UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Breeding for improvement of tracheid characteristics in interior spruce Ivkovich, Milosh 2000

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

Item Metadata

Download

Media
831-ubc_2001-611183.pdf [ 9.52MB ]
Metadata
JSON: 831-1.0090560.json
JSON-LD: 831-1.0090560-ld.json
RDF/XML (Pretty): 831-1.0090560-rdf.xml
RDF/JSON: 831-1.0090560-rdf.json
Turtle: 831-1.0090560-turtle.txt
N-Triples: 831-1.0090560-rdf-ntriples.txt
Original Record: 831-1.0090560-source.json
Full Text
831-1.0090560-fulltext.txt
Citation
831-1.0090560.ris

Full Text

BREEDING FOR IMPROVEMENT OF TRACHEID CHARACTERISTICS IN INTERIOR SPRUCE by Milosh Ivkovich B.Sc , The University of Belgrade, Yugoslavia 1991 M . S c , Lakehead University, Canada 1995 A thesis submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in The Faculty of Graduate Studies (Department of Forest Science; Faculty of Forestry) We accept this thesis as conforming to the required standard The University of British Columbia December 2000 © Milosh Ivkovich, 2000 In p r e s e n t i n g this thesis in partial fu l f i lment of the r e q u i r e m e n t s for an a d v a n c e d d e g r e e at the Univers i ty of Brit ish C o l u m b i a , I agree that t h e Library shall m a k e it freely available fo r re fe rence and study. I further agree that p e r m i s s i o n fo r ex tens ive c o p y i n g of this thesis fo r scholar ly p u r p o s e s may b e g ran ted by the h e a d of m y d e p a r t m e n t o r by his o r her representat ives . It is u n d e r s t o o d that c o p y i n g o r p u b l i c a t i o n of this thesis for f inancial ga in shall no t b e a l l o w e d w i t h o u t m y w r i t t e n p e r m i s s i o n . D e p a r t m e n t of T h e Un ivers i ty of Brit ish C o l u m b i a V a n c o u v e r , C a n a d a DE-6 (2/88) ABSTRACT Options for incorporating wood quality in the British Columbia's interior spruce improvement program were investigated. Special attention was given to quantitative variation in tracheid characteristics and its effects on pulp and paper properties. Growth and wood density were examined for 160 half-sib families from Prince George and East Kootenay regions. Tracheid characteristics were examined for 90 of those families, by using an efficient technique for quantitative assessment. Univariate and multivariate R E M L estimates of genetic parameters were obtained. Estimates of genetic variances and heritabilities differed greatly across sites for a number of traits, especially after transplantation between the regions. No significant decrease in heritability was found for ring width and latewood percentage in successive rings. Genetic age-age correlations were generally high for those two traits, with a decreasing trend with increasing difference in age. Within different ring portions, tracheid characteristics had significant family variation. There were limited benefits from considering component traits for avoiding negative genetic correlation between wood density and growth rate. Selection for volume as a single objective could bring significant improvements, positively influencing dry weight, and some pulp and paper properties. Simultaneous improvement of volume growth and other objectives would always require trade-offs, and multiobjective optimization would be beneficial. Expected genetic response in volume was by far the most superior, and any additional objective would need to have a high relative value to justify risk reduction strategies, including diversification through a multiple- population breeding strategy. i i TABLE OF CONTENTS Page A B S T R A C T ii LIST O F T A B L E S viii LIST O F F I G U R E S x L I S T O F A B B R E V I A T I O N S xii A C K N O W L E D G E M E N T S xiii I I N T R O D U C T I O N 1.1 DEFINITION OF THE PROBLEM 1 1.2 RATIONALE 2 1.3 HYPOTHESES 3 1.4 MAJOR GOALS AND SPECIFIC OBJECTIVES 4 II L I T E R A T U R E R E V I E W 2.1 INTERIOR SPRUCE COMPLEX 7 2.1.1 Geographic Distribution And Evolution Of Two Spruce Species and Their Hybrids 7 2.1.2 Genetic Variation in Interior Spruce 8 2.1.2.1 Genetic Differences between British Columbia's Spruce Species 8 2.1.2.2 Variation among Populations 9 2.1.2.3 Variation within Populations 10 2.1.2.4 Variation of Economically Important Traits 10 2.2 PHYSIOLOGICAL PROCESSES AND GENETIC CONTROL OF CONIFER WOOD FORMATION 11 2.2.1 Physiological Processes during Wood Formation 12 2.2.2 Inheritance of Anatomical Traits 13 2.3 RELATIONSHIP BETWEEN GROWTH, WOOD DENSITY AND WOOD ANATOMY 14 3.4.1 Growth Rate and Wood Density in Spruce 14 3.4.2 Wood Density and Wood Anatomy 15 3.4.3 Opportunities for Breaking the Negative Correlation between Growth rate and Wood Density through Breeding 16 2.4 RELATIONSHIP BETWEEN FIBRE CHARACTERISTICS AND PAPER PROPERTIES 16 2.4.1 Chemical Pulps 17 i i i 2.4.2 Mechanical Pulps 19 2.5 M U L T I P L E TRAIT IMPROVEMENT TECHNIQUES WITH SPECIAL EMPHASIS ON BREEDING FOR W O O D PROPERTIES 20 2.5.1 Index Selection 20 2.5.2 Problems Specific to the Use of Selection Indexes in Forestry 21 2.5.3 Multiple-Index Selection 22 2.6 BREEDING STRATGIES FOR M U L T I P L E TRAITS 23 2.6.1 Current Breeding Systems 23 2.6.1.1 Examples of Multiple Population Breeding Systems 24 2.6.2 Risk Management by Using Multiple Population Breeding Stratregy 25 III M A T E R I A L S A N D M E T H O D S 3.1 M A T E R I A L FROM T H E PROGENY TESTS 27 3.2 D A T A SETS AND TRAITS STUDIED 29 3.2.1 Growth Traits and Wood Macro-Properties 29 3.2.1.1 Height Growth 29 3.2.1.2. Radial Growth 29 3.2.1.3 Wood Macro-Properties 30 3.2.2 Tracheid Characteristics 32 3.2.2.1 Traits Based on Fibre Cross-sectional Dimensions 32 3.2.2.2 Fibre Length and Microfibril Angle 35 3.2.2.3 Other Derived Traits 36 3.3 D A T A ANALYSES 37 3.3.1 Preliminary Study: Experimental Design, Sample Size, and Power of Statistical Tests 37 3.3.1.1 Experimental Design 3 7 3.3.1.2 Sample Size and Power of Statistical Tests 3 8 3.3.2 Estimation of Genetic Parameters 40 3.3.2.1. Assumption Checking and Necessary Transformations 4 0 3.3.2.2 Estimation of Variance Components and Heritability 41 3.3.2.3 Estimation of Covariance Components and Genetic Correlations 42 3.3.2.4 Variance and Distribution of Estimated Genetic Parameters 43 3.3.3 Stability of Family Performance 45 3.3.3.1 Genotype by Site Interaction 45 3.3.3.2 Age-Age Correlations 4 5 3.4 RELATIONSHIP BETWEEN G R O W T H R A T E AND W O O D DENSITY AND A N A T O M Y 46 3.4.1 MAXR Analysis of Anatomical Data 47 3.4.2 Principal Component Analysis of Anatomical Data 48 3.4.3 Selection Based on Density Components 48 3.4.3.1 Monte Carlo Simulations 49 iv 3.5 OPTIMIZATION OF SELECTION 49 3.5.1 Definitions of Objective Functions 49 3.5.2 Optimization of Single Objective Functions 52 3.5.2.1 Maximization with Genetic Response as Constraint 52 3..5..2..2 Sensitivity Analyses 53 3.5.3 Multiple Objective Functions 54 3.5.3.1 NIMBUS Methodfor Multiobjective Optimization 55 3.5.3.2 Choosing among Alternatives 56 3.5.3.2 Sensitivity Analyses 58 3.6 D E V E L O P M E N T AND EVALUATION OF BREEDING STRATEGIES 58 3.6.1 Selection Scenarios for Different Breeding Strategies 58 3.6.1.1 Single Breeding Population 59 3.6.1.2 Multiple Breeding Populations 61 I V R E S U L T S 4.1 H E R I T A B L E VARIATION IN G R O W T H , W O O D MACRO-PROPERTIES, AND TRACHEID CHARACTERISTICS 62 4.1.1 Growth and Wood Macro-Properties 62 4.1.1.1 Height Growth 66 AAA.2 Radial Growth 68 4.1.1.3 Wood Macro Properties 69 4.1.1.4 Correlations among the Traits 69 4.1.1.5 Stability of Family Performance 71 4.1.2 Tracheid Characteristics 78 4.1.2.1 Traits Related to Fibre Cross-sectional Dimensions 78 4.1.2.2 Fibre Length and MFA 84 4.1.2.3 Relationships among Anatomical Traits 85 4.2 RELATIONSHIP BETWEEN G R O W T H R A T E AND W O O D DENSITY AND A N A T O M Y 90 4.2.1 Growth Rate and Wood Density: Overall and Ring-by-Ring Relationship 90 4.2.2 Growth Rate and Within-Ring Anatomy 89 4.2.2 A Within Ring Distribution of Cell Wall Thickness and Cell Size 93 4.2.2.2 Linear Prediction of Ratio of Cell Wall Thickness to Cell Size 95 4.2.2.3 Avoiding Negative Genetic Correlation between Growth Rate and Wood Density 97 4.3 OPTIMIZATION OF SELECTION 98 4.3.1 Optimization of Single Objective Functions 98 4.3.1.1 Volume, Dry-Weight and Wood Density 98 4.3.1.2 Volume, Dry-Weight, and Pulp and Paper Properties 100 4.3.2 Multiple Objective Functions 103 4.3..2 A Non-inferior Solutions and Sensitivity Analyses 103 4.3.3 Multiple Breeding Populations 111 v V D I S C U S S I O N 5.1 H E R I T A B L E VARIATION IN G R O W T H , W O O D MACRO-PROPERTIES, AND TRACHEID CHARACTERISTICS 113 5.1.1 Growth and Wood Macro-Properties 113 5 AAA Height Growth 113 5 . 1 . 1 . 2 Radial Growth 1 1 5 5 . 1 . 1 . 3 Wood Macro Properties 115 5 . 1 . 1 . 4 Correlations among the Traits 1 1 6 5 . 1 . 1 . 5 Stability of Family Performance 1 1 7 5.1.2 Tracheid Characteristics 119 5 .1 .2 .1 Traits Related to Fibre Cross-sectional Dimensions 1 1 9 5 . 1 . 2 . 2 Fibre Length and MFA 121 5 . 1 . 2 . 3 Trends for Anatomical Traits 121 5.2 RELATIONSHIP BETWEEN G R O W T H R A T E AND W O O D DENSITY AND A N A T O M Y 122 5.2.1 Growth Rate and Wood Density: Overall and Ring-by-Ring Relationship 111 5.2.2 Growth Rate and Within-Ring Anatomy 121 5 . 2 . 2 . 1 Within Ring Distribution of Cell Wall Thickness and Cell Size 123 5 . 2 . 2 . 2 Linear Prediction of Ratio of Cell Wall Thickness to Cell Size 1 2 4 5 . 2 . 2 . 3 Avoiding Negative Genetic Correlation between Growth Rate and Wood Density 125 5.3 OPTIMIZATION OF SELECTION AND D E V E L O P M E N T OF BREEDING STRATEGIES 125 5.3.1 Relative Gains from Different Selection Scenarios 126 5 .3 .1 .1 Single Breeding Population 12 6 5 . 3 . 1 . 2 Multiple Breeding Populations 1 2 8 V I C O N C L U S I O N S A N D R E C O M A N D A T I O N S 6.1 CONCLUSIONS REGARDING M A J O R G O A L S AND SPECIFIC OBJECTIVES OF T H E STUDY 131 6.1.1 Genetic Control over Growth and Wood Macro-and Micro-Characteristics 131 6 .1 .1 .1 Growth and Wood Macro Characteristics 131 6 . 1 . 1 . 2 Tracheid Characteristics 13 2 6.1.2 Wood Density and Wood Anatomy 132 6.1.3 Effects of Selection on Value of Final Products 133 6 . 1 . 3 . 1 Single Breeding Population 13 3 6 . 1 . 3 . 2 Benefits from Multiple-Population Breeding Strategy 1 3 4 6.2 E V A L U A T I O N OF HYPOTHESES 134 V I I R E F E R E N C E S VIII A P P E N D I C E S v i A P P E N D I X 1: L O C A T I O N O F S E L E C T E D T R E E S F O R E A S T K O O T E N A Y A N D P R I N C E G E O R G E R E G I O N S A P P E N D I X 2: W O O D D E N S I T Y M E A S U R E M E N T : C O M P A R I S O N O F X - R A Y , P H O T O M E T R I C , A N D M O R P H O M E T R I C M E T H O D S A P P E N D I X 3: F A M I L Y P E R F O R M A N C E S O N D I F F E R E N T S I T E S A P P E N D I X 4: M E A N S A N D R A N G E S O F F A M I L Y M E A N S F O R R I N G W I D T H A N D L A T E W O O D P E R C E N T A G E F O R N I N E G R O W I N G S E A S O N S . A P P E N D I X 5: T A B L E S O F V A R I A N C E C O M P O N E N T S A N D H E R I T A B I L I T I E S F O R I N D I V I D U A L S I T E S A P P E N D I X 6: R A N K S O F F A M I L I E S A F T E R S E L E C T I O N A C C O R D I N G T O D I F F E R E N T R I S K R E D U C T I O N S T R A T E G I E S v i i LIST OF TABLES Page 2.2.1 Results regarding the inheritance of wood anatomical traits in Picea sp. 13 2.3.1 Examples of phenotypic and genetic correlations between growth and wood density for white and Engelmann spruce. 14 3.1.1 Identification and description of test sites used in this study. 28 3.2.1 Variables related to fibre characteristics and their abbreviations. 34 3.3.1 General analysis of variance format and expected mean squares for the interior spruce half-sib progeny tests. 38 4.1.1 Site means and ranges of family means, for the East Kootenay progenies on three test locations; And Prince George (PG) progenies on two test locations. 63 4.1.2 Estimated variance components for site, replication within site, family, and site by family effects, and heritabilities with standard errors individualy-based and family-based for East Kootenay progeny tests. 64 4.1.3 Estimated variance components for site, replication within site, family, and site by family effects, and heritabilities with standard errors (individualy-based and family-based for the Prince George (PG) progeny tests. 65 4.1.4 Percentage of trees per site and per replication recurrently attacked by weevil and forked, or heavily attacked by gall-aphids, in 1994 at age 20 for the East Kootenay progenies. Percentage of trees recurrently weeviled and forked , or initially attacked by weevil, in 1992 at age 20, for the Prince George progenies. 67 4.1.5 Genetic and Pearson's correlations (bottom value), for E K and PG progeny tests. 70 4.1.6 Genetic type-B correlations across PG and Q sites. 72 4.1.7 Variance components for ring width and late-wood percentage with at sites RR and JC, for the E K progenies, and RR and Q, for the PG progenies. 75 4.1.8 Site means and ranges of family means for the anatomical traits in East Kootenay, and Prince George progenies. 79 4.1.9 Variance components and heritabilities for tracheid characteristics for the East Kootenay progenies overall across all sites, at Red Rock site, at the Jumbo Creek and the Perry Creek sites. 80 viii 4.1.10 Variance components and heritabilities for tracheid characteristics for the Prince George progenies, for rings 20 and 25combined, overall across all sites, at the Red Rock site, and at the Quesnel site. 81 4.1.11 Eigenvalues for correlation matrix for the EK (left) and PG (right) progeny tests. 87 4.1.12 Eigenvectors for first four principal components for the EK (left) and PG (right) 87 progeny tests. 4.1.13 Pearson's and genetic correlations for tracheid characteristics for EK progenies at RR site. 88 4.1.14 Pearson's and genetic correlations for tracheid characteristics for PG progenies at RR site. 89 4.2.1 Genetic and phenotypic correlations of average relative density with average height and average ring width. 90 4.2.2 Mean statistics describing within ring distribution of duble wall, cell size, and their ratio; Correlation between ring width and those statistics. 94 4.2.3 Analysis of covariance on mean statistics describing within ring distribution of duble wall, cell size, and their ratio. 95 4.2.4 Results of multiple regression analyses for predicting average ring ratio of double cell wall to size from different component traits. 96 4.2.5 Direct and correlated genetic response to selection in ring width and ratio of wall thickness to cell size. 97 4.3.1 Genetic response (A) after one generation of truncation selection based on multiple trait selection: volume and dry weight. 100 4.3.2 Genetic response (A) after one generation of truncation selection based on multiple trait selection: pulp and paper properties. 102 A4.1 Means and ranges of family means for RW (mm), and LW%, for 9 growing seasons. 173 A5.1 Variance components for the East Kootenay progenies. 175 A5.2 Variance components for the Prince George progenies, for ring 25. 176 A5.3 Variance components for the Prince George progenies, for ring 20. 177 A5.4 Variance components for the Prince George progenies, for rings 25 an20 combined. 178 A6.1 Ranks of families after selection based on different objectives. 180 A6.2 Ranks of families after selection based on different objectives and risk reduction strategies. 181 LIST OF FIGURES Page 3.1.1 Location of test sites of the Interior-Spruce tree improvement program in the Prince George and East Kootenay regions. 28 3.2.1 Last latewood and first earlywood of two growth rings. Sampling transects across cell files along which data on wood anatomy were recorded. 34 3.5.1 Flowchart of NIMBUS algorithm. 57 4.1.1 Stability of family performance for the PG and EK progenies for cumulative radial growth (D) at tree age 20. 71 4.1.2a Trends of family means for ring width, for each growing season, and cumulative, for EK and for PG progenies. 73 4.1.2b The trends of family means for latewood percentage, for each growing season and cumulative, for EK and PG progenies. 74 4.1.3 Evolution of variance components with cambial age and resulting individual heritability for ring width and latewood percentage, for at EK progenies at RR and JC sites. . 76 4.1.4 Genetic age-age correlations for ring width and latewood percentage among the EK progenies at RR site. Individual and cumulative increments. 77 4.2.1 Pith to bark evolution of phenotypic and genetic correlations between average latewood percentage, and cumulative radial increments. 92 4.2.2 The average empirical distributions of ratio of cell wall to size, for rings with approximately average and maximum growth rates. 94 4.3.1 The approximate nature of trade off between improvement in volume growth and simultaneous response in tensile strength of produced pulp wet web, based on multiple objective optimization. 104 4.3.2 The approximate trade off between improvement in volume growth and simultaneous response in tear strength of produced paper based on multiple objective optimization. 105 4.3.3 The approximate trade off between improvement in volume growth and simultaneous response in tear strength of produced paper, based on multiple objective optimization. 106 4.3.4 The approximate trade-off line between improvement in volume growth with density held constant and simultaneous response in tensile strength of produced pulp wet web based on multiple objective optimization. 104 4.3.5 Set of Pareto optimal solutions representing trade-offs necessary for simultaneous improvement in volume growth, tensile strength of pulp wet web, and in tear strength of produced paper, based on multiple objective optimization. 109 4.3.6 Selection alternatives and trade-offs followed from an intermediate solution to maximum improvement in volume growth, tensile strength of pulp wet web, andin tear strength of produced paper. 110 4.3.7 Economic weights on objective TRi relative to objective VOL, their probabilities, and population choices A-C. 111 A3.la Plots of family performance in height (H) and diameter (D) at two site locations Jumbo Creek (JC) and Perry Creek (PC) within the EK region. 166 A3.lb Plots of family performance in average ring wdth (AVRW) and relative density (RD) at two site locations Jumbo Creek (JC) and Perry Creek (PC) within the EK region. 167 A3.1c Plot of family performance in average latewoo percentage (AVLW%) at two site locations Jumbo Creek (JC) and Perry Creek (PC) within the EK region. 168 A3.2a Plots of family performance in height (H) and diameter (D)at two site locations Prince George (PG) and Quesnel (Q) within the PG region. 