m vm 1 vm 2 xm 2 1 Zone Sub Variant 0.04 4- 4000 2000 CWH CWH CWHMH mm vm 1 vm 2 xm 2 1 Zone Sub Variant (12b) M>350 Stock M<350 (12b1) M>350 M=350 S o c k (12c) (12c1) 0.06 Prxr Rich Nutrients f 5000 Poor Rich Nutrients Figure 12. Proportion of damaged segments for classes of major independent variables, n = 6,715. # = number of observations; (bar) = proportion; (•) = number of segments, (a) - (c) for 25m segments and (a1) - (c1) for 50 m segments. (a/a1) = BEC_subzone_variant; CWHvml , CWHvm2, CWHxm2, MH mm1. (b/b1) Stock; I = immature conifers; M>350 = mature conifers with volume > 350 m 3 ; M<350 = mature conifers with volume < 50 m 3 . (c/c1) Nutrients; poor and rich. * C W H v m l = C o a s t a l W e s t e r n H e m l o c k s u b m o n t a n e v e r y w e t m a r i t i m e , C W H v m 2 = C o a s t a l W e s t e r n H e m l o c k m o n t a n e v e r y w e t m a r i t i m e , C W H x m = C o a s t a l W e s t e r n H e m l o c k v e r y d r y m a r i t i m e , M H m m l = M o u n t a i n H e m l o c k m o i s t m a r i t i m e . 71 (13a) (13a1) 4500 2500 500 Dry Fair Wet Moisture Dry Fair Wet moisture (13b) (13b1) 0.18 • | 0 .12 o 2 0 .06 0_ 0 l _ L _ l _ L _ L * J 4- 4 0 0 0 2 0 0 0 tt C & F M O Surface material C 0 .04 -f o •c o g- 0 .02 %— 0_ 4 0 0 0 2 0 0 0 tt C & F M O S u r face m a t e r i a I (13c)' (13c1) 0.3 | 0.2 o o 0.1 • 3 4 5 Mwspeed 4000 2000 * 0.09 • - 0.06 + o O 0 .03 4-: H n H 4 5 0 0 3 0 0 0 1500 0 2 3 4 5 6 + M w s p e e d Figure 13. Proportion of damaged segments for classes of major independent variables, n = 6,715. # = number of observations; (bar) = proportion; (•)= number of segments, (a) - (c) for 25m segments and (a1) - (c1) for 50m segments. (a/a1) moisture; dry, fair and wet. (b/b1) = surface material, C&F, colluvial and fluvial; M, morainal; 0 , organic. (c/d)mwspeed, mean annual windspeed (m/s). 7 2 4.4 MODELS Three sets of logistic regression models (initial models) were fit, for dependent variables WTCN, WTP60, and WTT20 (Table 10; see Appendix VI for WTP60 and WTT20 initial models). Independent variables in the overall best-fit models included topographic, ecosystem, stand and management variables (Table 11). The models for WTCN, WTT20 and WTP60 accurately predicted the damage status (damaged/undamaged) for between 67-73%, 71-73% and 68-72% of the test segments respectively. Testing datasets #2-#5 against models created from dataset #1 yielded good fits (Tables 10 and Appendix VI). For individual segments all models were better at predicting the status of undamaged segments than damaged segments based on a cut-off point of 0.2 and an a of 0.5 (Table 12). For segments grouped on the basis of predicted probability of damage (Hosmer-Lemeshow Goodness of Fit test) there is reasonable correspondence between actual and predicted damage for the 10 groups as shown in Table 13 and graphed in Figure 14. 73 Table 10. Variables and coefficients in initial logistic models (dependent variable is WTCN) Variable WTCN (Model 1) WTCN (Model 2) WTCN (Model 3) Intercept -7.2399 -6.7820 -8.9004 TPX100_1K -0.0071 -0.0054 -0.0066 MWSPEED 0.5377 0.4918 0.5608 Height 0.0289 0.0219 a CBRGCN 0.0126 0.0018 0.0126 DIREX 0.2073 0.2203 0.2149 CATTACK -0.0017 b a SITEBEC b b 0.1079 S1 b -1.4071 b S2 b -0.0654 b S3 b -0.2367 b S4 b 0 b M1 b 0.0160 b M2 b -0.7462 b M3 b -0.5416 b M4 b -12.4660 b Model building n=1512 n=1512 n=1512 data HL G of fit value 3.70 2.94 3.3 c-value ' 0.72 0.74 0.72 %-concordant 71.5% 73.8% 72.0% Model testing #2 #3 #4 #5 #2 #3 #4 #5 #2 #3 #4 Dataset # 2 - 5 Correct 0.71 0.72 0.73 0.72 0.71 0.69 0.70 0.69 0.69 0.67 o.e prediction HLG of fit value 7.9 7.9 9.9 9.8 7.7 13.2 30.1 30.2 12.2 16.6 6.1 *X2 (0.5) critical = 15.507; Cut-off point = 0.2; a = included but not selected, b = not included, S1 = western redcedar, S2 = amabilis fir, S3 = Douglas fir, S4 = Yellow cypress, M1 = organic soils, M2 = colluvial soils, M3 = fluvial soils, M4 = Bedrock. 74 Table 11. Variables and coefficients in the best-fit logistic regression models for combined datasets - overall models. Variable Overall WTCN model based on variables from initial model 3 Overall WTT20 model based on variables from initial model 2 Overall WTP60 model based on variables from initial model 1 Intercept -6.1280 -9.1634 -8.0121 TPX100_1K -0.0028 -0.0076 -0.0040 MWSPEED 0.4972 0.5419 0.5775 HEIGHT 0.0165 0.0331 not included C B R G C N 0.0083 0.0099 0.0109 DIREX 0.2500 not included 0.2625 CATTACK -0.0039 -0.0048 not included SITEBEC not included 0.0955 0.0568 Model building data n=6715 n=6715 n=6715 H L G of fit 6.3 13.2 12.5 c-value 0.68 0.72 0.70 %-concordant 68 % 71% 70% *%2 (o.5) critical = 15.507. See Appendix II for description of independent variables. Note: these equations calculate the 'logistic' value, which is then converted to 'probability' using equations 2 and 3 in Tables Vll. A TO Vll. C. 75 Table 12. Percent of correct predictions for individual segments using test dataset #2, n = 1425. WTCN WTT20 WTP60 Damaged -correctly predicted % of correct predictions usinq test data 52 58 53 Undamaged -correctly predicted 74 74 74 Total correct predictions 71 73 72 Cut-off point 0.2 0.2 0.2 Table 13. Percent of actual damaged segments and predicted probability of damage using test dataset #2 sorted by predicted probability of damage and divided into 10 groups, n = 1425. WTCN WTT20 WTP60 Group Actual Predicted Actual Predicted Actual Predicted 1 1 3 2 1 4 1 2 5 5 4 2 1 2 3 8 6 3 3 4 4 4 10 8 3 4 9 5 5 6 9 4 5 8 7 6 13 12 3 6 8 8 7 14 14 10 8 9 10 8 15 17 8 10 16 14 9 23 21 15 13 13 17 10 26 30 19 20 26 30 76 (a) Figure 14. Predicted probability of damage versus percentage of segments actually damaged for dataset #2 sorted by predicted probability of damage and divided into 10 groups, with 1:1 line for a) WTCN b) WTT20 and c) WTP60. 77 Table 14 shows the results of using the seven variables that occurred repeatedly in the best-fit models to fit equations for four levels of the dependent variable WTP (WTP20, WTP40, WTP60 and WTP80) using the entire dataset. Table 14. Effect of changing area damaged for classifying a segment as 'damaged' on variable coefficients for all independent variables, n = 6715 Parameter WTP20 WTP40 WTP60 WTP80 n 6715 6715 6715 6715 Damaged segments 694 679 551 383 INTERCEPT -8.0823 -8.0925 -9.0218 -8.1944 TPX100 1K -0.0046 -0.00475 -0.00484 -0.00491 DIREX 0.2489 0.2486 0.2685 0.3367 CATTACK -0.00242 -0.00242 0 0 HEIGHT 0.0254 0.0251 0.0232 0.0217 SITEBEC 0.044 0.042 0.0511 0 C B R G C N 0.0101 0.0102 0.0114 0.0111 MWSPEED 0.5289 0.5398 0.5744 0.5478 The seven independent variables were selected by the models for all four levels of WTP with the exception of WTP60, which left out CATTACK and WTP80, which left out CATTACK and SITEBEC. The coefficients for the variables MWSPEED and DIREX and TOPEX increased as segment area damaged increased indicating increasing importance of these variables in predicting high severity damage. The coefficients for the variables HEIGHT and SITEBEC decreased with increasing segment area damaged indicating decreasing importance in predicting high severity damage. 78 4.4.1 Comparing Port McNeill and NIT Models The formulation for the Port McNeill models for WTP10 and WTP50 damage thresholds are given in Appendix Vlll. The Hosmer-Lemeshow test indicates increasing levels of actual damage with increasing levels of predicted damage but over-prediction in each test group (Table 15). The latter result is not surprising since damage was more frequent in Port McNeill, at approximately 24% of segments compared to 13% of segments in the NIT (for a threshold of 10%, e.g. WTP10). The HL G test score indicated poor overall fit, particularly for the WTP10 model. However, the graph showed a reasonable fit for the WTP50 model. (Figure 15). The models were, therefore, refit for NIT data using the variables in the PM models. The resulting models relied heavily on management variables but only slightly on site and stand variables. All models had c-values greater than 0.60, and high Chi-square values (Appendix IX). 79 Table 15. Percent of actual damaged segments and the probability of damage predicted by the Port McNeill model sorted by predicted probability of damage and divided into 10 groups (simple model) using entire NIT dataset, n = 6715. HL G of fit value = 310 HL G of fit value= 119 Group WTP10 WTC10 (PM) WTP50 Actual % WTC50 (PM) Actual % Predicted % Predicted % 1 6 9 5 3 2 9 11 6 4 3 11 14 7 6 4 8 17 10 7 5 14 18 9 9 6 19 20 15 11 7 11 23 10 13 8 10 26 16 17 9 21 30 17 22 10 21 38 23 35 Percent of Port McNei ll's correct predictions for individual segments in the NIT WTP10 WTP50 Damage - 56 46 correctly predicted Undamaged - 55 72 correctly predicted Total correct 56 70 predictions Cut-off point 0.4 0.4 80 (a) (b) 3? 40 i 0 8 16 NIT's actual damage (%) Figure 15. Port McNeill model Predicted probability of damage (simple model) versus percentage of segments actually damaged for combined datasets sorted by predicted probability of damage and divided into 10 groups, with 1:1 linefora) WTCIOand b)WTC50. 81 4.4.2 Windthrow Hazard Map Creation A windthrow hazard map was created from the NIT WTT20 model (Appendix X). The series of formulas incorporated into the ArcView map calculator are described in Appendix Vll. The logit (log of the odds) was calculated and then converted to a probability. The map shows the probability of at least 30% of the area within a boundary segment being damaged with at least 20% canopy lost for cutblock segments oriented perpendicular to prevailing winds that are north of the opening centroid (e.g., with high wind exposure after harvest) within 7 years of harvest. This model was selected for mapping not based on its good statistical fit alone, but also based on visual inspection of distribution of damage as well as its inclusion of height and SITEBEC - the only two stand variables. A large part of NIT has a low (0-0.26) risk of succumbing to such damage (Figure 16) but concentrations of higher risk areas are clearly visible on the map (Appendix X). 80000 -i CD O d 60000 o o 40000 ai 20000 £ 2 0 0-0 .06 0 . 0 6 - 0 . 1 6 0 . 1 6 - 0 . 2 6 0 . 2 6 - 0 . 3 6 0 . 3 6 - 0 . 4 6 0 . 4 6 - 0 . 5 6 Probability Figure 16. Number of 100 m by 100 m cells for forested portions of NIT occupied by each damage probability range. 82 5 DISCUSSION Damage was more frequent in the first 25 m buffer than the 50 m, which, in turn, had more damage than the 75 m buffer. These results mean that Hypothesis 1, that damage will be greater at the immediate edge of cutblocks than further into standing trees, has merit. This is consistent with the literature. For example Moore (1977) stated, "blowdown is more a result of trees which were long protected from the full force of the wind, suddenly being exposed, than of excessively strong or freak winds". The proportion of segments damaged increased with increasing mean wind speed (MWSPEED), with increased exposure of segments to the south (CBRG, CBRGCN), and with increased exposure to local wind direction (CATTACK). Furthermore, the contribution of MWSPEED and DIREX increased in importance for predicting high severity damage. This results is consistent with Hypotheses 2 and 3, that damage will increase with mean wind speed, and as exposure of edge segments within cutblocks increases. At least three of the four measures of wind exposure (MWSPEED, CATTACK, CBRGCN, and DIREX) contributed to each model giving a clear indication of the importance of wind exposure to windthrow risk. This result is consistent with the prediction of other researchers (e.g., Ruel 1995). The greater vulnerability of south facing segments (BRG) is consistent with wind directionality in British Columbia Hydro's wind resource data for the area, which shows a strong southerly component to mean wind speeds. 83 The variable MWSPEED incorporates topographic effects, but was not well correlated with TOPEX_1K (r=0.1). The inclusion of TOPEX_1K in the models with a negative coefficient (negative T O P E X 1 K indicates high topographic exposure) suggests that this variable accounts for some localized topographic sheltering effects not accounted for by MWSPEED. Hypothesis 4, damage will increase with topographic exposure therefore has merit. This is consistent with Mitchell et al. (2001) who used TOPEX_2K. Hannah et al. (1995) found TOPEX_2K to correlate well with annual windspeeds over twenty-one sites in Scotland, and Ruel et al. (2002) found strong correlation between local windspeeds and TOPEX_1K in Quebec. Out of the six forms of TOPEX tested in this study (derived from 50 and 100 m resolutions for distances of 1000, 2000 and 3000 km), only the TOPEX_1K at 100 m resolution was selected for by the models. However, all six TOPEX variables were highly correlated (r > 0.84). Successive ridgelines within NIT generally lie from 1000 to 3000 m apart meaning that the main topographic features with a significant influence fall within this range. More than 90% of the edge segments in NIT were in mature stands with volume greater than 350 m 3 (STOCK=2). Although the NIT data had little variability in terms of stocking, fully stocked mature stands were more frequently damaged than immature stands or mature stands with low stocking. Accordingly, Hypothesis 5, damage will increase with increasing stocking, does not have merit 84 as stated. In the Port McNeill study by Mitchell et al. (2001), damage was most frequent in high-density immature hemlock-amabilis fir stands (90 years old, STOCK=1), least frequent in open-grown cedar-hemlock stands (STOCK=3) and intermediate in frequency in mature stands that occupied 65% of the edge segments (STOCK=2). Other coastal studies (e.g., Moore 1977; Harris, 1989), also found that very dense tall stands were most vulnerable and short open stands least vulnerable. However, the increased frequency of damage in high density immature stands in Port McNeill applied only for the lowest damage severity thresholds. Holmes (1985) studying stands within NIT, on the other hand, reported lower damage with increasing stand density and volume. The increasing frequency of damage in mature stands may reflect the loss of inter-tree crown damping during sway, support of neighbors and the reduction of wind penetration into the stand (e.g., Smith et al. 1987). Hypothesis 8, damage will be lower on organic soils, was not consistent with the results. The proportion of segments damaged in organic soils is less than for mineral soils in the 50 m buffer. The reverse was however the case in the 25 m buffer where damage was more frequent on organic soils. Rooting on organic soils has been found to be weaker than on mineral soils in winching studies (Anderson et al. 1989), which would lead to the conclusion that windthrow would be more common on this soil type for comparable stands. However, both Mitchell et al. (2001) in Port McNeill and Harris (1989) in coastal Alaska found the opposite result. The conflicting results may reflect differences in stand/soil 85 patterns between Port McNeill and NIT. In Port McNeill and coastal Alaska, the organic soils were occupied by low productivity open-grown cedar stands, whereas in NIT they were occupied by moderate productivity hemlock