169 A3.2b Plots of family performance in average ring width (AVRW) and average letewood percentage (AVLWP) at two site locations Prince George (PG) and Quesnel (Q) within the PG region. 170 A3.2c Plots of family performance in relative density (RD) at two site locations Prince George (PG) and Quesnel (Q) within the PG region. 171 x i LIST OF ABBREVIATIONS average latewood percentage average relative density average ring width burst factor coarseness cell number cross sectional area radial cell size tangential cell size diameter double wall dry weight East Kootenay earlywood fibre length height heritability selection index Jumbo Creek lumen size latewood microfibril angle Mork's index multiple population breeding cell perimeter Perry Creek Prince George correlation coefficient coefficient of determination ratio DW/CS relative density Red Rock ring width tensile strength of mechanical pulp tensile strength of paper tear strength of paper transition wood tensile strength of pulp wet webs volume X l l ACKNOWLEDGMENTS I would like to thank my supervisor Dr. Gene Namkoong, and advisory committee members Dr. Sally Aitken, Dr. Alvin Yanchuk, and Dr. Simon Ellis for guidance during my research work. I am also very grateful to Dr. Mathew Koshy for his advice and help. Thanks to Dr. Gyula Kiss and Mr. Dave Walden, of BC Ministry of Forests, Kalamalka Research Station, for providing me with study material. Thanks to Ms Jodie Krakowski for her laboratory work. This research study was supported by FRBC research grant FR 96/99-199. I received additional financial support from the van Dusen Graduate Fellowship in Forestry. I am grateful to fellow students and friends who helped, or just gave me moral support, especially to late Sampson Yaw Bennuah, a colleague and a friend, whose life has been inspiring to me in many different ways. xlii if all trees would he tl?e same, T^en all wood would also be the same. An ancient Chinese saying Si tows les arbres seraient identique, lors tout le bois serait eaalement \bent\aue Une enunciation antique de C^inois I INTRODUCTION 1.1 D E F I N I T I O N O F T H E P R O B L E M Interior spruce is the common name for white spruce (Picea glauca (Moench) Voss), Engelmann spruce (P. engelmanni Parry), and hybrids between these two species. It is one of the most heavily harvested trees in British Columbia (B.C.). Wood of interior spruce is used both for lumber and pulp. Long fibres, light colour and low resin content make it an excellent pulping material. Difficulties with natural regeneration of spruce in the interior of B.C. have been encountered by foresters. Even after the application of site preparation techniques, the success of natural regeneration is inconsistent. Artificial reforestation is necessary, and planting is being preferred over seed sowing. The projected annual seedling requirement in year 2000 is estimated to be over 90 million. Only one half of this number will be genetically improved stock (Coates et al. 1994). In 1968, the B.C. Forest Service initiated the interior-spruce breeding program aimed at production of genetically improved material for reforestation purposes. Selection of "plus trees" was done throughout the wide range of the two spruce species, which occupies a large portion of B.C. In general, three selection units were recognised: East Kootenays, Prince George, and Smithers. Clones propagated from the selected trees were established in clone banks in Vernon (southern Okanagan), and Red Rock (Prince George region). Progeny trials were planted on several sites within the area of selection. These progeny trials have generated extremely useful information. Efforts were directed mostly towards examining opportunities for genetic improvement of growth characteristics, and insect resistance (Kiss and Yeh 1988, Kiss and Yanchuk 1991). Wood quality has not yet been thoroughly assessed, and only limited information exists on wood density variation in the progeny trials (Yanchuk and Kiss 1 1993). In the changing economy of forest industries, with a trend towards value added production, wood properties and fibre characteristics may play a significant role. It is now apparent that more attention ought to be given to the simultaneous improvement of the wood quality traits. The purpose of this study is to provide more information necessary for breeding, and to examine the potential for multiple trait improvement. Special emphasis will be placed upon options for incorporating wood quality, (i.e. tracheid characteristics), in the interior spruce improvement program. 1.2 RATIONALE For a geneticist who is trying to improve wood properties there are three main questions to address: 1) "What can be done? - Which fibre properties can be genetically improved and to what extent?" 2) "What should be done? - What are the important fibre properties for different end-products?" 3) "At what cost can the improvement be done?" (Dr. John Hatton personal communication.1) To address the first question, extensive data on genetic variability in wood properties are needed. Only after reliable estimates of relationships among wood and growth traits are obtained, can we have a good basis to assess possibilities for multiple trait selections in the tree improvement program. Regarding the second question, although wood density is generally considered the single most important wood property, there is evidence of anatomical differences among trees of same wood density (Jagels and Telewski 1990). It is not known whether such differences have a genetic basis. Clearly, it is of importance to examine not only the wood density, but also its components, especially fibre characteristics. If variation in wood anatomy that is not reflected in the wood density exists, it is important to estimate the level of inheritance of the anatomical traits and the possibilities for their direct improvement. From the biological point 1 R e s e a r c h S c i e n c i s t , P A P R I C A N , V a n c o u v e r . 2 of view, it is also easier to understand the physiological processes affecting wood density by looking at the component traits. Additional related questions could be asked. For example, how does selection for growth and wood density affect wood anatomy? Can selection be more efficient i f density components are included in selection indices? Do density components have more favourable genetic correlations with growth rate than overall density itself? How are these traits related to the value of final products? Similarly, the utility of improving fibre length is often questioned, since in conifers, fibres are already relatively long and increasing the length through genetic improvement may not warrant the effort. However, breeding might be necessary to maintain the current fibre lengths, especially in short rotation plantations, where mostly juvenile wood is produced. Another important trait for consideration is microfibril angle. Little is known about the inheritance and possibility for improving microfibril angle in spruce. Generally, the utility of incorporating a certain trait into a tree improvement program should be assessed interactively by considering the genetic aspects such as heritability and genetic correlations together with the economic value functions including that trait. The third major question can be addressed by comparing the rapidity and relative effectiveness of different strategies in breeding for paper quality. Simple breeding systems are not appropriate for this purpose because of the complexity and uncertainty regarding the value of end-products as a function of constituent traits. Multiple index selection, employed as part of a multiple population breeding strategy, has been viewed as possibly the most viable option for incorporating multiple traits and multiple objectives into a tree improvement program (Namkoong et al. 1988). However, this technique has not been applied in the existing interior-spruce tree-improvement program, and its advantages and disadvantages need to be evaluated relative to the other more conventional techniques. 1.3 H Y P O T H E S E S Four major hypotheses related to the above questions are formulated as follows: 3 1) There is a substantial amount of genetic variation for growth and wood properties, including fibre characteristics, in the interior spruce populations. Therefore, subtantial genetic gains can be achieved through direct selection for those traits; 2) The use of anatomical component traits, instead of wood density as a composite trait, can result in higher gains from selection, especially i f a negative correlation exists between growth and wood density; 3) The relative importance of traits considered for simultaneous selection is substantially influenced by the property of the value function that relates the traits to product quality; and 4) If complexity and uncertainty regarding the value functions are considered, multiple index selection as part of a multiple-population breeding strategy is more efficient than traditional index selection within a single population. 1.4 M A J O R G O A L S A N D S P E C I F I C O B J E C T I V E S The major goals of this study were: 1) To examine the genetic basis of variation in growth and wood macro- and micro-characteristics among spruce trees, with special attention given to the genetic basis of quantitative differences in fibre characteristics. Fibre properties studied in detail included radial dimensions, length, and microfibril angle. The specific objectives were: a) to estimate genetic variances and heritabilities of, and phenotypic and genetic correlations among growth and wood traits, as well as sampling distributions of these parameters; b) to estimate the magnitude and nature of interaction between environment and genotypes; and c) to examine the pattern of change of genetic parameters for different traits over age and growing seasons. 2) To examine basic relationships of traits related to growth with wood density and its component traits. Overall wood density is the only wood quality trait that is at present considered for selection in the interior spruce tree-improvement-program. The relationships 4 specifically looked at for different families at different ages and in different environments were between: a) height, radial, and volume growth and overall wood density; b) growth rate and wood density on a ring-by-ring basis, i. e., juvenile-mature trend; c) growth rate and within ring distribution of density, i.e., within ring frequency distribution of ratio of wall thickness to cell size; and d) growth rate and other features of wood anatomy. The use of component traits, instead of wood density as a composite trait, for breaking existing negative genetic correlation between growth and density of wood is evaluated. Underlying physiological processes, which affect the relationship between growth and wood properties, could be better understood by considering a variety of basic and composite traits. 3) Based on the relationships between fibre properties and paper quality, examine the effectiveness of selection for the value of final products made from improved trees. The particular objectives are: a) to identify goals of selection and derive approximate value functions; b) to derive economic weights for traits based on maximization of each particular value function, with constraint defined as the maximum genetic response; and c) to examine the relationship between various value functions during the process of single and multiple objective optimization. 4) To examine benefits of developing a multiple-population breeding strategy, aiming at simultaneous improvement of growth and tracheid properties in the tree improvement program. The specific objectives were: a) to derive different scenarios for multiple-trait multiple-objective selection indices involving different value functions; b) to compare the efficiency of the multiple index selection to the conventional selection considering the range of variation in estimated population parameters and precision of value functions; c) to consider also the effects of reduction in size of multiple populations, such as reduced selection intensities, poorer estimates of genetic parameters, changes in genotypic and phenotypic variance-covariance matrices, and the effects of possible inbreeding within each population; and 5 d) to present results, including merits and demerits of each selection strategy, in such a manner so that decision makers are presented with an array of possibilities to chose from. 6 II LITERATURE REVIEW 2.1 I N T E R I O R S P R U C E C O M P L E X 2.1.1 Geographic Distribution and Evolution of Two Spruce Species and Their Hybrids White spruce (Picea glauca [Moench] Voss) has a transcontinental range, and occuring in the northeastern part of British Columbia, generally east of the Coast Mountain range. It extends south to the north Okanagan valley, and down into the USA in the Rocky Mountains. In British Columbia, it grows mostly below 1000m elevation (Coates et al. 1994). Engelmann spruce (Picea engelmannii Parry), on the other hand, is a high elevation species. It has a more scattered distribution, extending from north central British Columbia through the Rocky Mountains into New Mexico and Arizona, and through the Cascade Mountains into California. It may be found as high as 3250m in the central and southern Rocky Mountains (Krajina et al. 1982). Throughout British Columbia, the ranges of the two species overlap and hybridisation occurs at intermediate elevations 600-1500m. In the regions of hybridisation, it is difficult to distinguish the two species from their hybrids, and they are collectively called the "interior spruce". In northwestern British Columbia, the white-Engelmann spruce complex overlaps with the range of Sitka spruce (Picea sitchensis [Bong.] Carr.) and hybridisation, mostly between white and Sitka spruce, occurs. In the northeastern region, the range of white spruce overlaps with that of black spruce (Picea mariana [Mill.] Brit.) and there is also evidence of hybridization. Taxonomic studies, based mostly on differences in cone scale morphology, show a clinal (gradual) variation with the pure species representing the extreme forms. This pattern of variation is a result of introgressive hybridisation, followed by selection and adaptation of a fraction of the hybrid swarms (Coates et al. 1994). White and Engelmann spruce, and their hybrids occur in all but two biogeoclimatic zones of British Columbia. White spruce and hybrids with mostly white spruce characteristics 7 dominate the Boreal White and Black Spruce (BWBS), Sub-Boreal Spruce (SBS), and Spruce-Willow-Birch (SWB) zones of the central and northern interior. Interior spruce is common but not dominant in the Interior Cedar-Hemlock (ICH), Interior Douglas-fir (IDF), and Sub-Boreal Pine-Spruce (SBPS). Engelmann spruce and its hybrids dominate the Montane Spruce (MS) and Engelmann Spruce-Subalpine Fir (ESSF) zones (Coates et al. 1994). A fossil equivalent of modern Engelmann spruce, Picea lahontense, existed from early Oligocene (35 million yr. BP) and into the early Pliocene (5-2 million yr. BP). Engelmann spruce, however, is not a progenitor of white spruce, which is a recent boreal species (Gordon 1996). A l l four spruce species that occur in B.C. survived the ice age either in refugia along the eastern flank of Rocky Mountains in Alberta and B.C. ( P. glauca, P. engelmannii, and P. mariana), coastal refugia (P. sitchensis), and northern refugia (P. glauca and P. mariana) perhaps in the west-central Yukon. It is believed that from these refugia, the species reinvaded the central and south-central B.C. (Neinstaedt and Teich 1972). Outlier populations of white spruce are found in South Dakota, Wyoming, and Montana (Hansen 1955). 2.1.2 Genetic Variation in Interior Spruce 2.1.2.1 Genetic Differences between British Columbia's Spruce Species There are numerous documented differences between white and Engelmann spruce. Cone scale morphology is typically used for distinguishing the two species (Roche 1967). In a chemo-systematic study, terpenes of the two species were found to be distinct, but there was a gradient of intermediate forms between high and low elevations (Ogilivie and von Rudolff 1968). More recently, species-specific restriction fragment length polymorphisms in mitohondrial D N A structure were detected for white, Sitka, and Engelmann spruce. The fragmentation pattern was consistent for all trees representing three spruce species from widely separated populations. Nevertheless, the mitochondrial D N A appears to be very highly conserved based on the number of restriction digests that showed no interspecific polymorphisms (Sutton et al. 1991). In an investigation of growth and phenology of immature populations of diverse geographic origin in British Columbia, white and Engelmann spruces are the extreme forms of a clinal pattern of variation associated with altitude (RocheT967). The author concluded: 8 r 1 "The environmental pressures which result in microevolution, i.e. variation within species, differ only in degree rather than in kind from the environmental pressures which result in macroevolution, i.e. speciation". Macroevolution, however, is often caused, not only by selection pressures, but also through genetic processes caused by isolation or historical events such as populational bottlenecks (Morgenstern 1996). 