stands. Stands on colluvial and fluvial soils were less frequently damaged than morainal soils in both the 25 m and 50 m buffers. In the Port McNeill study, the opposite was true. Stand types appear to be more important than soils in evaluating risk. Hemlock dominated stands were more frequently damaged than stands dominated by redcedar in both the 25 and 50 m buffers. Redcedar was the least frequently damaged among all species. Accordingly, Hypothesis 9, damage will be less frequent in cedar stands compared to hemlock stands, has merit. The less frequent and less severe damage in stands dominated by cedar and more frequent damage in hemlock and amabilis fir stands observed in this study are very consistent with other coastal studies (e.g., Moore 1977; Harris 1989; Beese 2001; Rollerson & McGourlick 2001; Rowan etal. 2001). The differential between hemlock and cedar was even more pronounced in the 50 m buffer indicating that hemlock stands are more vulnerable to more deeply penetrating damage. Interestingly, hemlock was not the leading species in terms of percent canopy lost. While only rarely damaged, Douglas-fir leading stands had high levels of canopy loss. This is likely because most of the Douglas-fir leading stands were immature, uniform-canopied stands. Where winds were sufficient to cause damage, all trees within the patch blew down. 86 The proportion of segments damaged was higher on wet-rich soils. In all of these cases, the trends in vulnerability observed for a given variable were even more pronounced from the 25 m buffer to the 50 m buffer. The variable SITEBEC, the median site index for a particular site series, was important for three out of the four models with its coefficients positively related to damage probability. Since variables with high correlations were selected against, the coefficients could be interpreted with caution. This means that Hypotheses 7, damage will increase with fertile and moist soils, had merit. This is in agreement with other studies showing increased damage on wet, rich sites (e.g., Holmes 1985; Harris 1989; Mitchell et al. 2001). On both poor and rich sites, there was almost no damage in stands with height classes between 10 and 30m. The probability of damage increased with height, peaked at height class 50-60m and then decreased. Hypothesis 6, damage will increase with stand height, therefore, has merit. This is consistent with Smith et al. (1987) who found that black spruce susceptibility to wind damage increased after a height of 21 m is reached. The contribution of HEIGHT and SITEBEC were less important for predicting high severity damage. It appears that boundary configuration and topographic attributes that affect wind exposure are more important than stand or soil variables in predicting high risk in NIT. Profile curvature, a topographic variable that shows the concave or convex nature of a surface, contributed to one of the three initial models for WTT20. 87 However, this variable was not generally selected. Hypothesis 10, the new variable profile curvature will contribute to damage prediction, therefore, did not have merit. Harris (1989) found flat valley bottoms to be more vulnerable in coastal Alaska and the variable profile curvature was expected to feature well in the models to differentiate between moisture shedding and moisture receiving areas. This variable however had a strong positive correlation (0.7) with TOPEX, which was preferred by the models rather than profile curvature. The high correlation may be an indication that topex and profile curvature were capturing a similar topographic attribute on the landscape and that SITEBEC better captured soil moisture effects. A key objective of this project was to check for model robustness and portability. The models were robust. When tested on datasets from the operating area that were not used for fitting the models, acceptable levels of prediction and good fits were obtained. The Port McNeill model predictions over-predicted the risk of damage relative to actual outcomes in NIT for the WTC10 (low severity) model. However, the relative ranking of low to high-risk groups was reasonable for the WTC50 (high severity) model and the results of this model would be useful to forest managers in NIT for distinguishing high from low risk situations. Hypotheses 11 and 12, the models will be robust within NIT and the Port McNeil models will be portable to NIT, therefore have merit. 88 However, there are several qualifications concerning model portability. The Hosmer-Lemeshow scores indicated poor fit for both the WTC10 and WTC50 models fitted using all data. These high scores reflect the sensitivity of this test to the large number of observations, n=6715, (Tabachnick and Fidell 2001). Using only the variables in the PM models to refit a model for the NIT data yielded a model with mainly management variables. The relative importance of management, stand and site variables differed substantially between the two models. This likely reflects the differences in stand types and topography in the two study areas. Most of the boundaries studied in NIT were in mature stands greater than 35m in height and there were few of the open grown cedar or dense second growth stands typical of Port McNeill. There is no equivalent of the open coastal plain near Port McNeill in NIT. While the Port McNeill model seems portable enough to yield useful results in NIT, the recommendation for now remains to test and refit empirical windthrow risk models in areas with characteristics that differ from the original fitting dataset. The dependent variable used for map production was WTT20. This variable is useful, because it captures both segment area damaged and canopy loss. The best-fit model for WTT20 included stand, site, topographic, management and wind speed variables and had the best prediction rate of all the models. The map resulting from this model was examined and compared to maps resulting from other well fitting models. The pattern of higher risk damage areas produced with 89 this model was more logical in relation to the landscape features such as elevation, topex and salvage locations than for other models. The models predict the probability of damage to cutblock edges under given combinations of environmental and management conditions. Stand level data for these maps are obtained from broad scale inventories indicating conditions at the stand level and not at the microsite or tree level. Furthermore, wind damage detection and interpretation from aerial photograph may miss low levels of damage. Such damage might be important in riparian areas or areas of unstable terrain. Hence, the hazard maps are intended for strategic planning during the development plan stage and windthrow risk and potential impacts should be evaluated in the field during cutblock layout. 90 6 CONCLUSIONS Windthrow is a product of complex interactions of climate, site and management factors. Geographic information systems and aerial photograph interpretation provide the opportunity to gather stand or landscape level information for large areas, generate spatial variables, analyze and map the results, avoiding laborious and expensive ground surveys. Adequately fitting, local windthrow risk prediction models can be developed using stand, site, ecosystem, topographic, wind and management variables. New management variables (CATTACK and DIREX), and simulated wind data were useful new additions to predictive modeling. Spatial autocorrelation was found to be significant for the variables under study, and should be taken into consideration in the construction of model-fitting and testing datasets. As expected, damage was more frequent in the 25 m buffer (13%) around the cutblock edge and decreased further into standing timber. As has been found in other studies, exposure to wind due to boundary configuration and topographic position are major contributors to damage prediction. While the wind data are still very coarse resolution and represents mean rather than extreme winds, it contributed substantially to damage prediction. 91 The main discrepancy between the results of the NIT study and those of the Port McNeill study was in the relationship between damage and soils and stand attributes. Since the results of the GIS/aerial photo studies were consistent with ground-based studies conducted in the same areas (Rollerson and McGourlick (2001) in Port McNeill and Holmes (1985) in NIT) it seems that the differences reflect true differences in soil and stand interactions in these locations. In spite of these interactions, richer soils and taller stands were more vulnerable to damage in both Port McNeill and NIT, and redcedar leading stands were less damaged than hemlock leading stands in terms of frequency and severity in all buffers. A comparison of this study to a similar one in Port McNeill highlighted a potential for portability of empirical models to areas with broadly similar conditions and with similar levels of damage for given severity thresholds. However, the differences in coefficients suggest caution. Until further studies on portability are done, it will be safe to say that the windthrow risk prediction model works best for landscape sub-units within the area from which data is obtained. Windthrow management requires prediction of damage, assessment of consequences, and mitigating strategies. Although windthrow damage is complex, it is predictable. The resulting hazard map for the North Island Timberlands should be useful for strategic development planning. The regression equations can also be used in spreadsheet form for evaluating different cutblock designs during cutblock layout, however they would be more useful in a fully 92 developed GIS-based cutblock design tool. Since the models only predict the probability of damage of a given severity, it is still necessary for managers to assess the consequences of damage and design mitigative treatments that reduce the risk of damage to acceptable levels. How forest managers can best do this remains a challenge to be investigated in operational research. 93 7 RECOMMENDATIONS The following recommendations are proposed for those who share an interest in windthrow research: 1. Windthrow is complex, but can be predicted. Hence, windthrow hazard maps should be developed and used during development planning stages and as a guide in silvicultural prescriptions in operating areas with a history of wind damage. 2. The location, timing, severity and orientation of windthrow should be systematically detected and mapped. This feedback can provide a way of validating and refining the hazard ranking and mapping process. 3. Wind exposure is a major contributor to the models. Simulation of extreme wind speeds and direction should be completed for windthrow prone areas of the province. 4. This system of ranking and evaluating risks should be automated into a computer program or routine such that it can be easily used to create hazard maps for other windthrow prone forest areas. 94 5. The issue of model robustness and portability should be further investigated with data from other areas with the aim of developing generic models for coastal BC and for portions of the BC interior. 6. The application of variable retention in British Columbia means that models for cutblocks with smaller openings and reserve patches should be developed in the very near future. 7. This process of evaluating the relative risk of cutblock designs should be incorporated into a GIS-based tool for designing cutblocks, which takes visual quality, aesthetics and other factors into consideration. 95 8 R E F E R E N C E S Alexander, R.R. 1964. Minimizing windfall around clear cuttings in spruce-fir forests. For. Sci. 10:130-142. Anderson, C. J., Coutts, M.P., Ritchie, R.M., Campbell, D. J., 1989. Root extraction force measurement for Sitka spruce. Forestry 62:127-137. Bandolier L. 1994. Testing a test. Bandolier Journal Library. http://www.ir2.ox.ac.uk/bandolier/band3/b3-1.html. April 1994;3-1. Beese, W. J. 2001. Windthrow monitoring of alternative silvicultural systems in the montane coastal forests. In Proceedings of the windthrow researchers workshop. S.J. Mitchell and J. Rodney {compilers). Jan 31- Feb. 1,2001 Richmond, B.C. University of British Columbia Faculty of Forestry and Forestry Continuing Studies Network, Vancouver BC. Pp. 2-11. British Columbia Hydro. 2001. Wind energy resource mapping consulting services for British Columbia. True Wind Solutions, LLC. Albany, New York. pp8. British Columbia Ministry of Forests. 1996. Land Management Handbook. Field Guide Insert 3. Site index curves and tables for British Columbia - coastal species. Ministry of Forests research program. Victoria, BC. ,2001a Resources Inventory Branch. Resources Inventory Committee website: 2001 a.http://www.for.gov.bc.ca/resinv/homepage.htm. , 2001b. Tree Farm Licence 39. Rationale for the annual cut determination. Ministry of Forests. Victoria, BC. 96 Brokaw, N. V. L. and L. R. Walker. 1991. Summary of the effects of Caribbean hurricanes on vegetation. Biotropica 23:422-427. Burrough, P.A. and R.A. McDonnell. 1998. Principles of geographic information systems. Oxford Univ. Press, New York. Canned, M. and M. Coutts. 1988. Growing in the wind. New Scientist 117:42-46. Carr, J. R. 1995. Numerical analysis for the geological sciences. Prentice Hall, Englewood Cliffs, New Jersey, USA. Chen, J. M., T. A Black, D. Novak, and R. S. Adams. 1995. A wind tunnel study of turbulent airflow in forest clearcuts. In Wind and Trees. M. P. Coutts and J.Grace (eds.). Cambridge University Press. Cambridge, pp. 71-87. Cook, N. J. 1985. The designer's guide to wind loading of building structures; background, damage survey, wind data and structural classification. Building Research Establishment Report. Butterworth, Sevenoaks. Coutts, M. 1983. Root architecture and tree stability. Plant and Soil 71:171-88. Cremer, K. W., C. J. Borough, F. H. McKinnel and P. R. Carter. 1982. Effects of stocking and thinning on wind damage in plantations. N. Z. J. For. Sci. 12:245-268. Dean, J. D. and E. D. Ford. 1983. Modelling root structure and stability. Plant and Soil 71:189-195. 9 7 Dietrich, W., D. M Windsor & T. Dunne. 1982. The physical environment and hydrology of Barro Colorado Island. In The ecology of a tropical forest: sensational rhythms and longer-term changes. Leigh, E. G. J., D. M. Windsor and A. S. Rand (eds.). Washington DC, Smithsonian Institute Press. Dillon W. R. and Goldstein M., 1984. Multivariate analysis methods and applications. John Wiley and Sons, Toronto. Ennos, A. R. 1995. Development of buttresses in rainforest trees: the influence of mechanical stress. In Wind and Trees. M. P. Coutts and J. Grace (eds.). Cambridge University Press. Cambridge, pp. 293 -301. Environment Canada. 1989. Climatological station catalogue: British Columbia. Atmospheric Environment Service. Downsview, Ontario. Environment Canada. 1992. Marine weather hazards manual, a guide to local forecasts and conditions, west coast edition. 2 n d ed. Gordon Soules Book Publishers Ltd., Environment Canada and the Canada Communication Group-Publishing, Supply and Services Canada. West Vancouver, BC. Environment Canada. 