2.1.2.2 Variation among Populations The broad geographic and ecological range of interior spruce has resulted in a high degree of phenotypic variation among populations. This implies not only a plastic response to the environment, but also local adaptation through natural selection. Differences in aggregation of environmental conditions including photoperiod, temperature, water and nutrient availability, soil composition and other factors have caused adaptational changes among populations in light, water, temperature and nutrient relations. These changes are reflected in clinal variation in phenology, growth, and hardiness associated with environmental gradients. The genetic basis of this type of variation has been studied in provenance trials. It was hypothesized that the genetic constitution of a natural population is to a certain degree determined by the photoperiod and temperature regimes prevailing in that region. Roche (1967) studied genetic differences in germination and seedling growth behaviour among 150 interior spruce provenances in a nursery trial. Early growth was correlated with altitude and some growth measures were correlated with the latitude of provenance. The correlation of flushing with the temperature at provenance origin was much weaker than the correlation of growth cessation or dormancy initiation with photoperiod. The cessation of growth and initiation of dormancy are two critical factors for development and survival of all woody perennials including interior spruce. Generally, a clinal pattern of provenance variation, i.e., gradual change over latitude and altitude, was found. High-elevation provenances are the first to break bud in the spring, the first to cease elongation and set buds in the fall, and the slowest growing (Fowler and Roche 1976, Lester et al. 1990). Population adaptation to different soil types in British Columbia, i.e. acidic and calcareous, was recently identified (Xie et al. 1998). There is evidence that some interior spruce provenances can outperform and be as hardy as local sources, over a wide geographic range. For example, superior provenances 9 have been identified in the "wet-belt transition zone" (east of Williams Lake southwards to Shuswap Lake), and are recommended for planting throughout the interior B.C. south of 55°N latitude (Jaquish 1982, Lester et al. 1990). Genetic differences among populations also arise through mechanisms other than adaptation, including mutation, gene flow, random genetic drift, inbreeding, and hybridisation. In turn, these genetic processes depend in whole or in part on both mating system and effective population size. A study of isozyme variation showed a limited amount of variation among locations within a breeding zone (Fst=0.04), indicating extensive gene flow ( N m = 7). Three sub-populations were sampled, representing each of three watersheds in the Shuswap-Adams low-elevation zone of interior B.C.. (Stoehr and El-Kassaby 1997). 2.1.2.3. Variation within Populations Variation within populations is a mostly consequence of genetic segregation and recombination, and not a result of selection and adaptation. Spruce is an outcrossing conifer and it is highly heterozygous. Nevertheless, in population studies of white spruce in eastern Canada inbreeding at the half-sib level (F=0.125) was found among neighboring trees in natural populations (Park et al. 1984). Genetic selection within spruce populations can be promising considering the extent of additive and non-additive variation (Namkoong 1966, Morgenstern 1996). A high degree of genetic variation within spruce populations has been observed in a number of traits (Neinstadt and Teich 1972; Fowler and Roche 1976). 2.1.2.4 Variation of Economically Important Traits It has been suggested that some economically important traits, such as volume growth, are only indirectly influenced by the evolution of the species, since they do not directly affect survival or reproduction (Langlet 1963). Similarly, wood properties have been considered influenced by differences in phenology and growth rate (Zobel and Jett 1995). Therefore, genetic variability of some of these economically important, perhaps non-adaptive, traits is reviewed separately in this section. The results are mostly regarding growth characteristics from field trials in BC. In replicated tests at Buckthorn and Aleza Lakes near Prince George, a provenance from the southern part of the Columbia forest region was 30% taller than the experimental 10 average at 4 years of age (Revel 1969). Kiss and Yeh (1988) estimated heritability of height in spruce based on a progeny test including 174 open pollinated families, on four sites in the Prince George region. Narrow sense heritability on an individual tree basis was moderate, and decreased with age from 0.52 at age 3, to 0.36 at age 6, and 0.29 at age 10. Differences in site quality and site preparation had a major influence on the early growth of spruce. Resistance to white pine weevil (Pissodes strobi Peck) was estimated at age 15, on three sites. The data suggested that there was moderate heritability for this trait, and that selection for growth might improve resistance to weevil attack (Kiss and Yanchuk 1991). In addition to height growth and weevil resistance, inheritance of wood density has been estimated on a limited number of families (Yanchuk and Kiss 1993). Observed differences among 40 families were moderate, which was reflected in an individual heritability of 0.47. Genotype by environment interaction does not seem to have a major influence on variation in growth and wood traits in the progeny tests. Families that grew best at one test site were generally above the average on the other sites (Kiss and Yehl988). Relative stability of families in terms of resistance to weevil attack was also observed (Kiss and Yanchuk 1991). 2.2 P H Y S I O L O G I C A L P R O C E S S E S A N D G E N E T I C C O N T R O L O F C O N I F E R W O O D F O R M A T I O N One way of improving growth is through provenance transfer, by moving material longitudinally, latitudinally, or altitudinally. Widely separated sites are influenced by differences in photoperiod, temperature, precipitation, and other factors that eventually lead to the evolution of different genetic types. In spruce, northern and high elevation sources usually start and finish growth early (Roche 1967, Worrall 1975, Westin et al. 1995). Provenance transfer can certainly result in changes in wood structure. For example, in Sweden, lowland sources of Norway spruce brought to the North had long continued growth and produced earlywood of reduced density, which in turn resulted in snow breakage (Kennedy 1961). To avoid such phenomena it is important for tree breeders to understand the physiological processes involved in wood formation. 11 2.2.1 Physiological Processes during Wood Formation According to the hormonal theory of cambial growth (Larson 1969, 1994), activity of terminal buds influences cambial growth and in turn wood structure. Auxin from developing leaves and shoots flows in phloem sap downward towards developing xylem. When the terminal growth ceases, the transition from early- to latewood occurs. In general the ratio of auxin to gibberellic acid controls nature of the wood produced, but other hormones such as coniferin may also be involved (Savidge 1996). If duration of cambial growth is similar for different genotypes, the ones with slow and prolonged terminal growth would be expected to produce more earlywood. Others with a rapid terminal growth would produce more latewood. This would give rise to a positive correlation between wood density and height growth rate. One could perhaps search for trees with a short and intensive terminal growth-period and prolonged radial growth, in order to increase latewood percentage and ultimately wood density. From phenological studies it appears that, in British Columbia's interior spruce, there is less variation in timing of terminal growth initiation (bud break) than in its cessation (transition to latewood) and cessation of cambial activity (Roche 1969, Krasowski et al 1993). By influencing phenology, environmental factors indirectly influence wood anatomy. Generally, warm spring and mild autumn weather would extend the growing season, and increase latewood production and its density. Sometimes, initiation of latewood early in the season results in production of so called "long day latewood", i.e. large diameter cells with thick walls. This implies the existence of a relationship between photoperiod and wood anatomy. An increase in light intensity and the level of available photosynthate caused also an increase in cell wall thickness (Savidge 1996). At high elevations, it seems that temperature governs the initiation of latewood formation more than precipitation (Rolland and Schueller 1994). In controlled experiments, for example, increased night temperatures up to 25°C resulted in increased cell diameter and length in Sitka spruce seedlings (Brown 1970). Wood density depends on the ratio of wall thickness to cell size. These two anatomical features are determined by two different physiological processes, and perhaps governed by two different gene complexes. Cell enlargement through the process of expansion of the primary cell wall is regulated by hormone production. The process of secondary wall 12 deposition and thickening is linked to the nutritional aspects, i.e. availability of photosynthates for assimilation into wall components (Larson 1969, Savage 1997). It is not known how different genotypes use the two processes to produce wood of different densities. Variation in tracheid length was found to be regulated by the rate of cambial division (Bannan 1956). Tracheid length increases with temperature, but decreases with local increase of carbohydrate levels (Brown 1970). This may be because many shorter cambial derivatives are lost i f they lack contact with rays, which are the source of carbohydrates. The relative significance of specific environmental factors and their interaction with genetic factors that influence a certain wood property are difficult to evaluate. Only when a factor becomes limiting does the correlation become more obvious. Climatic and phenological observations can be used to ascertain what limiting conditions are affecting tree growth. The numerous interactions, however, make it difficult to predict wood quality, and its relation to growth rate as determined by both genetics and environment. 2.2.2 Inheritance of Anatomical Traits Recently, a number of studies on the quantitative inheritance of anatomical traits, such as cross-sectional dimensions, fibre length, and M F A have been reported (Hannrup 1999, Shelbourne et al. 1997, Donaldson and Burdon 1996, Kibelwhite 1991). This is due mostly to the availability of the sophisticated equipment such as Kajaani® (King et al. 1998) and SilviScan® (Evans et al. 1996). The few available reports on spruce species are given in the Table 2.2.1. Although important, genetic relationships between the anatomical traits and growth, wood density, pulp and paper quality etc. are still largely unknown. Some reports exist, on phenotypic relationships, and these are reviewed in sections 2.3 and 2.4. Table 2.2.1 Reports on inheritance of wood anatomical traits in Picea sp. Species Reference Result Picea Ford 1993 Variation in wood anatomy: number of cells, cell area and sitchensis shape, and cell wall area, was found both within and among 4 examined clones Picea Boyle 1987 Tracheid diameter appeared to be under strong additive mariana genetic control 13 Table 2.2.1 continued Picea Khalil 1985 All characters related to fibre anatomy, including cell lumen mariana diameter and fibre length, had h2>0.3 except wall thickness Picea abies Olesen 1977 Tracheid diameter exhibited considerable variability 2.3 RELATIONSHIP AMONG GROWTH RATE, WOOD DENSITY AND WOOD ANATOMY It is desirable to develop genotypes with both fast growth rate and high wood quality. Some biological characteristics of spruce that may limit achieving this goal in the tree improvement program are reviewed below. Special attention is given to wood density, the trait commonly used to represent the overall wood quality. 2.3.1 Growth Rate and Wood Density in Spruce The relationship between growth rate and wood density in Picea sp. has generally been found to be negative. This was documented extensively by Zobel and van Buijetenen (1989), Zobel and Jett (1995) and Rozenberg and Cahalan (1997). Some of the results for white and Engelmann spruce are given here in the Table 2.3.1. Table 2.3.1 Examples of phenotypic and genetic correlations between diameter growth and wood density for white and Engelmann spruce Species Refernce Result P. glauca Corriveau et al. Phenotypic correlations of diameter and density for 27 1990 provenance means was -0 .45** P. glauca Corriveau et al. Genetic correlations of diameter and density for 39 families 1991 w a s - 0 . 6 3 P. glauca X Yanchuk and Kiss Phenotypic and family mean correlations of diameter and Engelmannii 1993 density were -0.46** and -0 .28 respectively ** statistics significant at 0.01 level The correlation between growth rate and density was reported to significantly change according to ring cambial age (Lewark 1982 in Rozenberg and Cahalan 1997). Kennedy 14 (1995) reported that some individual trees form denser wood early in life, in spite of accelerated growth. Ford (1993) saw differences in the production rate of wood volume and weight as an opportunity for their simultaneous improvement. A more detailed analysis on a ring by ring basis might reveal some opportunities for breaking this negative relationship through breeding and silviculture. Additional opportunities may be found after an examination at the within ring level. 2.3.2 Wood Density and Wood Anatomy Two within-ring classes earlywood (EW) and latewood (LW) are usually used to examine the relationship between growth rate and wood density for individual rings and among cells within ring. By separating the ring into classes, its total density is decomposed into the sum of densities of each class multiplied by its proportion. The criterion to separate EW from L W is usually defined as the point in the ring where density equals the mean of the minimum and maximum density values (Vargas-Hernandez and Adams 1994). Some other definitions are used for the separation criterion: a fixed value of density (Parker and Jozsa 1973) and Mork's index (Denne 1988). Larson (1969) introduced a third within-ring class: the transition wood (TW). This was necessary for connecting his ideas about physiology of cambial activity with resulting wood anatomy. Wide rings conceivably have an extra component of less dense earlywood, causing negative correlation between ring width and density (Worrall 1970, 1975, Taylor et al. 1982, Lindstrom 1997). This means that the total ring density is determined largely by the density of earlywood, i f the proportion of latewood is small, as in spruce. It was also found that changes in radial tracheid diameter with ring width are more pronounced in earlywood than in latewood. The anatomy of earlywood therefore deserves special attention. Neverthless, the accumulation of extra wood during a growing season, can be achieved either by more rapid rate of cell production in and/or by prolonging the radial growth. In the latter case, more latewood is produced. Variations in wall thickness with ring width were found to be more pronounced in latewood (Gregory and Wilson 1967, Michell and Denne 1997). Recently, the availability of extensive X-ray and anatomical imaging data make it possible to look at frequency distributions of wood density and anatomy among many within-1 5 ring classes. A n indefinite number of density classes can be represented by distribution functions, and different methods can be used for describing those distributions. 2.3.3 Opportunities for Breaking Negative Correlation between Growth Rate and Wood Density through Tree Breeding Different strategies have been proposed for avoiding the negative correlation between growth rate and wood density. They were recently reviewed by Rozenberg and Cahalan (1997). Most authors suggested selecting genetic units (clones, families, etc.) for which the correlation is less negative. Corriveau et al. (1991) found that some growth traits have less negative correlation with density, e.g., height in black spruce. Zhang and Morgensten (1995) suggested use of dry-weight instead of volume, as the trait for selection, to preserve wood density, in the same species. Rozenberg and Cahalan (1997) also suggested use of gross weight yield as selection criteria, i f pulp is the main product. However, since weight yield is a product of wood density and volume, loss in density may be offset by gain in volume. Zobel and Jett (1995) have suggested that that selection for cell diameter and wall thickness could improve both quality and yield of some types of paper. Ford (1993) also hinted at the possibility of using anatomical measures instead of density or yield for selection in Sitka spruce. 2.4 RELATIONSHIP BETWEEN TRACHEID CHARACTERISTICS AND PAPER PROPERTIES Tracheid characteristics determine to a great extent the properties of pulp fibres (separated tracheids), and therefore influence the physical properties of pulp and paper. Tracheid characteristics are important for chemical pulp, because pulp quality is largely determined by the properties of fibres. Tracheid characteristics also play an important role in determining the energy required for mechanical pulping and the quality of resulting pulp (Karnis 1994). Relationships between fibre anatomy and properties of their webs, i.e. pulp and paper, have been examined both empirically and theoretically. The complexity of factors influencing pulp and paper production, development of new technologies, and ever-changing market conditions are challenges for tree breeders attempting to establish breeding objectives for pulping processes (Raymond and Greaves 1997). 16 2.4.1 Chemical Pulps Traditionally, wood relative density has been used as an overall indicator of wood anatomy, and it has been considered the single most important wood property with major effects on both yield and quality of pulp. At higher wood density, higher chipping energy is required, more time for cooking is needed, and bulkier paper is produced. The paper drains faster and is more absorbent. On the other hand, wood with lower density generally has a higher percentage of earlywood. Earlywood fibres are flexible and collapse readily into ribbons during sheet formation and form strong interfibre bonds. They produce dense transparent sheets, with high burst and tensile, but low out-of-plane tear strength. For this reason relative density of above approximately 0.45 is negatively correlated with tensile, burst, and modulus of elasticity, and positively with tear strength of paper (Zobel and vanBuijtenen 1989). However, wood density cannot directly control pulp quality. Some other fibre properties that correlate well with wood density must be the controlling factor (Rudie 1998). Tracheid diameter (tangential and radial), length, wall thickness, and micro-fibril angle are the basic properties that influence its performance in pulps. From the basic properties a number of parameters describing tracheid geometry and strength can be derived. These parameters can be used to predict properties of either pulp wet-webs, or dry paper sheets. Generally, the larger the tracheid perimeter or its lumen diameter the easier fibres collapse and form more even sheets, with a higher bonding strength. Wall thickness is also directly related to the "collapsibility". Thick walled fibres give higher yield, but coarser, rougher, and bulkier sheets. They are harder to beat to a low freeness, have higher tear strength, but lower burst, tensile, and folding strength. The ratio of double wall thickness to lumen diameter, or Runkel-ratio, is considered as an important indicator of pulp quality. Fibre length is critical up to certain limit, because a minimum length is required for inter-fibre bonding. Longer fibres have greater contact area, tolerate greater stress, have higher tear, and to a less extent burst, tensile strength, and folding endurance. They are washed more easily, but screened with more difficulty, and give poorer sheet formation, which is a handicap for printing (Zobel and vanBuijtenen 1989). 