1993. British Columbia climate normals 1961-1990. Atmospheric Environment Service. Downsview, Ontario. Environment Canada. 1994. Canadian monthly climate data. Atmospheric Environment Service. Downsview, Ontario. ESRI, 2000. Using Arcview GIS. Environmental Systems Research International Inc., Redlands, CA. 98 Everham III, E. M. 1995. A comparison of methods for quantifying catastrophic wind damage to forests. In Wind and Trees. M. P. Coutts and J. Grace (eds). Cambridge University Press. Cambridge, pp. 340-357. Everham, E. E. and N. V. L. Brokaw. 1996. Forest damage and recovery from catastrophic wind. The Botanical Review 62:114-185. Fleming, R. L. and R. M Crossfield. 1983. Strip cutting in shallow-soil upland black spruce near Nipigon, Ontario. Windfall and mortality in the leave strips: preliminary results. Information report O-X-354. Great Lakes Forest Research Centre. Ontario. Foster, D. R. 1988. Species and stand response to catastrophic wind in central New England, USA. J. Ecol 76:135-151. Foster, D. R. and E. R Boose. 1995. Hurricane disturbance regimes in temperate and tropical forest Ecosystems. In Wind and Trees. M. P. Coutts and J. Grace(eds.). Cambridge University Press. Cambridge, pp. 305-339. Fridman, J. and E. Valinger. 1998. Modelling probability of snow and wind damage using tree, stand and site characteristics from Pinus sylvestris sample plots. Scan. J. For. Res. 13:348-356. Gardiner, B.A., G.R. Stacey, R.E. Belcher and C.J. Wood. 1997. Field and wind tunnel assessments of the implications of respacing and thinning for tree stability. Forestry 70:233-252. 99 Greene, R.N., and K. Klinka. 1994. A field guide for site identification and interpretation for the Vancouver forest region. Land management handbook number 28. BC Ministry of Forests, Research Branch. Victoria, BC. Guitard, D. G. E. and P. Castra. 1995. Experimental analysis and mechanical modelling of wind-induced tree sways. In Wind and Trees. M. P. Coutts and J. Grace (eds.). Cambridge University Press. Cambridge, pp. 182-194. Hall, J.B., P.C. Pierce and G.W. Lawson. 1971. Common Plants of the Volta Lake. University of Ghana, Department of Botany, Legon. Hannah, P., J.P. Palutikof and C P . Quine. 1995. Predicting windspeeds for forest areas in complex terrain. In Wind and Trees. M. P. Coutts and J. Grace (eds.). Cambridge University Press. Cambridge, pp. 113-129. Hanns-Christof, S. and F. Bruechert. 2000. Basic biomechanics of self-supporting plants: wind loads and gravitational loads on a Norway spruce tree. For. Ecol. and Manage. 135:33-44. Harris, A.S. 1989. Wind in the forests of southeast Alaska and guides for reducing damage. USDA For. Serv. PNW-GTR-224. Holmes, R. H. 1985. An analysis of windthrow along clearcut boundaries in the Tsitika watershed. B.A. thesis, University of British Columbia, Vancouver, B.C. 100 Howes, D.E. 1981. Terrain inventory and geological hazards: northern Vancouver Island. Province of British Columbia, Ministry of Environment. APD Bulletin 5:9-26. Ingestad, T. and G. I. Agren. 1991. The influence of plant nutrition on biomass allocation. Ecol. Applic. 1:168-174. Jensen, V.S. 1941. Hurricane damage on the Bartlett Experimental Forest. USDA Northeastern Forest Experiment Station Technical Note 42. Kramer, M.G., A.J. Hansen, M.L. Taper, and E.J. Kissinger. 2001. Abiotic controls on long-term windthrow disturbance and temperature rain forest dynamics in South Alaska. J. Ecol. 82:2749-2768. Laiho, O. 1987. Susceptibility of forest stands to windthrow in southern Finland. Folia For. 706:23-24. Lekes, V and I. Dandul. 2000. Using airflow modelling and spatial analysis for Defining wind damage risk classification (WINDARC). For. Ecol. and Manage. 135:331-344. Manly, B. F. J. 1991. Randomization and Monte Carlo methods in biology. Chapman & Hall, New York, New York, USA. Manly, B. F.J., 2000. Multivariate statistical methods a primer. 2 n d edition. Chapman & Hall, New York, New York, USA. 101 Matheck, C. 1989. Engineering components grow like trees. Kernforschungszentrum Karlsruhe, KfK 4648. Cited in Telewski, F. W. 1995. Wind induced physiological and developmental responses in trees. In Wind and Trees. M. P. Coutts and J. Grace (eds.). Cambridge University Press. Cambridge, pp. 237 -263. Matheck, C. 1991.Trees: the mechanical design. Springer-Verlag. Berlin. Mayhead, G. J. 1972/73. Windthrow problems. Coedwigwr,25:16 - 20. Cited in Smith et al. 1987. The practice of silviculture: applied forest ecology. 9 t h ed. John Wiley and Sons. New York. Mayhead, G. J., J. B. H. Gardiner, and D. W. Durrant. 1975. A report on the Physical properties of conifers in relation to plantation stability. Forestry Commission Research & development Division, Edinburgh. Mergen, F. 1954. Mechanical aspects of windbreak and windfirmness. J. For. 52:119-125. Metzger, K. 1893. Der wind als massgebender Faktor fur das Wachsthum der Baume. Mundener forstl. Hefte 3, 35 - 86. Cited in Telewski, F.W., 1995. Wind induced physiological and developmental responses in trees. In Wind and Trees. M. P. Coutts and J. Grace (eds.). Cambridge University Press. Cambridge, pp. 237 -263. Miller, K.F. 1985. Windthrow hazard classification. Forestry Commission leaflet 85. Forestry Commission, London. 102 Milne, R.1995. Modelling mechanical stresses in living Sitka spruce stems. In Wind and Trees. M. P. Coutts and J. Grace (eds.). Cambridge University Press. Cambridge, pp. 165-181. Mitchell, S. J. 1995. A synopsis of windthrow in British Columbia: occurrence, implications, assessment and management. In Wind and Trees. M. P. Coutts and Grace (eds.). Cambridge University Press. Cambridge. pp. 448-559. Mitchell, S. J. 1999. Assessing and promoting windfirmness in conifers in British Columbia. Ph.D.Thesis, University of British Columbia, Vancouver, B.C. Mitchell, S. J. 2000. Stem growth response in Douglas-fir and Sitka spruce following thinning: Implications for assessing wind-firmness. For. Ecol. and Manage. 135:105-114. Mitchell, S. J., T. Hailemariam and Y. Kulis. 2001. Empirical modelling of cutblock edge windthrow risk on Vancouver Island, Canada, using stand level information. For. Ecol. and Manage. 154:117-130. Moore, M. K. 1977. Factors contributing to blowdown in streamside leave strips on Vancouver Island. BC Min. For. Land. Manage. Rep. No. 3. Victoria, BC. Moore, J. R. 2000. Differences in maximum resistive bending moments of Pinus radiata trees grown on a range of soil types. For. Ecol. and Manage. 135:63 -71. Moore, J. R. and A. Somerville. 1998. Assessing the risk of wind damage to plantation forests in New Zealand . N. Z. For. 43:25-29. 103 Murphy, B, and P.L. Jackson. 1997. Extreme value analysis, return intervals of severe wind events in the central interior of British Columbia. McGregor Model Forest Association. Prince George, BC, Canada. Oliver, H.R. and G.J. Mayhead. 1974. Wind measurements in a pine forest, during a destructive gale. Forestry 47:185-195. Papesch, A. J. G. 1974. A simplified theoretical analysis of the factors that Influence windthrow of trees. 5 t h . Australian conference on hydraulic and fluid mechanics. Univ. Canterbury, New Zealand, pp. 235-242. Peltola H. and S. Kellomaki. 1993. A mechanistic model for calculating windthrow and stem breakage of Scot pines at stand edge. Silva Fennica 27:99-111. Peltola H., B. Gardiner, S. Kellomaki, T. Kolstrom, R. Lassig, J. Moore, C. Quine and J. Ruel. 2000. Wind and other abiotic risks to forests. For. Ecol. & Manage. 135:1-2. Persson, P. 1972. Stand treatment and damage by wind and snow - survey of younger thinning Experiments. Department of Forest Yield Research. Royal College of Forestry. Stockholm. Research Notes. No. 23. Putz, F. E., P. D. Coley, K. Lu, A. Montalvo and A. Aiello. 