17 Fibre flexibility also is one of the keys to softwood pulp quality. Various ratios are used to express fibre flexibility, but the ratio of fibre mass to length appears to be the best. This ratio is called coarseness, and has a strong relationship with the strength of both chemical and mechanical pulps (Seth 1996, Karnis 1994). The basic fibre properties: diameter, wall thickness, length, resulting coarseness and surface area are all correlated. For example, a larger fibre of the same coarseness as a smaller one has much greater surface area, is much more flexible, and will collapse better on drying. Since paper strength is significantly influenced by fibre surface area, a measure of fibre surface area relative to its mass is perhaps more useful measure of its performance than coarseness used as a single predictor (Rudie 1998). Increase in microfibril angle reduces the strength and elasticity of individual fibres. It thereby relates directly to the stretch of sheets, also determines sheet strength, provided there is good inter-fibre bonding, (Page et al. 1977 and 1979). The relationship of parameters related to tracheid dimensions and strength with pulp and paper properties have been examined using multiple regression (Ledell 1970, Horn 1974, Matolcsy 1975, Williams 1994, Kibellwhite et al. 1997, King et al. 1998, Corson 1999). The relationships were not completely clear, however, because of considerable interdependence among tracheid properties and changes brought about by using different techniques of pulp and papermaking. The numerous regimes of chemical and mechanical pulping make the prediction a complex problem. Theoretical and semi-empirical models, which relate fibre properties to properties of pulp and paper, have also been developed. Theoretically developed formulations are less dependent on a particular process or product type, and therefore may be desirable for use as breeding objective functions. For chemical softwood pulp and paper, certain measures of fibre characteristics were used in semi-empirical models based on the statistical geometry to predict their physical strength. For example, tensile strength of pulp wet-webs (with 30% solids) was modelled as proportional to fibre length and inversely proportional to the fibre coarseness or its square (Williams 1983, Page 1993, Seth 1995). The tensile strength of wet-webs is considered extremely important because new high-speed paper rolling machines require high tensile strength. The wet-webs have bonds between fibres, which are not determined by hydrogen 18 bonds, but by surface tension. These bonds are much weaker than modulus of individual fibres. Tensile strength of dry paper sheets depends, however, beside fibre length and coarseness on the strength individual fibres (Page and Seth 1988). The individual fibre strength is usually measured as zero-span tensile strength. It is mainly determined by the microfibril angle (Page et al. 1977). A theoretical relationship between those wood anatomy characteristics and the tensile strength of dry paper sheets has been given in explicit equations (Page 1969, Karenlampi 1995). Tensile, or in-plane strength, has been considered more important than tearing, or out-of-plane strength, for determining fracture toughness of chemical softwood pulps (Seth 1996). The relationship between the two most commonly used physical properties of paper tends to be curvilinear (MacLeod 1986). Tearing strength was examined empirically by Clark (1985) for weakly bonded papers and found to be proportional to, among other factors, zero span tensile strength and fibre length, and inversely proportional to the square root of coarseness. Page and Seth (1988) obtained different results regarding fibre coarseness. According to the authors, for well-bonded papers, tearing strength depends on fibre strength, and is directly proportional to fibre coarseness and length. Results of burst fracture and resistance tests are usually well correlated. When bursting pressure is applied, bonds are loosened in all directions, but fibre length preserves integrity of the sheet. Some other paper properties such as elasticity or folding endurance, printing and optical properties relate closely to fibre properties. 2.4.2. Mechanical Pulps Wood density has a large impact on the manufacture of mechanical pulps. It influences the mechanical energy requirement. Energy absorption is unevenly distributed within a growth increment, however. Differential energy consumption between earlywood and latewood portions explained low pulp strength and efficiency of energy use in refining of pine pulps (Rudie 1994). For juvenile wood in spruce, a species with a gradual transition to latewood and relatively few thick walled fibres, this is probably of less concern. Average fibre dimensions are more likely to control the strength of mechanical pulp rather than the energy consumption. 19 2.5 M U L T I P L E T R A I T I M P R O V E M E N T T E C H N I Q U E S W I T H S P E C I A L E M P H A S I S O N B R E E D I N G F O R W O O D P R O P E R T I E S 2.5.1 Index Selection Virtually any type of selection is effectively multiple trait selection due to the existence of correlations among traits. Selection on multiple traits can be either successive or simultaneous over generations. Baker (1986) and more recently Lynch and Walsh (1999) provided the comprehensive reviews of multiple trait selection techniques including tandem selection (selection of one trait per generation), independent-culling levels (selection where truncation levels are set for phenotypic values of each trait), and index selection. Index selection aims at simultaneous improvement of several traits giving each candidate tree an aggregate index value. This selection method, takes into account both the genotypic and phenotypic correlations among traits. A linear index (1) is a linear function of phenotypic values for traits (Xi), each of which is weighted by a coefficient (b,), such that the index value relates phenotype to the genotypic worth of that tree. The genotypic worth (H) is composed of breeding values (gi) weighted by their relative economic values (a,) per unit change: I = IbiXi H = Zdigi The vector of index weights (b) is obtained as partial regression coefficients of genotypic worth on phenotypic values: b = F'Ga where P is the matrix of phenotypic variances and covariances among traits; G is the genetic variance-covariance matrix; and a is the vector of economic weights. A special group of restricted selection indices restricts the response of some traits to either zero or some other predetermined value while maximizing others. The derivation of a restricted index involves the so-called "constrained optimization problem", which can be solved using Lagrange's multipliers. The first original restricted index, derived by Kemthorne and Nordskog (1959), maximized some traits while other traits were kept constant, i.e., their change was restricted to zero. 20 Tallis (1962) extended this approach, so that response in some traits can be maximized while others are fixed to an amount other than zero (Tallis' optimum index). A similar and widely used index called is "desired gain index" (Pesek and Baker 1969). Harville (1975) derived an index with proportionality constraints, which may be more effective than the Tallis optimum index. Gibson and Kennedy (1990) argued that for every restricted index there is a set of implied economic weights. Yamada (1995), however, argued that where the profit function is non-linear, the desired-gain index could be more efficient than the economic index. Economic weights are usually determined by changing different traits marginally by one unit, while holding all other traits constant, and then estimating the sum of change in profit. A non-linear profit function causes difficulty because the economic value of a trait is not constant, but rather changeable as the population mean changes. There are two groups of selection indices for the "non-linear situation": a group of linear selection indices and a group of non-linear ones. Godarad (1983) pointed out that if component traits are inherited additively, genetic progress is always better if based on a linear index. Economic weights for linear selection indices with non-linear profit functions can be derived by the method of Itoh and Yamada (1988), or approximated by using partial derivatives of the profit function evaluated at the means before selection. 2.5.2 Problems Specific to the Use of Selection Indices in Forestry The application of selection indices requires estimates of genetic parameters, but errors of these estimates may be high. At the same time, long generation intervals characteristic of forest trees make determination of relative economic values for the traits included in selection indices particularly difficult. In the case of uncertain economic weights, their effects can be examined through the sensitivity analysis. Extensive economic analyses and partial regression techniques have been applied, but the results may not be valid by the time progeny of selected trees are ready for harvesting (Cotterill and Jackson 1984, Talbert 1984, Borralho 1993, Aubry 1998). In this situation of high uncertainty, when the value function remains unknown and weights unpredictable, use of sensitivity analysis to find a mini-max solution, which maximizes minimum gain regardless of which value criteria applies at harvesting time, have 21 been suggested (Namkoong et al. 1988). However, this is a highly conservative option, which may result in loss in potential gain. 2.5.3 Multiple-Index Selection As already mentioned above, the complexity of factors influencing pulp and paper production, the development of new technologies, and the ever-changing market conditions often cause tree breeders to chose conservative strategies for selection. They are reluctant to clearly declare the breeding objectives of their tree improvement programs (King et al. 1988, Namkoong et al. 1988, Magnussen 1990). This problem, however, can be avoided to some extent, i f a multiple-population breeding strategy is adopted, and a different breeding objective is given to each of several populations. The multiple-index selection technique, described by Namkoong (1976), can then be employed. The breeding population can be divided into several smaller ones and within each population a different selection index can be applied. By doing so, a whole array of possible future alternatives can be explored. This strategy can be viewed as an optimization problem with multiple objective functions. Multiple objectives can be accommodated among populations. If more than one objective function is used within one of the populations then multiobjective optimization can be employed to maximize the genetic gain (Land and Mattheiss 1983, Dijkstra 1984, Miettinen and Makela 1995). The feasible set for such a program would be bounded by the selection response surface. Genotype by environment (GxE) interactions also can be significant for composite traits even in the absence of GxE in individual traits (Namkoong 1985). In that case, multiple populations may be needed for efficient breeding. As Namkoong and Johnson (1986) stated: "Selection must jointly consider the multiple nature of trait-by-environment responses, and the multiplicity of value functions" It has been suggested, however, that in early generations, more individuals should be retained in a breeding population, as most genes of future interest will be still at low frequencies (Robertson 1960). Theoretically, after a few generations of selection, the frequencies of the favoured genes are expected to increase and their chances of being lost decrease. From this reasoning comes the idea of importance of maintaining large population sizes in initial phases of tree improvement. If, on the other hand, multiple populations were 22 assigned based on co-ancestry, random allele loss may occur in any one of them, but gene frequencies should remain constant on average. This latter approach is based on Wright's shifting balance theory, and it was empirically tested by Williams et al. (1995). For smaller breeding populations, variance-covariance matrices over generations will be altered, depending on the selection intensity in each population. Reduction in effective population size could result in inbreeding, resulting in further changes in the matrices. For selection based on a certain selection index, changes in genetic variance in a breeding population from one generation to another can, to some extent, be theoretically predicted (Villanueva and Kennedy 1990). 2.6 BREEDING STRATEGIES FOR MULTIPLE TRAITS 2.6.1 Current Breeding Systems Modern breeding programs usually maintain four types of population to achieve genetic gains: 1) The base population is a large and diverse sample of the natural gene pool, managed to maintain variation for the future use. 2) The breeding population is derived from the base population and improved through successive generations of selection, mating, and more recently genetic engineering. The population can have a hierarchical structure or it can consist of multiple populations. Multiple populations can be replicate (sub-lines), diversified, or heterotic populations (Barnes 1994). In diversified populations existing differences are used, or new differences are created, by selecting for different traits including wood properties. 3) The propagation population is a sub-set of genotypes from the current breeding population, and it is used for producing commercial quantities of seed or vegetative propagules. 4) The production population is the commercial plantation of improved genotypes (Whiteman et al. 1996). In New Zealand, for example, high levels of genetic improvement of radiata pine {Pinus radiata) were achieved. The breeding population contains over a thousand selections 23 in clonal archives, but these are divided into two main sub-lines to manage inbreeding. Breeds (full-sib families or clones) have been developed for growth and form, long internodes, disease resistance, wood density, reduced spiral grain and more recently they are evaluated for fibre and pulp properties. The clones that emerge as superior are then crossed to form a variety of smaller populations designed for specific quality or site requirements. Production populations are based on controlled crosses among first- and second-generation selections, followed by vegetative propagation (Sorensson et al. 1997, Kibellwhite 1997). For coastal Douglas fir (Pseudotsuga menziesii (Douglas), in British Columbia, breeding population of 500 parents is divided into 12-parent sub-lines. Poly-cross and circular mating is used to generate progeny from which selections will be made for the next generation orchard and breeding population. Emphasis in most of the sub-lines will be on volume; only a few sub-lines emphasise other traits such as stem quality and wood density. It was estimated that by the fourth generation inbreeding will reach a level at which combining of sub-lines is necessary (Lester 1993). 2.6.1.1 Examples of Multiple Population Breeding Systems In breeding some of the southern pines, sub-lines have been used (Lowe and vanBuijtenen 1986), and the use of multiple populations, diversified for adaptability and/or different traits, has been advocated (Namkoong 1997). A version of this management system is used for loblolly pine populations, where different traits are selected in different regions. For example, fusiform rust resistance is only important in the southern part of the species range. A few selected populations are bred for special wood properties. In Zimbabwe, the breeding program is required to work with many species for many sites and the multiple population breeding (MPB) was applied in response to this need through the C A M C O R E co-operative. For a particular species, a breeding population is structured as follows: one base population that covers several regions, several main populations that consist of selections within each region, and elite populations that consist of selections within each main population. The structure allows for different selection intensities at different levels of structure (Barnes 1994, C A M C O R E 2000). For long-term breeding of Norway spruce (Picea abies) and Scots pine (Pinus sylvestris) in Sweden the MPB system is applied, aiming at developing improved stock while 24 maintaining genetic variability (Danell 1991). The spruce meta-population consists of 20 independent populations, each with 50 parent trees selected per generation. A l l selections are made within families to maintain genetic diversity. The populations are adapted to a specific combination of photoperiod and temperature. Short-term goal is development of elite lines. The M P B has been considered as an important option for white spruce in the lake states in US (Neinstaedt and Kang 1987), and in eastern Canada (Park and Adams 1993). The strategy, however, has not yet been proposed as a population management strategy within existing seed-orchard planning-zones for British Columbia's interior spruce (Lester 1993). 2.6.2 Risk Management by Using Multiple Population Breeding Strategy If outcomes of our decisions are known with certainty we have a "perfect knowledge". With enough information, so that we can estimate the probability function of outcomes, we can assess "risk". In addition, i f the probability function cannot be estimated we are dealing with "uncertainty" (Knight 1921). Anderson (1997) challenged the above concepts as naive since all probabilities are subjective, to a certain degree, because they are related to subjectively chosen decision situation. The polemic is actually a form of hair-splitting, since both risk and uncertainty similarly influence outcomes, which are our primary interest. Risk (and/or uncertainty) in forestry and agriculture is often associated with variation in yield. Generally, variance and covariance of decision alternatives are the most important measures of risk. Utility of decision alternatives is subject to the decision-maker's attitude towards risk, which determines the shape of the utility function (Bunn 1984). It is, however, often too difficult to establish a specific utility function. To deal with the problem, techniques such as Expected Value-Variance (E-V) analyses were developed (Klee 1996). Diversification is a frequently used risk management strategy that involves participating in more than one activity. The motivation for diversifying is based on the idea that when one enterprise has low returns the others are likely to have higher returns. Mutual fund managers, for example, tend to hold many stocks, thus limiting the losses when one stock is doing poorly. Theory suggests that diversification is useful when one is faced with greater risks, is relatively risk averse, and faces small reduction in returns in response to diversification. Depending on the situation, the costs of diversifying may sometimes outweigh the benefits (Harwood 1999). Considering that other methods of risk reducing, such as future 25 contracts, may not be as applicable in forestry as they are in agriculture, diversifying can be a viable approach to risk management in tree improvement. As a way of diversifying, the MPB strategy can be used to optimize genetic gains (Roberds and Namkoong 1989). Under certain realistic assumptions, this approach can be evaluated in comparison to conventional selection strategies using a single breeding population. 26 Ill MATERIALS AND METHODS 3.1 MATERIAL FROM PROGENY TESTS The interior-spruce breeding program in British Columbia includes progeny tests planted throughout the wide range of spruce habitats. Three general selection units, East Kootenays, Prince George, and Smithers are represented by 132, 178, and 168 plus-trees respectively. Open-pollinated progeny trials were planted on 40 sites. The progeny trials were established between 1973 and 1975, using 2+1 seedlings, at 2.5m spacing (Kiss 1976). For the present project, sampling included 80 open pollinated families, whose parents were originally from the Prince George (PG) selection unit, in two plantations at the Red Rock (RR) Tree Improvement Station near Prince George and Quesnel (Q). Each family was replicated in 6 blocks, within each site. The sampling also included another 80 families originally from the East Kootenay (EK) selection unit, planted on 3 test sites: Red Rock tree improvement station, with 6 blocks per site, Invermere - Jumbo Creek (JC), and Cranbrook -Perry Creek (PC), both with 3 blocks per site. (Note that the E K progenies are represented not only within their own selection unit, but also at RR site in the PG region.) Location of origin of selected parent trees is given in the APPENDIX 1. Identification and description of test sites are given in the Table 3.1.1, and their location shown in Figure 3.1.1. One tree from each block was sampled. Altogether more than 1920 large size (10mm) increment cores were taken. 27 Table 3.1.1 Identification and description of test sites used in this study Breeding unit Test Site Rep's Latitude (deg°) Longitude (deg°) Elevation ( m ) Soil EK RR 6 53.75 122.73 762 Alluvial gravel loam J C 3 50.37 116.48 1372 PC 3 49.55 116.03 1463 P G RR 6 53.46 122.43 610 Alluvial gravel loam Q 6 52.59 122.14 915 Clay silt-loam Source: BC Ministry of Forests, Kalamalka research station . Figure 3.1.1 Location of test sites of the Interior-Spruce tree improvement program in the Prince George and East Kootenay regions. 28 3.2 D A T A SETS A N D TRAITS STUDIED 3.2.1 Traits Related to Growth and Wood Macro-Properties For the analyses of growth and wood macro-properties, a data set obtained from the B.C. Ministry of Forests, containing periodical height and diameter data, was combined with the data from additional field measurements of height, diameter, and laboratory assessments of wood properties. Increment cores were used for laboratory analyses. Ring based observations on increment cores, such as measurements of ring width and latewood percentage, were obtained by photometry, a method which utilises a reflective-light scanning instrument, and an image analysis system. For determining wood density of individual rings and within-ring density profiles, a choice had to be made among three different methods: the reflective-light based photometry, direct reading X-ray densitometry, and microscopic morphometry. A small comparative study was conducted, based on a data set obtained by applying those three methods to the same wood samples (Ivkovich and Koshy 1997)(see A P P P E N D I X 2). 3.2.1.1 Height Growth B.C. Ministry of Forests provided data for tree height in the progeny tests for ages 1, 3, 5, 7, 10, and 15 years. I made the final height measurements (H) in the spring of 1996, at age 20 for E K and 22 years after planting for PG progeny. 3.2.1.2 Radial Growth Basal area might have been a more meaningful variable to use as a measure of radial growth than diameter or ring width, because it relates closer to the volume of wood accumulated around a tree stem. However, basal area was not normally distributed, and it had a non-linear relationship with wood density, so it did not lend itself to straightforward conclusions from statistical analyses. Therefore radial growth on sample cores taken above the sixth whorl of branches, was examined as cumulative and individual ring width (RW). 29 Diameter at breast height Diameter at the breast height was obtained from the Ministry of Forests for both sets of progeny at ages 1, 3, 5, 7, 10, 15, and 20. I measured it at the start of 1996 growing season, at age 22 after planting, for the PG progenies (D). Ring width Width of annual increments (RW) in each year was obtained by photometry of reflected light. A polished surface of wood on each core was scanned using a reflective-light scanning instrument. Scanned images were analysed using the WinDendro® system specially designed for growth ring analysis (Regent Instruments 1999). Usually only the 9 to 11 outermost rings on increment cores were free from the compression wood, and these were included in the analyses. 3.2.1.3 Wood Macro-Properties Evaluation of macro-scale wood properties included two traits: average latewood percentage (AVLW%), and average relative density (AVRD). Latewood Percentage Latewood percentage (LW%) was measured using photometry. The point of transition from earlywood to latewood was defined to be at the pixel with 50% difference between the minimum and maximum light intensity within a ring. There was no apparent need for any smoothing of light intensity profiles. Normally, to determine the relative amount of latewood within each ring, a relative threshold is applied to within-ring wood density profiles obtained by X-ray scanning. In a pilot study, we obtained data by both X-ray and reflective-light scanning. Although a universal calibration of reflected light intensity that would correspond to X-ray transmission seemed impossible, within single rings, correlation of the reflected light intensity and wood density determined by X-ray measurement was typically high. Consequently this should enable one to determine the relative amount of latewood within each ring based on the reflected-light intensity profile. These relative amounts can indeed be compared among different trees (for details see APPPENDIX 2). From a practical point of view, since L W % is easy to obtain, relates well to wood density, and can be used to relate to other wood properties, it is a very useful variable to be used in the evaluation of progeny tests. 30 Relative density The average relative density (AVRD) of small wood pieces was determined before within-ring analyses. A wood piece consisted of the first 4 outermost rings cut from the cores sampled from the E K tests, and the first 6 outermost rings in the PG tests. The sampled rings corresponded to cambial ages from 12 to 15 years, and from 12 to 17 years, respectively. Since the increment cores were taken at a position determined relative to the total tree height, they almost always had the same ring count. In that way, variability between samples due to both cambial age and calendar year was avoided. For determining wood relative density of small pieces, the "Maximum Moisture Content" (MMC) method of Smith (1954) was used. This method is also known as gravimetric method because it only requires measuring of weight, without knowing the volume of a particular piece of wood. In a preliminary study, the M M C method produced almost identical results as the volumetric method, and showed good agreement with the results obtained by X-ray measurements (r2= 0.93). After measuring ring width, the small pieces were cut from the rest of the increment core. In order to determine the maximum weight (Wmax), these pieces were put into distilled water, and placed in a desiccator. To ensure that the wood has reached the point of maximum moisture content, they were left submerged until the last three measurements of weight were the same. Usually, it takes 2 to 3 weeks to reach the maximum moisture content after complete submersion of the specimen. A n analytical scale was used for weight measurements. Just before weighing each piece, excess water was blotted with a wet paper towel to remove surface water but not water within the wood, as suggested by Gonzales and Kellogg (1987). After a constant maximum weight (Wmax) is recorded, the ring sections are dried in the oven for 48 hours at 105 ± 3°C to obtain the oven-dry weight (W0). The ratio of maximum to oven-dry weight is calculated to two decimal places, and the appropriate value of relative density can be found in Fogg's (1967) table, which is based on the following formula: RD = ! ETmax ~ Wo 1 Wo  + 1 . 5 3 RD = basic relative density, which is determined on a green volume basis (W0 /Vg), Wmax = maximum weight (weight of wood with maximum moisture content), W„ = oven dry weight, 1.53 = relative density of wood substance. 31 3.2.2 Tracheid Characteristics Several options for within ring evaluation of wood properties at a micro-scale were considered. Within-ring density profiles were first obtained for the purpose of comparison by using two different techniques: photometry (described above) and morphometry (described in the following sub-section). Data obtained by X-ray densitometry (Parker and Jozsa 1973) were used for reference. Results of the pilot study showed that the photometry, i.e., measurements of reflected light from the polished surface of spruce wood was not an adequate substitution for commonly used X-ray transmission measurements. For the purpose of density comparison among trees, a universal calibration of reflected light intensity corresponding to X-ray transmission was needed, but it seemed impossible. The regression coefficients and intercepts obtained were characteristically inconsistent. On the other hand, morphometric analysis is suitable not only for approximation of wood density profiles from double cell wall to lumen ratios, but also for examination of other features of cell morphology. Results indicate that the morphometric measurements have several advantages over X-ray measurements. The important one is that most X-ray density measurement systems introduce a slight smoothing in the density due to internal reflections and limitations of detector sensitivity. The resolution of the morphometric measurements is higher and that type of smoothing is minimized. Nevertheless mean ring density, estimated by X-ray and morphometric measurements had strong correlation (r2 = 0.82). 3.2.2.1 Traits Based on Tracheid Cross-sectional Dimensions We have developed a technique for quantitative assessment of tracheid characteristics by measuring cross-sectional dimensions. (Ivkovich and Koshy 1997 appended as APPENDIX2) Samples, 15 um thick, including 4 outermost rings (EK) or 6 outermost rings (PG samples), were prepared using a sliding microtome. Sections were stained with aniline safranin, and mounted with Cytoseal™. Monochrome images were captured with a video camera and analysed by the SigmaScanPro® image processing software (Jandel Corp. 1995). Frame size of an image was 640x480 pixels. The average pixel intensity was recorded along several 11 pixels (approximately 7 urn) wide line across the annual rings. The lines were drawn in the radial direction that was perpendicular to cell walls. A threshold value was 32 chosen to separate the low and high intensity pixels along the lines. An algorithm was designed such that the cell lumen was assigned value of zero, and the cell wall value of one. Each column of zeros and ones was analysed, to yield values for cell wall thickness and lumen radial diameters for each cell, and their ratios. A balanced threshold was found for the width of walls in the first earlywood and the last latewood such that no cell walls were skipped in the first earlywood and no cell lumens were lost in the last latewood (Figure 3.2.1). Only the complete files are considered to avoid tapering close to cell tips. The values represent averages for sampling transects taken across several cell files. More than fifty variables were derived from the morphometric data, but only those deemed the most important are described below and summarised in the Table 3.2.1. The first three definitions were proposed by Larson (1969). Earlywood percentage Earlywood percentage (EW%) was defined as region with cells along sampling transects that have both lumen bigger and adjacent double cell wall thinner than average. Transition-wood percentage Transition-wood percentage (TW%) comprised of cells along sampling transects that had either lumen smaller or adjacent double cell wall thicker than average, but not both at the same time. Latewood percentage Latewood percentage (LW%) comprised of cells that had both smaller lumen and thicker double cell wall than the average. Mork's index (modified) Mork's (MI) index was defined as the percentage of cells along a sampling line that have double wall thickness to lumen width ratio greater than one quarter. Originally the Mork's index was defined as the percentage of cells with the ratio greater than one half. However, this definition was designed for mature spruce wood, and in juvenile wood the ratio of one half might not be present (Denne 1988). Number of cells per ring Average number of cells (CN) along the sampling transects in the radial direction was represented by this variable. 33 Figure 3.2.1 Last late-wood and first early-wood of two growth rings. The superimposed lines indicate sampling transects across cell files along which data on wood anatomy were recorded. Table 3.2.1 Variables related to fibre characteristics and their abbreviations. V a r i a b l e W h o l e ring E a r l y - w o o d (EW) T rans i t i on -wood (TW) La te -wood (LW) E a r l y - w o o d percentage E W % Trans i t ion -wood percentage T W % L a t e - w o o d percentage L W % Mork ' s Index M l N u m b e r of ce l l s C N R a d i a l cel l s i z e C S r Tangent ia l ce l l s i z e est Rad ia l lumen s i z e L L e w Ltw L lw Doub le wal l t h i c k n e s s D W D W e w D W t w DWIw Ratio: doub le wal l t h i ckness /ce l l s i z e R R e w R t w R l w Fibre length F L Microfibri l ang le M F A Ce l l c r o s s - s e c t i o n a l a r e a C S A Ce l l per imeter P C o a r s e n e s s C 34 Mean radial cell size Size of each individual cell was taken to be the sum of a radial lumen diameter and the adjacent radial double wall. Mean radial cell size (CSr) represents the mean of all cells that were sampled along several transects. Mean lumen radial sizes of cells that were classified according to Larson's definition as early-wood (Lew), transition-wood (Ltw), and late-wood (Llw), were also obtained. Mean tangential cell size This variable (CSt) represented the mean cell size in tangential direction, i.e. the tangential dimension of lumens plus a double cell wall thickness. Mean double wall thickness Double wall (DW) is comprised of two adjacent radial cell walls. It represents the mean for all double cell walls measured. Mean double wall thickness was also estimated separately in early-wood (DWew), transition-wood (DWtw) and in late-wood (DWlw). Mean ratio DW/CS The ratio DW/CS was chosen instead of the ratio of double wall to cell lumen, because the distribution of this variable was closer to normal. The ratio was estimated for the whole ring (R), and separately in early-wood (Rew), transition-wood (Rtw), and late-wood (Rlw). 3.2.2.2 Fibre Length and Micro-fibril Angle Fibre length Fibre length (FL) was evaluated for the central portion of the same growth rings that were included in the morphometric measurements. The fibres were macerated by Franklins's (1945) method. The images of fibres from each sampled ring were digitally measured using the WinCell© package (Regent Instruments 1999). Micro-fibril angle (MFA) For measuring micro-fibril angle (MFA) a measuring technique using angles of cracks in the cell wall, often extending from the slit-like pit apertures of ray parenchyma, was applied to macerated fibres. The cracks were induced by rapidly drying and re-wetting sample pieces before maceration. The technique was first proposed by Cockrell (1974) for wood of giant sequoia (Seqoiadendron giganteum), and also used by Hirakava and Fujisawa (1995) for wood of sugi tree (Cryptomeria japonicd). It was assumed that the orientation of slit-like pit 35 apparatuses relative to the local fibre axis approximately follows the M F A . The orientation of slit-like pit apparatuses was also used by Kyrkjeedeby (1990), for measuring M F A in Norway spruce. Data obtained by this method showed good correlation (r>0.80) with the data obtained by the standard method of Senft and Bendtsen (1985). 3.2.2.3 Other Derived Traits Cross-sectional area With an assumption that all cells have a uniform rectangular shape, cross sectional area (CSA) of cell solid fraction was calculated as follows: CSA = CSr*CSt-(CSr-DW)*(CSt-DW) Cell perimeter Cell perimeter (P) was calculated as follows: P = (CSr+CSt)*2 Coarseness Coarseness (C) is defined as the weight of fibres per unit length. Since weight of a fibre is the product of its cross-sectional wall area, its length and the specific gravity of cell wall material, croarseness can be simply represented as fibre cross-sectional wall area (CSA) (Kibbelwhite and Shelbourne 1997). _ ((CSr* CSt)- ((CSr- DW)* (CSt- DW)))* F L * 1.56 FL A l l of the above variables have shown good repeatability (ratio of among-sample to total variation >0.95). It was mentioned earlier that mean ring density for 30 annual rings obtained by X-ray measurements had high correlation with morphometric measurements of ratio R (r = 0.82). In spite of the fact that variation in cell shape has not been considered in calculation of R, its within ring variation followed closely also X-ray density profiles. The correlation between the two types of profiles was typically high, with an average r2 = 0.89 and ranging from 0.77 to 0.92 (APPENDIX 2 ) . 