1983. Uprooting and snapping of trees: structural determinants and ecological consequences. Can. J. For. Res. 13:1011-1020. Quine, C. P. 1995. Assessing the risk of wind damage to forests: practice and pitfalls. In Wind and Trees. M. P. Coutts and J. Grace (eds.). Cambridge University Press. Cambridge, pp. 379-403. 104 Quine, C. P. and M. S. White. 1998. The potential of distance-limited topex in the prediction of site windiness. Forestry 71:325-332. Rizzo, D.M. and T.C. Harrington. 1988. Root movement and root damage of red spruce and balsam fir on subalpine sites in the White Mountains, New Hampshire. Can. J. For. Res. 18:991-1001. Rollerson, T.P. 1979. Queen Charlotte Woodlands Division windthrow study. Nanaimo: MacMillan Bloedel Limited, Woodlands Services. Rollerson, T.P. and K. McGourlick. 2001. Riparian windthrow - North Vancouver Island. In Proceedings of the windthrow researchers workshop, S.J. Mitchell and J. Rodney (compilers). Jan 31- Feb. 1,2001 Richmond, B.C. University of British Columbia Faculty of Forestry and Forestry Continuing Studies Network, Vancouver, BC. pp. 139-156. Rowan, C , S. J. Mitchell and T. Hailemariam. 2001. Edge windfirming treatments in coastal British Columbia. In Proceedings of the windthrow researchers workshop. S.J. Mitchell and J. Rodney (compilers) Jan 31- Feb. 1, Richmond, B.C. University of British Columbia Faculty of Forestry and Forestry Continuing Studies Network, Vancouver. BC. pp. 205-222 Ruel, J.C. 1995. Understanding windthrow: silvicultural implications. For. Chron. 75:434-445. Ruel, J . C , S.J. Mitchell, M. Dornier. 2002. A GIS based approach to map wind exposure for windthrow hazard rating. North. J. Appl. For. 19:183-187. 105 SAF. 1993. Task force report on sustaining long-term forest health and productivity. Society of American Foresters, Bethesda, MD. Cited in Quine, C. P. 1995. Assessing the risk of wind damage to forests: practice and pitfalls. In Wind and Trees. M. P. Coutts and J. Grace (eds.). Cambridge University Press. Cambridge, pp. 379-403. Edinburgh. SAS Institute Inc. 2001, SAS/STAT User's Guide, Version 8.2 edition. SAS Institute Inc. Cary. NC. Slodicak, M. 1995. Thinning regimes in stands of Norway spruce subjected to snow and wind damage. In Wind and Trees. M. P. Coutts & J. Grace (eds.). Cambridge University Press. Cambridge, pp. 436-447. Smith, D.M., B.C. Larson, M.J. Kelty and P.M. Ashton. 1997. The practice of silviculture: applied forest ecology. 9 t h ed. John Wiley and Sons, New York. Smith, K. 1992. Environmental Hazards: assessing risk and reducing disaster. Routledge, London. Smith, V. G., M. Watts and D. F. James. 1987. Mechanical stability of black spruce in the clay belt region of northern Ontario. Can. J. For. Res. 17: 1080-1091. Somerville, A. 1979. Root anchorage and root morphology of Pinus radiata on a range of ripping treatments. N.Z. J. For. Sci. 9:294-315. Stathers, R.J., T.P. Rollerson and S.J. Mitchell. 1994. Windthrow handbook for British Columbia forests. BC Min. For. Research Branch working Paper 9401, Victoria, BC. 106 Stokes, A., A. H. Fitter and M. P. Coutts. 1995a. Response of young trees to wind: effects on root growth. In Wind and Trees. M. P. Coutts and J. Grace (eds.). Cambridge University Press. Cambridge, pp. 264-275. Stokes M.E., C S . Davis and G.G. Koch. 1995b. Categorical data analysis using SAS system, SAS Inc., Cary, NC. Studholme, W. P. 1995. The experience of management strategy adopted by the Selwyn Plantation Board, New Zealand. In Wind and Trees. M. P. Coutts and J.Grace (editors). Cambridge University Press. Cambridge, pp. 468-476. Tabachnick, B. G. and Fidell L. S., 2001. Using multivariate statistics. 4 t h edition. Allyn and Bacon, Toronto. Talkkari, A., H. Peltola, S. Kellomaki and H. Strandman. 2000. Integration of Component models from the tree, stand and regional levels to assess the risk of wind damage at forest margins. For. Ecol. and Manage. 135: SOS-SIS. Telewski, F. W., 1995. Wind induced physiological and developmental responses in trees. In Wind and Trees. M. P. Coutts and J. Grace (eds.). Cambridge University Press. Cambridge, pp. 237 -263. Telewski, F. W. and M. J. Jaffe.1986. Thigmomorphoginesis: field and laboratory studies of Abies fraseri in response to wind or mechanical perturbation. Physiologia Palantarum 66: 211-18. 107 Valinger, E., L. Lundqvist and L. Bondesson. 1993. Assessing the risk of snow and wind damage in Scots pine stands using tree characteristics. Forestry 66: 249-260. Waring, R.H. and W.H Schlesinger. 1985. Forest ecosystems: concepts and management. Academic Press, New York. Whitney, R. D. 1989. Root rot damage in naturally regenerated stands of spruce and balsam fir in Ontario. Can. J. For. Res. 19:295-308. Wilson, J.D. 1994. Determining a topex score. Scott. For. 384:251-256. Wood, C. J. 1995. Understanding wind forces on trees. In Wind and Trees. M. P. Coutts and J. Grace (eds.). Cambridge University Press. Cambridge, pp. 133-164. Wright, J. A. and C. P. Quine, 1993. The use of a geographical information system to investigate storm damage to trees at Wykeham Forest, North Yorkshire. Scottish Forestry. 47:166-174. 108 APPENDICES 1 0 8 a APPENDIX I SUMMARY OF PROCEDURES USED TO BUILD THE WINDTHROW RISK MODEL Table I. A. Windthrow risk model procedures Step Procedure 1 Information Assembly • Obtained GIS layers for ecosystem, stand, logging history, roads and hydrology data. • Obtained 1:5,000 salvage paper maps. , • Obtained 1:15,000 aerial photographs. • Obtained TRIM point elevation data. • Obtained BC Hydro wind resource data (1km scale). 2 Data Translation to ArcView format • Coverages received in Arclnfo interchange (*.e00) format were converted to ArcView shape files using ArcView import wizard. 3 Correcting Mapping Inconsistencies • Forest cover boundary outlines which where spatially inconsistent with logging history boundaries were corrected with the help of ortho photos. 4 Creation of sample units • Systematically selected eight 1:20,000 mapsheets, out of 20 for the entire NIT, to create the sample dataset. • Created three buffers of 25m, 50m and 75m along the cutblock boundaries from logging history coverage. • Divided each buffer into 25m long*25m deep segments. • Used Avenue Scripts to calculated edge exposure scores using coordinates of segments. 5 Windthrow detection and mapping • Identified and mapped potential edge windthrow on 1:15,000 color aerial photos taken in August/September 1994. • Scanned photographs at 2400dpi, ortho-rectified photos and digitized windthrow. • Estimated percent canopy loss within each windthrow polygon. • Verified windthrow status of 5% of segments in field from helicopter. 109 Table I. A (con't). Windthrow risk model procedures Step Procedure 6 Determination of topographic variables • Produced Digital Elevation Model (DEM) using interpolation between TRIM elevation points. • Determined topographic variables aspect, elevation, slope and ground curvature. • Calculated TOPEX-to-distance scores using Avenue scripts. 7 Construction of segment database • Overlaid coverages with edge segments and extracted segment database. • Kept segments with forested boundaries (height >=10m) for analysis (22,304 for all three segments). 8 • Initial data analysis • Imported database into SAS version 8.2. • Calculated % of segment damaged and created set of response variables. • Created contingency tables. • Calculated correlation between independent variables. • Catered for spatial correlation. 