36 3.3 DATA ANALYSES 3.3.1 Preliminary Study: Experimental Design, Sample Size, and Power of Statistical Tests 3.3.1.1 Experimental Design Sampling of the spruce progeny tests was primarily dependent on the available resources. Collecting and processing of 2000 increment cores was deemed feasible. Within this limit, tradeoffs had to be made between the number of families to be evaluated and the number of individuals per family. Optimal numbers were found by using Robertson's (1959) formula. Parameters used in the formula were based on the range of variation previously observed in the same tests (Kiss and Yeh 1988, Yanchuk and Kiss 1993). Allocation of sampling among sites and blocks within sites was done according to recommendations of Loo-Dinkins (1992), based on the size of detectable difference between families in standard deviations. It was concluded that a minimum variance of estimated heritability and genetic correlations would be obtained from 80 open-pollinated families with 12 individuals per family within each of two regions, Prince George and East Kootenays. Progenies representing parent trees from the East Kootenay region were sampled on two test sites in the East Kootenays, Jumbo Creek (JC) and Perry Creek (PC), and in the Prince George region at the Red Rock (RR) site. Sites JC and PC had three blocks each, while the RR site had six blocks. Prince George progenies were sampled at the Red Rock (RR) and Quesnel (Q) test sites, with six blocks per site. The general linear model and analysis of variance (ANOVA) format (Table 3.3.1) are given below: Yyki = M + Si + R(i)j + 8(i)j + Fk + SFik + RF(i)jk + emi z=l ,2 , (3*);y=l ,2 . . .6(3*); k = 1, 2 , 8 0 ; /=1. where: Yyki value of a variable observed on the Ith tree from the &th family in/h replication (block) within z'th site, ju overall mean, Sj random effect of i t h site, R(i)j random effect offh replication (block) within z'th site, 37 S (yj (random) effect of randomisation of replication plots within the./"1 site (the restriction error of Anderson and McLean 1974), Fk random effect of the kth family, SFlk interaction effect of the /'* site with the kth family, RF,(j)k interaction effect of the/h replication (block) with the kth family, e(ijkji random effect of the single tree from the klb family from the/ h, replication within /' site. Assumed to be identically and independently distributed with N(0,s2). * In the East Kootenay progenies there were 3 sites with 3 and one site with 6 replicates per site. Table 3.3.1 General analysis of variance format and expected mean squares for the interior spruce half-sib progeny tests. (The design is similar for Prince George and East Kooteney progenies, but the analysis does not include comparison between two regions because there is no true replication of both on a common site.) Source of variation Degrees of freedom EMS S R(S) *1(2) 10(9) cr2 + 6C?SF + S0C?R + 240o * s cr2 + 80cfR ** F 79(79) cr2 + 6CT 2 SF + <^F SF 79(158) cr2 + 6 C T 2 SF RF 790(711) + (o2RF - assumed to be 0) Within 0(0) a 2 Total 959(959) * Degrees of freedom for the East Kootenay tests are given in the brackets; ** The broken line divides two sets of factors, which cannot be tested mutually because of the restriction error. 3.3.1.2 Sample Size and Power of Statistical Tests For estimating the variance components and determining the significance of family variance relative to the error variance, the power of tests equals: F, KK* > 1 + ( 1 f 2 ^ 2 / 2 CT oF / T + I / \ r ) J } or for site by family interaction as: l + r 38 where: <jF2 family variance component, OSF site by family variance component, a2 error variance component, s number of sites, r number of replicates per site. Since families are considered a random effect N(0, o~/r) the calculations involve the central F-distribution with a significance level a and vy, v2 degrees of freedom associated with family and error mean squares, respectively. Thus the power for detection of significant family differences depends on the actual variance ratio, the number of trees per family, and the chosen level of significance (Wearden 1959, Lynch and Walsh 1997). For preliminary determination of the power of detecting the best families, the method of Cotterill and James (1984) was used. This method is based on the standardised difference between the true means, number of trees per family, and a certain level of heritability. The above procedures for sample size and power calculations were done for growth (H, D, RW) and wood properties (LW%, AVRD) . To determine necessary sub-sample size for the traits related to wood anatomy, i.e., the number of sample transects per core, or fibres per ring, calculations were based on preliminary results. Within a particular site, the power for detecting family differences equals: l - a , V | , v 2 1 + r r } ss where: sub-sample variance component, sub-sample size, i.e. number of sample transects within a ring or fibres per ring. A sub-sample size of four sample transects within each ring was found to be sufficient for most of the traits related to wood anatomy. For calculating the number of microfibril angles per fibre per ring or the number of sampling transects per cross-section per ring, the above formula can be extended to: F, l - a , v , l + r\ } 39 where: cr,.,.,2 sub-sub-sample variance component, q sub-sub-sample size, e.g., number of micro-fibril angles measured on each fibre within each ring. Power and sub-sample size for detecting significant differences in fixed family effects for the traits related to wood characteristics can be determined in a similar fashion, using the tests involving non-central F distribution. Repeatability of Measurements Two different samples were taken from opposite sides of each of 30 increment cores, and the ratio of the variance among samples to the total variance was calculated as an indicator of repeatability of measurements. Some of the traits related to wood characteristics did not show high enough repeatability when preliminary results were analysed. Those were mostly the variables related to measurements of maximum and minimum of the above described morphological traits. Variables retained in the subsequent analyses all had repeatability greater than 0.95. The question still remains about the relationship between a sampling point fixed (approximately at the breast height) and the whole tree value, which was considered to be relatively good for some traits including wood density (Zobel and vanBuijtenen 1989, Loo-Dinkins and Gonzales 1991). Good correspondence between brest height and whole tree 2 2 values was found, ranging from r =0.83 for tracheid wall thickness to r =0.91 for tracheid perimeter in radiata pine {Pinus radiata) (Evans et al. 1996). 3.3.2 Estimation of Genetic Parameters 3.3.2.1. Assumption Checking and Necessary Transformations Assumptions Normality of the distribution of family effect in blocks within sites was checked using the Interactive Data Exploration option in SAS/ASSIST® procedure. For a test of normality, the hypothesized distribution is a normal distribution function with parameters population mean (ju) and variance (o2) estimated by the sample mean and standard deviation. The probability of a larger test statistic is obtained by forming the modified Kolmogorov's statistic 40 D. Normality of distribution of residuals was tested by both Kolmogorov D and Shapiro-Wilk's statistic W. Homogeneity of family -variances across blocks and sites was checked graphically by plots of "studentised" residuals as function of predicted values. The traits were analysed after some outlying data resulting from measurement errors were removed. Transformations To satisfy the assumption that the errors have a normal distribution with constant variance, a logit transformation was used to transform the percentage or ratio data (p) (EW%, TW%, L W % , MI, or R): l0g i t (» = 111(0/(100-/7)) The above variables were measured on an approximately continuous scale, and in the case where the percentage is not too extreme {i.e., proportion close to zero or one) the logit transformation can give a distribution close to normal. The arcsine(square-root(x)) transformation was used as a remedy for heteroscedasticity for the same traits. Since there are difficulties with interpretation of results obtained from transformed data, and there are usually no significant differences in genetic gain calculated from transformed and raw data (Jansson and Danell 1993), transformations were omitted from the calculations of genetic gain. 3.3.2.2 Estimation of Variance Components and Heritability Variance components were estimated using the mixed model analyses of variance using procedure M I X E D of SAS/STAT® software (SAS Institute Inc. 1997), which utilises the restricted maximum likelihood algorithm (REML) (Patterson and Thompson 1971). The variance components in a mixed model are estimated by maximum likelihood, using only the residual or error contrasts of the data set, /. e., only those data contrasts with zero expectation containing no information on treatment differences. For mixed models without fixed effects, this is equivalent to maximum likelihood (ML) estimation of variance components. When fixed effects are present in the model, the method adjusts the estimates for the degrees of freedom used in estimating treatment effects. For balanced data, R E M L estimates of variance components are the same as those obtained from expected mean squares in A N O V A . 41 If the assumptions of the likelihood estimation that the random effects are independent and arise from a normal distribution with equal variance are satisfied, a likelihood ratio testing procedure can be used to establish the best linear model. To test the null hypothesis that one of the variance components equals zero {a2 = 0), the linear model can be fitted both, with or without the variance component. The change in deviance (-21og of the residual likelihood evaluated at the estimated parameter values) between the two models is used as a test statistic. For large samples, under the null hypothesis cr2 = 0, the change in deviance has a %2 distribution with one degree of freedom. If more than one component is omitted from the model at once, the degrees of freedom for the test are equal to the number of components omitted. Estimates of variance components, however, are not independent; the change in deviation depends on the order in which the terms are omitted from the model (Horgan and Hunter 1993). Individual heritability (/z,) and family heritability (h/) were estimated as follows: h2= c s l ' 2 , 2 , 2 o> +<ySF +<Jms) H2= ?JL 1 2 rr 2 c x2+ ^ + °jn^ s sr where: c = 4 (for half-sib families), s = number of sites, r = number of replicates (blocks). The constant c equals 4 in case of random mating, but this is an approximation that may be very well overestimated due to the lack of random mating (Namkoong 1966). 3.3.2.3 Estimation of Covariance Components and Genetic Correlations Since PROC M I X E D does not perform the multivariate R E M L , covariance components and genetic correlations were calculated using the program package Quercus, which facilitates the multiple-trait R E M L algorithm for quantitative genetic data (Shaw 1987). The programs perform maximum likelihood analysis of data from a population, using a model involving different types of variance components. They estimate genetic and environmental variance 42 and covariance, as well as fixed effects using the Bio-model of Cockerham and Weir (1977). The M L or R E M L estimate of variance or covariance components can be constrained to satisfy feasibility requirements or for hypothesis testing. Significance of a covariance component was tested using likelihood deviance in the similar fashion as for a variance component. Genetic correlations (ra) between traits x and y were calculated using additive genetic variance and covariance as: Since the estimation error for genetic correlations is inversely proportional to heritability of the examined traits (section 3.3.2.4), the estimates were the best for the sites where the heritabilities were the highest. 3.3.2.4 Variance and Distribution of Estimated Genetic Parameters Falconer (1989) gave the approximate formula for standard errors of heritability and genetic correlation: The formula, however, can give too narrow a confidence interval (Roff and Preziosi 1994, Koots and Gibson 1996). To approximate standard errors of heritabilities and genetic correlations "delta method" has been recommended (Lynch and Walsh 1997). Expectation and moments of a non-linear function of several variables can be approximated, by using Taylor expansion and ignoring all but the first-order terms, e.g. variance is: The technique uses the asymptotic variance-covariance matrix for estimates obtained as the second derivative of the likelihood (objective) function (SAS Institute Inc. 1997). ra- Oaxyl (Oax Oay) 43 The results obtained by the above technique were in good agreement with those obtained by using jackknife technique (Roff and Preziosi 1994). Using the SAS/IML softwere and SAS macro processing (SAS Institute Inc. 1988 and 1987), a jackknife estimate of e.g. genetic correlation r,- and other parameters can be obtained as described. First the genetic correlation is estimated as described above, then a sequence of pseudo-values is computed by dropping in turn each of the families and using the formula 0N,i = NrN-(N-\)rN.u where: &NJ = i'h pseudo-value, N = total number of families, rN = correlation estimated using all TV values, rN_u = correlation estimated by dropping i'h family alone. There are N pseudo values and the jackknife estimator (rj) is simply the mean of the pseudo-values i=N I Variance of the estimated genetic correlation is C T r j = ' f J ( 0 N , l - r J ) 2 / N ( N - \ ) Provided that the true value of genetic correlation is not close to ±1, its sampling distribution is approximately normal (vanVleck and Henderson 1961, Brown 1969, Hui Liu et ai 1997). A jackknife estimate of heritability and its variance can be obtained in a similar fashion. For narrow sense heritability, a regression model can be used to predict quantiles and confidence limits of, either directly or via a generalized beta distribution model. Heritabilities follow a beta distribution under normally distributed environments and additive genetic effects. Estimates of the expected heritability and Taylor approximations (delta method) of standard errors are needed as input to the quantile model (Magnussen 1992). 44 3.3.3 Stability of Family Performance 3.3.3.1 Genotype by Site Interaction The genotype by environment interaction effect (GxE) was first evaluated by plotting family means at one site against means of the same families at another site. The GxE was subsequently evaluated through statistical significance of the FxS variance component. In addition the type-B genetic correlations (rE.s1.s2) of Burdon (1977) were calculated. rBsi.s2= OaSLS2l (aaSi*aasl) where: 0a.s1.s2 = additive covariance between a trait expressed at sites SI and S2 respectively, (7as, and cxaSi.s2 = additive variances at sites SI and S2 respectively. These correlations can be used for predicting the correlated response (AGsi.52) to selection among progenies, on site SI for use on site S2: AGSLS2= i hs,hS2 rBsi.s2 &Ps2 where: /' = selection intensity, 2 Ops2 = phenotypic variance at site S2, hx = heritability at site SI. Multiple traits are usually combined into selection indices, and even when there is no GxE on a trait-by-trait basis it can be present for an index (Namkoong 1985). In such case, performance of overall, averaged, and specific variance-covariance matrices is compared as a basis for index derivation (Lynch and Walsh 1999). A method based on maximum-likelihood, which is incorporated in the QUERCUS software, was used for testing the similarity of genetic variance-covariance matrices (G) (Shaw 1991). Finally, it was possible to generate three-way interactions from our data set by including the growing seasons as an additional factor and this problem will be addressed in the following section. 3.3.3.2 Age-Age Correlations Analysis of measures on each experimental unit over years provides the information on time trends (Kuehl 1994). Differences in time trends for RW or L W % among individual trees and families were evaluated. The main experimental effects S, F, and SxF were in a factorial 45 relationship with the effects calendar year or cambial age (Y). Within tree errors (e) were not independent, however, because RW or L W % of the adjacent rings tend to be more correlated than in rings several years apart. Formation of cambial initials always in the previous growing season provides a simple explanation for this correlation (Fritts 1976). The covariance structure of errors can be modeled by using statement REPEATED in procedure MIXED of SAS/STAT®, which provide different structures for within subject variance-covariance matrices. The most appropriate one, with the property of correlation being larger for nearby rings than for those far apart, is auto-regressive of order 1 (AR1). This AR1 correction is important for the inferences about the main experimental effects. Alternatively, due to the large computer memory required to perform the above procedure, statement REPEATED in procedure G L M was used for the analysis. This is equivalent to using the unstructured covariance for multivariate tests of main effects, or compound symmetry for Greenhouse-Geisser or Huynh-Feldt adjusted univariate F tests of time (within subject) effects (Littell et al. 1996). Evolution of genetic and phenotypic variance, and associated heritability was followed over years. Phenotypic and genetic age-age correlations were calculated and plotted for different lags between measurements. The role of GxE for these correlations was also evaluated by calculating site-specific parameters. 