9 Model fitting and testing • Creation of model fitting and model testing data sets independent of each other in terms of spatial correlation. Created 125m*125m panels and retained only 1 segment per panel. • Dataset 1 contained 1512 segments, dataset 2 contained 1425 segments, dataset 3 contained 955 segments, dataset 4 contained 960 segments and dataset 5 contained 595 segments. • Fitted logistic regression models using dataset one. • Tested predictions using other four datasets. • Used variables from best performing model to refit overall model using complete dataset. • Repeated fitting process for 25m buffer with 3 different response variables: damage present at centroid (WTCN), segment area damaged threshold (WTCP60), and segment area damaged threshold plus percent cover lost threshold (WTT20), refer to table 2. • Port McNeill models were compared with the NIT data for portability. 110 APPENDIX II DEFINITION OF INDEPENDENT VARIABLES Table II. A. Summary of independent variables. Variable Name Description Type* BLKAREA Block Area Area in hectares of each block cn ELV50, 100 and 200 m Elevation Elevation value from 50,100,200 cn m resolution DEMs (°) ASP50, 100 and 200 m Aspect Aspect value from 50,100,200m cn resolution DEMs (°) SLP50, 100, and 200 m Slope Slope value from 50,100,200 m cn resolution DEMs (°) TOPEX_1K, 2k and 3K Topographic Topex-to-1km from 50 and 100 m cn exposure to resolution DEMs (°) 1,2 and 3km PRCURV50, 100 and Profile curvature Shows concave or convex cn 200 m 50m surfaces MWSPEED Mean annual 1km windrose windspeed data ct wind speed MS Main Species Main species within a stand cn CLASS Class of type Mature, immature, etc. in cn numerical order SI Site index Describes potential productivity cn of site VOLC Volume Actual vol in m3/ha of conifers cn coniferous AGE Age Age cn HEIGHT Height Height cn C_T_NAME Class type name Mature, immature, etc. ct BGC_ZONE Biogeoclimatic BC BGC system ct zone (BGC) BGC_SUBZ Biogeoclimatic BGC occurring within particular ct sub zone zones 111 Table II. A (con't). Summary of independent variables. Variable Name Description Type BGC_VRT BGC Variant Biogeoclimatic subzones ct occurring within particular zones SITE_S1 Site series Categorized sites based on their ct ability to produce specific climax vegetation within a particular BGC subzone or variant SURFM 1 SURFM 1 Surficial Material Component 1 The formative geomorphological ct process of the first stratum of surficial material of ct component 1 of the current terrain polygon. (Howes & Kenk) TIMELOG DIREX Time since logging DIREX 1994 minus time since logging cn Number of segment exposed directions out of 8 cardinal directions cn BRG;CBRG BRGCN;CBRGCN Bearing; Cosine of Bearing Bearing to opening centroid; cosine of BRGCN Bearing at right angles to cn boundary inward towards block (°); the cosine of BRG Bearing from segment to cn cutblock centroid (°); the cosine of BRGCN SITEBEC Sitebec Median site index for a given site cn series ATTACK; CATTACK Attack; cosine of Angle between boundary bearing cn attack and peak wind direction; the cosine of ATTACK S1 Western red-cedar S2 Amabilis, grand and true firs S3 Douglas fir S4 Yellow cypress M1 Organic M2 Colluvial soils M3 Fluvial soils M4 Other *cn = continuous variable, ct = categorical variable 112 APPENDIX III CREATING CLASS VARIABLES Decision rules for creating Stock class variable: If C_t_name ="IMM CONIF" or C_t_name= "IMM DECID" then STOCK=1 If C_t_name="MAT CONIF" and Volc<=350 then STOCK =3 If C_t_name="MAT CONIF"" and Volc>350 then STOCK =2 Cutblocks were grouped into three size classes: if 1.5ha< BLKAREA <= 15 then block size = small, if 15ha< BLKAREA <= 50 then block size = medium, if BLKAREA > 50 ha then block size = big. Blocks adjacent to each other were dissolved together into single blocks to better represent the wind (fetch) effects within openings. Consequently there are few blocks in the smallest size class. Site series were grouped on the basis of soil nutrient and moisture regime. If BGC_SUBZ_VRT = CWH vm 1 and SITE_S1 in (1,2,3,6,13) then nutrient=1, if BGC_SUBZ_VRT = CWH vm 1 and SITE_S1 in (4,5,7,8,14) then nutrient=2, if BGC_SUBZ_VRT = CWH vm 2 and SITE_S1 in (1,2,3,6,9,10) then nutrient=1, if BGC_SUBZ_VRT = CWH vm 2 and SITE_S1 in (4,5,7,8,11) then nutrient=2, if BGC_SUBZ_VRT = CWH xm 2 and SITE_S1 in (1,2,3,6,11) then nutrient=1, 113 Appendix III (con't) if BGC_SUBZ_VRT = CWH xm 2 and SITE_S1 in (4,5,7,12) then nutrient=2, if BGC_SUBZ_VRT = MH mm 1 and SITE_S1 in (1,2,4,6,8) then nutrient=1, and if BGC_SUBZ_VRT = MH mm 1 and SITE_S1 in (3,5,7,9) then nutrient=2 *BGC_SUBZ_VRT = BEC zone, subzone and variant, *SITE_S1 = site series, 1 = poor-medium, and 2 = rich-very rich. Similarly the class variable moisture was created using the dry, fair and wet divisions of the soil moisture regime of the BEC classification system. Height class Stands were assigned to 10 m height classes using the height variable. Heights greater than 70 m were lumped together into one height class since they are few. 114 APPENDIX IV TOPEX CALCULATIONS For use in producing windthrow risk maps, Topex_2K is calculated using the following steps: 1. DEM grids of 50 50 m and 100100 m resolutions were built directly from TRIM point elevation data using ArcView spatial analyst and 3D analyst extensions (from Arcview 3.2). 2. UTM coordinates for each grid cell and Z value representing elevation were obtained. 3. For each grid cell, within the area of interest the following analysis were carried out: a. the vertical angle from the 'point of observation' grid cell to the grid cell at 2000 m distance to the North was calculated; b. calculation 3 a. was repeated from the point of observation grid cell to the grid cell at 1900m distance to the North, and then for each cell working back towards the point of observation grid cell; c. retained the x,y coordinates, z- value, and vertical angle of the cell for which the largest angle from point of observation cell is obtained; d. calculations a-c were repeated for each of the other cardinal directions: NE, E, SE, S, SW, W and NW. 1 1 5 4. The eight vertical angles obtained using calculations a-d were summed up. This was the TOPEX_2K score for the first point of observation cell. 5. The next cell in the grid was selectedand steps 2 to 5 were repeated. 6. The same procedures above were then repeated for calculating TOPEX_1 K and TOPEX 3K. 116 APPENDIX V NUMBER OF DAMAGED SEGMENTS FOR COMBINATIONS OF S E G M E N T AREA AND CANOPY LOSS Table V. A. Number of damaged segments for combinations of segment area and canopy loss (total n = 6715.with damage on 874 segments). TJ CD Percentage Canopy Lost -1 o (Ti 10 20 30 40 50 60 70 80 90 Total intage S Dam 10 0 0 2 0 0 0 0 0 0 2 intage S Dam 20 0 0 0 0 0 0 0 0 0 0 eg » 30 1 1 1 0 0 3 1 1 0 8 CD T Q . d CD 40 6 3 3 1 2 4 2 0 0 21 50 20 10 4 3 6 10 5 1 1 60 > 60 36 18 12 6 22 15 3 1 2 115 CD CO 70 22 27 15 19 15 16 2 1 0 117 80 26 16 13 8 10 17 5 0 1 96 90 31 27 11 17 9 24 7 1 1 128 100 52 49 47 36 43 78 15 6 1 327 Total 194 151 108 90 107 167 40 11 6 874 117 APPENDIX VI INITIAL MODEL VARIABLES AND COEFFICIENTS (Table VI. A and VI. B) Table VI. A. Variables and coefficients in initial logistic models - WTT20 Variable WTT20 model 1 WTT 20 model 2 WTT 20 model 3 Intercept -5.5130 -8.6316 -8.9470 MWSPEED 0.4819 0.5718 0.5729 C B R G C N 0.0117 0.0117 0.0134 PRCURV 100 -1.4453 b b CATTACK -0.0040 -0.0042 b TPX100_1K b -0.0090 -0.0082 DIREX a b 0.01369 HEIGHT a 0.0210 a SITEBEC a 0.1635 0.1734 Model building n=1512 n=1512 n=1512 data H L G of fit 7.7 11.4 15.1 critical c-value 0.72 0.75 0.73 %-concordant 71.8% 74.3% 72.6% Model testing #2 #3 #4 #5 #2 #3 #4 #5 #2 #3 #4 #5 Dataset #2 - 5 Correct 0.71 0.71 0.71 0.71 0.71 0.73 0.71 0.71 0.73 0.73 0.73 0.73 prediction H L G of fit 15.6 11.4 22.5 15.9 9.1 12.6 11.5 11.0 9.6 32.5 18.87.1 x (0.5) critical = 15.507; cut- off point = 0.2; a = included but not selected; b = not included 118 Table VI. B. Variables and coefficients in initial logistic models - WTP60 Variable WTP60 (Model 1) WTP60 (Model 2) WTP 60 (Model 3) Intercept -9.8543 -9.6693 -5.7657 MWSPEED 0.7386 0.7858 0.7161 CBRGCN 0.0128 0.0128 a CBRGCNXBRG b -0.00002 a TPX100 1K -0.0079 -0.0071 -0.00539 SITEBEC 0.1102 0.1416 b DIREX 0.1842 0.2355 0.2233 HEIGHT b b 0.00129 CATTACK b b -0.00492 Model building n =1512 n =1512 n =1512 data HL G of fit 9.2 7.6 .11.7 critical c-vaue 0.73 0.74 0.75 %-concordant 72.6% 73.8% 74% Model testing #2 #3 #4 #5 #2 #3 #4 #5 #2 #3 #4 #5 Dataset # 2 - 5 Correct prediction 0.70 0.69 0.69 0.69 0.72 0.72 0.71 0.68 0.72 0.71 0.69 0.69 HL G of fit 8.7 8.8 16.0 18.3 14.8 16.4 17.5 32.6 13.6 12.2 17.0 21.4 x^(o.5) critical = 15.507; cut- off point = 0.2; a = included but not selected; b = not included 119 APPENDIX Vll MAPPING FORMULAE FOR THE NORTH ISLAND TIMBERLANDS Tables Vll. A to Vll. C. A series of formulas incorporated into ArcView map calculator for hazard map creation. These formulas first calculate the logit, and then convert it to probability. Table Vll. A. Map formulae used to calculate windthrow risk for WTCN model MAP CALCULATION 1 ((4.AsGrid * 0.250) - (LAsGrid *-6.1280) - ([TPX100 1K] * 0. 0.0028) + ([MWSPEED] *0.4972) + ([HEIGHT] *0.0165) + (180.AsGrid*0.0083) - (1 .AsGrid*0.0039)) MAP CALCULATION 2 (2.71828.AsGrid.Pow([Map Calculation 1])) MAP CALCULATION 3 ([Map Calculation 2] / ([Map Calculation 2]+1)) VARIABLES: a) [TOPEX 1K1000 (100 m resolution)] /m b) [HEIGHT] /m c) DIREX, number of exposed directions (held constant at 4) / dimensionless d) [MWSPEED], mean windspeed / km/h e) CBRGCN, bearing at right angles to the centroid of a block, (also held constant at 180°)/ degrees f) CATTACK / degrees *AsGrid = is used in ArcView map calculations to represent a constant. The ArcView map calculator does all calculations with shapefiles converted to grid layers, hence, constants in equations has to be included as grids by the AsGrid' command. * See Appendix II for definition of variables. 120 Table Vll. B. Map formulas used to calculate windthrow risk for WTT20 model MAP CALCULATION 1 ([MWSPEED] * 0.5419) - (1 .AsGrid * 9.1634) - (1 .AsGrid * 0.0048) + ([HEIGHT] * 0.0331) + (180.AsGrid * 0.0099) -([TPX100JK] * 0.0076) + ([SITEBEC] * 0.0955) MAP CALCULATION 2 (2.71828.AsGrid.Pow([Map Calculation 1])) MAP CALCULATION 3 ([Map Calculation 2] / ([Map Calculation 2]+1)) VARIABLES: a) [TOPEX_1K]/m b) [SITEBEC] / dimensionless c) [HEIGHT] /m d) [MWSPEED] / km/h e) CBRGCN, cosine of bearing at right angles to the centroid of a block (held constant at 180°) / degrees f) CATTACK, angle between segment bearing to cutblock and peak windspeed direction (held constant at 1°) / degrees Table Vll. C. Map formulas used to calculate windthrow risk for WTP60 model MAP CALCULATION 1 ((4.AsGrid * 0.2625) - (TPX100 1K] * 0.0040) - (1 .AsGrid * 8.0121) + ([MWSPEED] * 0.5775) + ([SITEBEC] * 0.0568) + (180.AsGrid*0.0109)) MAP CALCULATION 2 (2.71828.AsGrid.Pow([Map Calculation 1])) MAP CALCULATION 3 ([Map Calculation 2] / ([Map Calculation 2]+1)) VARIABLES: a) [TOPEX_1K]/m b) [SITEBEC] / dimensionless c) [MWSPEED], mean windspeed / km/h d) CBRGCN, cosine of bearing at right angles to the centroid of a block, (held constant at 180°) / degrees e) DIREX, number of exposed directions (held constant at 4). / dimensionless *AsGrid = is used in ArcView map calculations to represent a constant. The ArcView map calculator does all calculations with shapefiles converted to grid layers, hence, constants in equations has to be included as grids by the 'AsGrid' command. * See Appendix II for definition of variables. 121 APPENDIX Vlll PORT MCNEILL EQUATIONS USED IN MODEL COMPARISON (Table Vlll. A and Vlll. B) Table Vlll. A. Port McNeill (PM) simple model #3 Damage Threshold Formula WTC10 -3.2071 + SITE*0.4564 - STOCK*-0.4 + TIMELOG * 0.0943 + BRGP*0.6911. WTC50 -5.2782 + SITE*0.4542 + TIMELOG*0.1164 + BRGP * 0.9526 - TPX2000*0.00997 + COAST * 1.0510 Table Vlll. B. Port McNeill (PM) full model #1 Damage Threshold Formula WTC10 -0.3081 -HEIGHT*0.1713-TOPEX_2K*0.0076-BRGCN*1.2007 - M1*0.9663 + M2*0.7766 + B R G C N 2 * 0.3394 + B R G T I M E L O G * 0.0446 + BRG*SITE*0.2205. WTC50 -3.2022 - TOPEX_2K*0.012 + COAST*0.9417 -M1*0.6161 + M2*0.9472 + BRG*TIMELOG * 0.0585 + BRG*SITE * 0.1342 + BRG*BRGCN*0.1147. * S e e A p p e n d i x II f o r d e f i n i t i o n o f v a r i a b l e s . 122 APPENDIX IX COMPARISON OF PORT MCNEILL AND NIT MODELS (Table IX. A and IX. B) Table IX. A. Percent of actual damage segments and probability of damage predicted by the Port McNeill model sorted by predicted probability of damage and divided into 10 groups (full model) Model testing data n = 6715 H L G of fit critical = 466 WTP50 Actual % WTC50 Predicted % Model testing data n = 6715 HL G of fit critical = 548 Group WTP10 Actual % WTC10 Predicted % 1 9 0 8 0 2 7 0. 6 0 3 7 13 7 13 4 7 89 6 89 5 9 99 9 99 6 11 100 11 100 7 17 100 16 100 8 21 100 20 100 9 13 100 12 100 10 27 100 24 100 % of correct predictions using NIT as test data WTP10 WTP50 Damage -correctly predicted Undamaged -correctly predicted Total correct predictions 82 32 38 82 31 38 Cut-off point = 0.35 123 Table IX. B. Variables and coefficients for PM model refit with NIT combined dataset Variable PM's WTC10 PM's WTC50 PM's WTC10 PM's WTC10 Model #1 on Model #1 on Model #3 on Model #3 on NIT's WTP 10 NIT's WTP50 NIT's WTP 10 NIT's WTP50 Intercept -2.7661 -3.4597 -4.5815 -4.1032 COAST b 0.8694 b 0.8367 SITE b b a a STOCK b b 0.6770 b BRG b b 0.6394 0.6318 HEIGHT a b b b TPX100 2K a b b a C B R G C N a b b b C B R G C N S Q 0.000035 b b b B R G T I M E L O G 0.0329 0.0327 b b CBRGCN*BRG b 0.00178 b b BRG*SITE 0.0607 0.0669 b b TIMELOG b b a 0.0460 M1 a a b b M2 -0.2430 -0.2720 b b n n = 6715 n = 6715 n = 6715 n = 6715 H L G of fit 9.7 18.2 12.6 56.8 critical c - value 0.64 0.67 0.61 0.66 % concordant 63.5% 66.9% 42.5% 64.1% X2(o,5) c r i t i c a l = 1 5 . 5 0 7 . a i n c l u d e d b u t n o t s e l e c t e d b n o t i n c l u d e d 124 APPENDIX X WINDTHROW HAZARD MAP AND LEGEND WINDTHROW RISK FOR NIT - WTT20. Figure IX. A. Windthrow Hazard Map - North Island Timberlands Operation Area. The map show the probability of damage occurring within7 years of harvest to the south facing boundaries of large openings if located in a given 100*100m map cell. For the 'probability for WTT20' (WTT20 equation, Table Vll. B) map, this is the probability of at least 30% of the area within a boundary segment being damaged with at least 20% canopy lost. This highlights moderate to severe damage areas. The map legend is explained in Table X. A. 125 > Table X. A. Explanation of Map Legend Symbol Meaning WTT20 Windthrow probability calculated using the formula in Table VII.B for 100 m grid cells. Height<= 10m Height of the trees less or equal then 10 m. Salvaged Areas that were salvaged after damage according to records. Mapped Windthrow mapped on 1:15,000 color aerial photos. windthrow Non-forest Polygons with descriptor field, NSRA, industry, water, scrub, grassland, rock, slide, swamp and non-productive. Alpine Alpine areas. No data Polygons with no height data - includes recent cutblocks and non-forest types. 126 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0075094/manifest

Comment

Related Items