3.4 R E L A T I O N S H I P B E T W E E N G R O W T H R A T E A N D W O O D D E N S I T Y A N D A N A T O M Y The relationship between traits related to tree growth, wood density and wood anatomy were examined more closely using the methods described above. Relationships of height, radial, and volume growth rate with overall wood density for different genotypes in different environments were first studied. Basal area and volume growth had non-linear relationships with wood density. Therefore, phenotypic and genetic correlations with A V R D were calculated for A V H and A V R W , with an approximately linear relationship. The relationship between growth and wood density on a ring-by-ring basis at different cambial ages was assessed indirectly either by considering L W % or anatomical parameter R. 46 Both showed strong positive correlation with average RD, at the phenotypic as well as at the genetic level. To approximate within ring density profile, R was used. Influence of growth rate on within-ring density was examined by looking at its within-ring distribution. Parameters (moments) that describe the distribution, i.e., mean (ju), standard deviation (cr), skewness (co), and kurtosis (K) were first summarized, across sites. The distributions for rings with low and high growth rates were graphed and visually compared. To statistically test the effect of growth rate on the within ring density distribution, correlation between RW and statistics describing within ring frequency distribution for the anatomical traits, DW, CS, and R were calculated. After within-ring distribution parameters were adjusted for growth rate as a covariate in an analysis of covariance (ANCOVA), their random variation was partitioned between families and within families. The random regression coefficient model was applied using PROC M I X E D of SAS/STAT® (Littell et al. 1996). In such a model, regression coefficients for each family are assumed to be a random deviation from some population coefficients. The analysis was used to assess the relative contribution of family differences to the overall variation in shape of within-ring frequency distributions of DW, CS, and R. Similar analyses were performed on DW and CS, using transformed R as a covariate. Relationships between growth rate and anatomical traits that constitute density components were also studied. The use of component traits instead of wood density as a composite trait for avoiding the existing negative genetic correlation between growth and wood density was examined. Additional statistical techniques described in the following sections were employed for that purpose. 3.4.1 MAXR Analysis of Anatomical Data Whole ring density, or in our case R, can be partitioned into component traits (modified from Vargas-Hernandes and Adams 1991): R = E W % * (DWew/CSew) + TW% * (DWtw/CStw) + L W % * (DWlw/CSlw) Regression analyses were obtained using mean ring R and those various tracheid properties. Data for R and L W percentage were transformed so they were approximately 47 normally distributed. The multiple regression technique M A X R was used (SAS/STAT®, SAS Institute Inc. 1997). It uses a form of stepwise selection to fit the best (maximum R 2 ) one-variable model, the best two-variable model, and so on. The stepwise selection is similar to forward selection except the variables do not necessarily stay in the model. The analyses were done as an exploratory model building exercise rather than with the objective of predicting ring density from anatomical traits. 3.4.2 Principal Component Analysis of Anatomical Data Another descriptive statistical tool, principal component (PC) analysis was applied to family means based on the correlation matrix for anatomical data. The purpose of principal component analysis is to derive a small number of linear combinations (PCs) of a set of variables that retain as much of the information in the original variables as possible. Principal component analysis can also be viewed as an attempt to uncover approximate linear dependencies among variables. Only those principal components with eigenvalues greater than one were retained. SAS/STAT® software (SAS Institute Inc. 1997) was used to perform of this task. 3.4.3 Selection Based on Density Components The possibility of circumventing the negative genetic correlation between growth and wood density, as measured by RW and R, was examined by looking at the correlated response (CR) in those traits, when selecting for component traits instead of directly on R. The multivariate response was calculated from the multivariate breeder's equation: CR = GF's where: P = p h e n o t y p i c m a t r i x o f var iances a n d c o v a r i a n c e s ; G = genetic v a r i a n c e - c o v a r i a n c e m a t r i x ; s = vector o f se lect ion intensit ies. Only component traits with statistically significant genetic variance at that site were included in the calculations. First, direct response to selection was calculated for RW and R. Selection intensity was then placed in turn on each trait that had a positive genetic correlation with R, and correlated responses in RW and R were compared "with direct response to 48 selection. Similarly, traits positively correlated with RW were used to replace it in the vector of selection intensities. Only one trait at a time was entered, considering the decrease in accuracy of prediction i f a larger number of traits are included in calculation. t 3.4.3.1 Monte Carlo Simulations Confidence limits on correlated response in the examined traits were obtained through Monte Carlo simulations. This computer intensive analysis accounts for every possible value that each variable could take and weighs each possible combination by the probability of its occurrence. Precision in determining the outcome distribution and the behaviour of the model is dependent on the number of iterations (Vose 1996). The estimated distribution of response was based on an assumption of normal distribution of estimates of genetic parameters. Initially the properties of their dispersion were obtained by both jackknife and delta techniques as explained previously. A package of add-in software @RISK® 3.5.2 for Excel (Palisade Corp. 1997) was used for the simulations. 3.5 OPTIMIZATION O F S E L E C T I O N For selection to be maximally efficient, relative economic weights on different traits according to particular objective functions are needed. Optimal index weights for selection indices with non-linear profit functions can be derived using the method of Itoh and Yamada (1988) described below. If more than one objective function is used, then multiobjective optimization can be employed to maximize genetic gain (Land and Mattheiss 1983, Dijkstra 1984, Miettinen and Makela 1995). When uncertainty about the value functions exists, multiple-index selection technique, described by Namkoong (1976), can be employed as a risk reduction strategy. 3.5.1 Definitions of Objective Functions Growth, wood, and fibre characteristics were combined to form following value functions: 49 Volume of wood (VOL) Standing volume per stem for the immature Interior spruce is given by the local volume equations (B.C. Ministry of Forests 1976): /0g/OVOL=-4.29+1.86 log10D + 1.01 log10ii Dry-weight of wood (DWT) Average dry-weight on a per stem basis was determined as the product of stem volume and the basic density of the wood: DWT = V O L * R D Tensile strength of pulp wet-webs ( T w ) According to the quantitative theory of the strength of wet-webs (Page 1993), for sheets of straight fibres, bonded so weakly that the fibre strength exceeds the bond strength, tensile strength is: _ b? L RBA where: T w w = tensile strength (moment) (Nm/g) b = shear strength of the fibre-fibre bond (load) (constant assigned 1.8* 10"6) (N/m2) P = fibre perimeter (m) of the average fibre cross-section L = fibre length (m) C = fibre coarseness (g/m) RBA = relative bonded area in the sheet (constant assigned 0.5). Tensile strength of paper (Tp) The following explicit equation for the tensile strength of paper in terms of a few basic fibre properties was derived by Page (1969) and empirically tested by Page and Seth (1988), andO'Neil etal. (1999): J _ _ _ 9 _ 12CSA p T p ~ 8Z + b? L RBA where: T p = finite-span tensile strength (moment) of paper (Nm/g), Z = zero-span tensile strength (a measure of fibre strength) (Nm/g), CSA = average cross-sectional area of fibres' solid fraction (m2), 50 P = fibre perimeter (m) of the average fibre cross-section (m), L = fibre length (m), p = density of cellulose (g/m3), RBA = relative bonded area in the sheet (constant assigned 0.5), b = shear strength of the fibre-fibre bond (load) (constant assigned 5.9* 10"2) (N/m2). Tearing resistance of paper (TR) Clark (1985) derived following formulae for pulp properties indicating the relative importance of basic fibre and paper sheet formation properties. For weakly bonded sheets the resistance formula is: T R ^ ^ L ' ^ C ' V V - 1 And for well bonded sheets: TR 2 = Jfr 2z 1- 5L 0- 5c-°VV 1- 5 where: - TRi = tearing resistance (moment) of paper, - K, = constant, - Z = intrinsic fibre strength, - L = fibre length, - C = fibre coarseness, - S = cohesiveness, - V = bulk or specific volume, Burst fracture resistance (BF) Clark (1985) defined bursting resistance for weakly bonded sheets as: B F ^ ^ ' L ' ^ C - ' W 0 ' 1 And for well bonded sheets: B F 2 = ^ 2Z 1 0L 0 1C- 0^°V 1 0 The following basic properties of mechanical pulps can also be considered: Strength of mechanical pulp (Tm) Rudie et al. (1994) cited a simple formula that under certain conditions gives a straight-line relationship with pulp breaking length, at a fixed specific energy consumption: 51 T m = P/CSA w h e r e : T m = tensile strength (breaking length) of mechanical pulp, P = fibre perimeter of the average fibre cross-section, CSA = average cross-sectional area of fibres' solid fraction. 3.5.2 Optimization of Single Objective Functions 3.5.2.1 Maximization with Genetic Responses as Constraints Derivation of selection indices was based on the objective functions defined above. For a linear selection index, index weights were first found by the analytical method of Itoh and Yamada (1988) that is considered optimal in case of a non-linear relationship among variables. The results were compared with those obtained by a simpler method proposed by Harris (1970), which uses linear approximations of the objective functions. Non-Linear Optimization If expected selection responses (d) form an ellipsoid, it is defined by: d'G'PGld=i2 w h e r e : G = genetic variance-covariance matrix P = phenotypic variance-covariance matrix / = selection intensity. Among all ds which satisfy the above equation we have to find those which maximize expectation: AE{x)) =fip + d) where: fj. = vector of population means Iterative maximization of these objective, and constraint functions were done by the Solver option in Microsoft Excel® (1997). Microsoft Excel Solver uses the Generalised Reduced Gradient (GRG2) nonlinear optimization code developed by Lasdon et. all (1972) and Lasdon and Smith (1992). 52 After obtaining d we can get the index weights b from: b = G'd The expected response in various objective functions can be obtained by this method and sets of parents selected based on different index weights compared. The expected correlated response in individual traits forming an index can be obtained from: aGb A./U = i 4bPb Optimization using linear approximations The above optimization procedure can be simplified by using the method based on Taylor series. The linear approximation of objective functions is f{x) =M + t'(x-M) This method uses the partial derivatives evaluated at the mean before selection as economic weights, in conventional constructing of selection index. The vector of economic weights is expressed as: df/dx2 t = ¥/d*„. X = /J / i s a objective function, and JC and // are phenotypic and mean vectors of component traits, respectively. Index weights can then be expressed as b = FxGt. 3.5.2.2 Sensitivity Analyses Sensitivity analysis provides information about how sensitive an optimization solution is to small changes in parameters of target formula or constraints. For non-linear models, the Solver® report provides values for reduced gradients and Lagrange multipliers. There are numerous factors that influence parameter values for the objective functions or the constrains, such as: 53 - Genetic parameters are estimated with sampling errors that can sometimes be large; - Heritability is decreasing over years (Kiss and Yeh 1988). If genetic correlations between traits change over time, assuming that the selections are still made on the basis of former correlations could lead to a substantial decrease in actual gains (Hansen in press); - Sampling errors can be confounded with effects such as GxE. For example, the gain and its estimation precision would be affected, i f environment-specific variance-covariance matrices were used instead of overall or averaged across all environments matrices (Cladwell and Weber 1965, Hanson and Johnson 1957); - Genotype by environment interactions for compound traits, i.e., selection indices, may exist even if there is no interaction for the component traits (Namkoong 1985); - Genetic parameters are estimated under assumption of a certain degree of inbreeding which is actually unknown (Namkoong 1966, Park et al. 1984); - The amount genetic variation over generations changes as a consequence of particular selection strategy (Villaneuva and Kennedy 1990); - Theoretical value functions are only approximations, and it is generally not known how well they fit empirical data. Limited information can be obtained from the original papers where empirical testing was done (Page and Seth 1988, Page 1993, O'Neil et al. 1999); - Including dispersion parameters instead of just mean value for a tree, we could perhaps obtain different values for objective functions. For example, Karenlampi (1995a,b,c) studied the question of physical strength the mixture of two pulps and further the effect of a distribution of fibre properties on pulp strength; - Value functions are defined for the present and not for uncertain future situations, and they may not actually predict future reality. These relationships and inter-dependencies, i.e., sensitivity of predicted genetic gain to variation in estimated population parameters and to exactness of value function prediction, were modelled through previously described Monte Carlo simulations. 3.5.3 Multiple Objective Functions Analogous to single objective functions, two or more objective functions can be optimized simultaneously. The optimization constraint, i.e., the ellipsoid of expected selection responses includes all traits on which those functions are based. Firstly, the maximization was done for 54 one objective function letting all other function values vary. Next maximization is done for another function, letting all other functions, including those previously maximized, to vary. Multiple intermediate solutions were generated for the range between alternative function maxima by decreasing the previously maximized functions. Eventually index weights could be derived for any chosen solution in the generated set of optimal solutions. 3.5.3.1 NIMBUS Method for Multiobjective Optimization An interactive method for multiobjective optimization problems NIMBUS (Nondifferentiable Interactive Multiobjective Bundle-based optimization System) of Miettinen and Makela (1995) can be used for finding non-dominated set of solutions for considered functions. A non-dominated solution is the one for which value of one function cannot be increased without decreasing another function of the same variables. NIMBUS is suitable for both differentiable and non-differentiable, multi-objective and single-objective optimization problems subject to nonlinear and linear constraints with bounds for the variables. The problems to be solved are of the form: minimise {/i(x),...,/k(x)} subject to gi(x) < 0 ... gmW < 0 x'<x<x" w h e r e k = n u m b e r o f t h e o b j e c t i v e f u n c t i o n s , m = n u m b e r o f t h e n o n l i n e a r c o n s t r a i n t s , x = d e c i s i o n v e c t o r a n d i t s l o w e r a n d u p p e r b o u n d s a r e n - d i m e n s i o n a l v e c t o r s , A l l the functions involved are assumed to be locally Lipschitzian in an n-dimensional Euclidean space, i.e., differentiable or convex functions are locally Lipschitzian. Pareto optimality and weak Pareto optimality are used as optimality concepts. A criterion vector z (consisting of the values of the objective functions at a point x) is Pareto optimal, if none of its components can be improved without impairing at least one of the other components. A criterion vector z is weakly Pareto optimal, i f there does not exist any other vector z' for which all the components are better (NIMBUS 2000). 55 3.5.3.2 Choosing among Alternatives In the NIMBUS method, the idea is that the user examines the values of the objective functions calculated at a current solution and divides the objective functions in up to five classes. The classes are functions whose values: should be decreased, should be decreased down till some aspiration level, are satisfactory at the moment, are allowed to increase up till some upper bound, and are allowed to change freely within the feasible set. A new problem is formed according to the classification and the connected information. This problem is solved alternatively by a multiobjective proximal bundle method (Miettinen and Makela 1995, 1996). The flowchart of the NIMBUS algorithm is given in the Figure 3.5.1. 56 User: Choose a starting point. Calculate the ideal criterion vector and the weakly Pareto optimal counterpart of the starting point. User: Classify the objective functions and specify aspiration levels, upper bounds and weighting coefficients, if necessary. Formulate the new problem and solve it. Present the old and the new solution to the user. I Calculate and present the alternatives to the user. User: Select the most preferred alternative. Figure 3.5.1 Flowchart of NIMBUS algorithm. The result of the multiobjective optimization is a vector, where the components are the values of the objective functions. When optimizing the functions individually and creating the vector of these values, the Ideal Criterion Vector (ICV) is obtained. The ICV tells us the best solution that exists for each objective, when the functions are treated independently. However, the ICV vector is infeasible and one must make compromises. The ICV represents the upper bounds of the criterion values in the set of Pareto optimal solutions. If the problem is complicated, i.e., non-convex, the actual components of ICV are difficult to calculate. Thus it is referred to as an estimated ICV. Nadir vector, on the other hand, consists of component User: Specify their number. 57 values for the 'worst case' solution scenario, i.e., the lower bounds of the criterion values in the Pareto optimal set. In practice, the Nadir vector is also only an estimate. When the first classification is done by NIMBUS, two different solutions can be created. Using the starting point as basis classification produces the first solution. The second is the result of a new classification. After determining i f the classification has moved the result into a desired direction, a number of alternatives in between the two solutions can be examined. The alternatives are simply intermediate solutions. After choosing one of the alternatives, the selected alternative is the basis of the new classification, and the iteration proceeds. More and more specific results from the classification are obtained and the best result is chosen for further examination. The iteration process depends on two factors: the right class is chosen for each function, i.e., direction for the objective values, and the choice of values allowed to decrease. 3.5.3.2 Sensitivity Analyses Beside factors mentioned, which influence parameter values of each objective function or related constraints, in the case of multiple objective optimization, choice between two alternatives may depend on additional complexities and interactions. Furthermore, changes in decision-maker's preferences are reflected in the amount of necessary trade off between some conflicting objectives. Sensitivity to such changes was also considered, and evaluated here. The analysis here was restricted to using a deterministic nonlinear multiobjective method, but promising new developments in multiobjective methods will be available in the future in researching the problems similar to the one at hand that is stochastic and fuzzy in nature (Miettinen 1999). 3.6 D E V E L O P M E N T A N D E V A L U A T I O N O F B R E E D I N G S T R A T E G I E S 3.6.1 Selection Scenarios for Different Breeding Strategies The results of the optimization processes were used as a basis for evaluating selection scenarios in different breeding strategies. Optimising selection for a single objective function within a single breeding population at a time was considered firstly. Multiple objective 58 functions within a single breeding population were considered secondly. Selection for multiple objective functions using two or more breeding populations was considered thirdly. 3.6.1.1 Single Breeding Population Single objective function Selection based on linear selection indices and non-linear objective functions were optimized. Selection indices were derived for each of the following objectives: one that maximizes response in volume ( I V O L ) , one that maximizes response in volume and places restriction on change in relative density ( I V O L _ R D ) , and one that maximizes response in dry-weight ( I D W T ) - The objective functions related to pulp and paper properties were also considered: tensile strength of wet-webs (T w w ) , tensile strength of paper (Tp), tear of weakly (TR|) and well bonded (TR2) paper, burst of weakly (BRi) and well bonded (BR2) paper, and tensile strength of mechanical pulp (Tm). Different sets of parents were obtained based on their ranks according to each of the above indices. Multiple objective functions Accommodating multiple objective functions within a single breeding population requires multiobjective optimization to find the optimal economic weights on selection index. The following selection scenarios were created: S c e n a r i o Breed ing Se lec t ion Object ive intensity ( / ) a V O L , T w w 1,2 b V O L , T p 1,2 c V O L , T R , 1,2 d V O L _ R D , T w w 1,2 e V O L , Tp, T R , 1,2 59 After maximization was done for one objective function letting other functions to vary, the next solution is found and multiple intermediate solutions were generated for the range between different function maxima. Those solutions represented the range of necessary trade-offs between improvement in different objective functions depending on how much emphasis was placed on each particular objective. The trade-offs were graphically presented. Allocation of objectives according to specific criteria was examined more closely. Those were maximum improvement possible in any objective function (MaxiMax), maximized average value (MaxiAvr), maximized minimum improvement (MaxiMin), and minimized maximum loss (MiniMax). Under uncertainty about relative values of objective functions in the future, those options represent different risk-management strategies, according to particular attitudes towards risk. Index weights were derived for some chosen solutions in the generated set of optimal solutions. Choosing different selection objectives would give different sets of selected parents. Value of a certain selection scenario was determined based on the potential for improvement in all objectives as an average. No assumptions were made as to what would be the future value of objective functions, at what probability would that value can be expected, or how the values interact. Some realistic values were considered, however. If the estimated values of objective functions can be improved, further optimization can be obtained through iteration using the multiobjective method described above. Here a range of alternatives is presented, and choosing between alternatives by classification of functions would depend on decision-maker's preference and risk attitude. Sensitivity between the two points of interest could be checked, by simply inserting the number of new alternatives. Value of a breeding population is estimated based on expected values of its objectives and expected value of genetic gain. According to Namkoong (1976), for a single population with a single objective function, i f estimated index weights have an error distribution around optimum f(b), and a value function is also a probability density function g(b), the value of the population is the product of the value it has at the point gj(b) and the probability fj(b) that b is at an optimum: v= J...{g(b)df(b). K b, 60 Expected gain from a particular risk management strategy would depend on both relative expected values and variances of g(b) and f(b). If uncertainty is high, diversification of breeding populations may be effective. 3.6.1.2 Multiple Breeding Populations To apply the multiple population breeding (MPB) option, multiple selection indices for multiple objective functions were incorporated. According the model of Namkoong (1976), diversification can provide gains especially if variance of g(b) is large relative to variance of f(b). Relative advantage of having additional populations, beside the one selected for volume, was examined first assuming that utility equals average of the potential improvement in all objectives. Other assumptions were made then as to what would be future relative importance for objective functions, and advantage of diversifying breeding population assessed again. 61 VI RESULTS 4.1 HERITABLE VARIATION IN GROWTH, WOOD MACRO-PROPERTIES, AND TRACHEID CHARACTERISTICS 4.1.1 Growth and Wood Macro-Properties Cumulative growth and wood macroscopic traits were evaluated first. Height (H) and diameter (D) measured at ages 20 and 22, for East Kootenay (EK) and Prince George (PG) progeny tests were studied. (There was a 2-year difference in age between E K and PG progenies.) Wood macro-properties included average ring width (AVRW), average late-wood percentage (AVRLW%) and average relative density (AVRD) for ages after planting 17-20 (cambial ages 12-15) for the EK, and for ages 17-22 (cambial ages 12-17) for the PG progenies. Site means and ranges of family means are given in Table 4.1.1. Estimated variance component ratios and heritabilities are given in the Tables 4.1.2 and 4.1.3. Heritability is given only i f family (F) variance component was found to be significantly different from zero. Site-specific heritabilities were also estimated. Phenotypic and genetic correlations among traits are given in the Table 4.1.5. 62 Table 4.1.1 Site means and ranges of family means, for the East Kootenay (EK) progenies on three test locations: Red Rock (RR) (Prince George), Jumbo Creek (JC) (Inveremere), and Perry Creek (PC) (Kranbrook); And Prince George (PG) progenies on two test locations Red Rock (RR) and Quesnel (Q). Traits are height (H), diameter (D), average ring width (AVRW), average late-wood percentage (AVRLW%) and average wood relative density (AVRD). Breeding Test Traits unit site H(m) D(cm) AVRW(mm) AVLW% AVRD EK RR 5.1 3.0-7.1 9.0 5.5-12.9 3.0 1.8-4.4 20.4 13.5-29.6 0.363 0.292-0.455 JC 4.3 2.7-6.1 7.0 4.3-10.3 2.7 1.5-3.8 26.7 15.6-43.5 0.383 0.298-0.501 PC 5.5 3.9-7.8 9.7 6.1-14.0 3.3 2.0-4.9 22.2 13.2-31.9 0.362 0.299-0.492 PG RR 5.2 3.5-6.9 9.0 5.0-11.3 2.2 1.5-2.9 27.0 13.5-29.6 0.375 0.325-0.413 Q 5.0 2.4-6.4 9.8 4.4-14.0 3.4 1.8-4.6 28.6 21.2-37.5 0.348 0.314-0.414 63 Table 4.1.2 Estimated variance components for site (S), replication within site (R(S)), family (F), and site by family (SxF) effects, and heritabilities with standard errors (individual-based h2, and family-based h2/) for East Kootenay (EK) progeny tests. Heritablilities are given as overall (ALL), at the Red Rock (RR) site only, at two sites in East Kootenays (EK) combined, or separately as Jumbo Creek (JC) and Perry Creek (PC). Traits are height (H) and diameter (D); average ring width (AVRW), average late-wood percentage (AVRLW%), and average relative density (AVRD). T e s t S o u r c e of Deg . of Traits Var iat ion F r e e d o m H D A V R W A V L W % a A V R D E K S 2 0 .346" 1 .298" 0 . 0 5 7 n s 0 . 0 0 0 n s 0 . 7 0 2 n s R(S) 9 0 . 0 2 5 n t 0 . 1 7 0 n t 0 . 0 4 0 n t 2 2 . 8 4 n t 0 . 9 0 5 n t F 79 0 . 2 5 1 " 0 . 3 5 8 " 0 . 0 3 7 n s 7.330** 2.256** (0.60) SxF 158 0.077* 0.316* 0.103** 4 .632* 1.284* Error 711 1.031 3.587 0 .586 4 6 . 2 2 0 13.538 h", A L L 0.74±0.16 0.34±0.12 - 0 . 5 U 0 . 1 4 0.52±0.14 R R 1.38±0.16 0.90±0.19 1.11±0.20 1.03±0.22 1.09±0.20 E K 0.38±0.16 - - - -J C - 0.73+0.29 0.99±0.31 0.66±0.30 P C - - - - -A L L 0.69±0.07 0.45±0.12 0.58±0.10 0.56±0.09 R R 0.76+0.07 0.64±0.06 0.70±0.05 0.51±0.07 0.69±G,51 E K 0.24±0.08 - - - -J C - - 0.40±0.11 0.49+0.09 0.66±0.30 P C _ _ a arcsine square root transformed proportion data. 'significant F statistics or likelihood deviance at a=0.05 level; "significant F statistics or likelihood deviance at 0.01 level; "'not tested; nsnot significant (if F or SXF sources were found insignificant the power of tests of significance is given in the parentheses). 64 Table 4.1.3 Estimated variance components for site (S), replication within site (R(S)), family (F), and site by family {SxF) effects, and heritabilities with standard errors (individual-based h2, and family-based h2}) for the Prince George (PG) progeny tests. Heritablilities are given as overall {ALL), or at the Red Rock (RR) or Quesnel (Q) site separately. Traits are height (H) and diameter (D); average ring width (AVRW), average late-wood percentage (AVRLW%), and average relative density (AVRD). T e s t S o u r c e of Deg . of Traits var iat ion f reedom H D A V R W A V L W % a A V R D P G S 1 0 . 0 0 7 n s 0 . 2 1 2 n s 0.749** 1 . 3 3 0 n s 3.598** R(S) 10 0 .136 n t 0 . 6 1 9 n t 0 . 0 4 6 n t 0 . 6 6 1 n t 1.359 n t F 79 0.251** 0 . 3 1 7 n s 0 . 0 2 0 n s 2 . 1 7 8 n s 1.070** (0.40) (0.70) SxF 79 0.098* 0 . 4 1 4 n s 0 . 0 2 0 n s 1 . 2 0 3 n s 0 . 3 3 6 n s (0.49) (0.62) (0.49) (0.45) Error 790 1.455 7 .895 0 .600 39 .117 9.826 h'< A L L 0.56±0.15 - - - 0.36±0.13 R R 0.70±0.18 0.30±0.15 0.25±0.15 - 0.46+0.17 Q 0.84±0.20 0.37+0.16 0.24±0.15 0.43±0.19 0.54±0.17 h2f A L L 0.60±0.15 - - - 0.50±0.18 R R 0.56±0.07 0.33±0.12 0.29±0.13 - 0.44±0.10 Q 0 .6110.07 0.38±0.11 0.28±0.13 0.42±0.12 0.48+0.09 a arcsine square root transformed proportion data. "significant F statistics or likelihood deviance at a=0.05 level; "significant F statistics or likelihood deviance at 0.01 level; "'not tested; nsnot significant (if F or SXF sources were found insignificant the power of tests of significance (/3) is given in the parentheses). 65 4.1.1.1 Height Growth For the E K progeny test, family means ranged widely between and within sites (Table 4.1.1). Site (S) differences were statistically significant. Highly significant overall family (F) variance component and high overall heritability estimate were obtained. There was also present SxF interaction, and site-specific heritability was especially high at the RR site (Table 4.1.2). For the PG test, family means ranged less between sites (Table 4.1.1), and site differences were not statistically significant. Substantial family variance component was found on both sites. Interaction SxF was significant (Table 4.1.3). The observed site differences and heritability estimates for H, however, may be significantly influenced by family differences in pest resistance. Some sites were damage-free while others were severely affected by weevil or gall-aphid attacks. For the E K progenies, the incidence of damaged leader, as assessed in 1994, was much higher for trees at the RR site than for trees at the JC and PC sites. For the PG progenies, the incidence of recurrent weevil attacks in 1992 was also higher at the RR site. But since the incidence of initial weevil attacks was high in that year at the Q site, it might have also affected the trait H , measured in 1995 (Table 4.1.4). When families were separated into two classes: resistant (free of attacks) and susceptible (attacked), within-class individual and family heritability for H of PG progenies at RR site among resistant families decreased significantly. It did not change significantly at the Q site and among E K progenies at the RR. 66 Table 4.1.4 Percentage of trees per site and per replication recurrently attacked by weevil and forked (WF), or heavily attacked by gall-aphids (G), in 1994 at age 20 for the East Kootenay (EK) progenies. Percentage of trees recurrently weeviled and forked (WF), or initially attacked by weevil (WI) in 1992 Br. Test Rep. Percent attacked Br. Test Rep. Percent attacked unit site W F G unit site W F WI EK PG RR RR 1 21 0 I 12 0 II 13 0 II 12 0 III 31 0 III 13 0 IV 25 0 IV 32 0 V 25 0 V 50 0 VI 16 0 VI 44 0 Avr. 22 0 Avr. 27 0 JC 1 13 3 II 04 3 Q III 03 5 I 19 15 Avr. 6 3 II 17 18 PC III 20 18 1 00 0 IV 21 17 II 01 24 V 17 16 III 04 24 VI 15 23 Avr. 2 16 Avr. 18 18 Note: Data - courtesy of the B.C. Ministry of Forests, Kalamalka Forestry Centre 67 4.1.1.2 Radial Growth Radial growth was examined on sample disks. Total cumulative radial growth was represented as diameter (D). Growth of wood with cambial age greater than 12 years was analysed separately. It was represented as average ring width (AVRW) in the older portion of juvenile growth. Diameter Among E K progenies, variation in D due to the site was more pronounced than for H (Table 4.1.1). Variation due to site was highly significant. Overall family differences were significant. They were somewhat less pronounced than for H, which resulted in a lower heritability estimate. Interaction SxF was also significant, and site-specific heritability was high at the RR (Table 4.1.2). Among the PG progenies, there were no statistically significant differences between the two site means. The site Q had somewhat wider within-site range of variation than the RR plantation. Overall, family variance component was not significant, while site-specific heritabilities were significant but low. Heritability of D was generally lower than that for H. Interaction SxF was not significant (Table 4.1.3). Average Ring Width For the E K progeny test, for A V R W (cambial ages 12-15) differences between sites were less pronounced than for D (Table 4.1.1). Variation due to site was not significant. Overall family variance component was not significant. Significant variation occurred in SxF interaction form, and site-specific heritabilities were high at the RR and JC sites (Table 4.1.2). Although, for both E K and PG progeny tests ranking of sites for A V R W was similar to the one for D, the magnitude of differences was not the same. This was especially so among PG progenies, where Q plantation had much higher A V R W than RR plantation. The difference was much stronger than for D, and variation due to site for A W R V was highly statistically significant. Perhaps differences in climate itself, in the 1990-95 period, or climate-induced maturation of cambium resulted in significantly narrower rings for PG progenies on the RR. Overall family variation was not significant. There was no significant genotype by environment interaction, and site-specific heritabilities were low (Table 4.1.3). 68 4.1.1.3 Wood Macro Properties Average Latewood Percentage A V L W % showed little variation due to planting site (Table 4.1.1). Site differences were significant neither for EK, nor for PG tests. Genetic variation was more pronounced among the E K progenies. There was also significant SxF interaction, and there was high heritability at RR and JC sites (Table 4.1.2). Locations of tests used for E K progenies are in both East Kootenay and Prince George regions, the latter being outside the region of origin of parent trees. For PG progenies estimates of genetic family variance component was significant only at the Q site. There was no significant SxF interaction {Table 4.1.3). Average Relative Density For E K progenies the lowest recorded A V R D family-mean was at the RR site, and the highest at the JC site. Those were also the sites with the highest and the lowest A V R W , respectively (Table 4.1.1). There was a significant overall family component of variation. There was also significant SxF interaction. High heritability was found at the RR site. (Table 4.1.2) . Although the sites were more diverse in E K tests, only PG tests exhibited significant variation among sites for this trait. For PG progenies, lower relative density recorded at the Quesnel site was probably related to its higher average ring width (Table 4.1.1). This result is different from the one of Yanchuk and Kiss (1993), where there was no site variation for 40 families evaluated at age 15. Considering the fact that A V R D was strongly related to radial growth rate, and that there was a significant difference between sites in A V R W in more recent growth at ages 12-17 for which A V R D was measured, this was expected. When A V R W was included the analysis as a covariate, the site effect for A V R D disappeared. Overall genetic variation was highly significant. But heritability estimates were somewhat lower than those reported by the above authors. There was no significant SxF interaction for A V R D (Table 4.1.3) . 4.1.1.4 Correlations among Traits Since heritabilities differed considerably among sites, and since estimation precision of 69 genetic correlation is directly proportional to heritability of traits (Falconer 1989), site-specific genetic correlations were calculated. For E K progenies, heritabilities were the highest at the RR site, and for PG progenies at the Q site, therefore genetic correlations were estimated for those sites (Table 4.1.5). Significant negative genetic correlation of growth with wood density, i.e., H , D, and A V R W with A V R D was found for the E K test. No significant genetic correlation could be found in the PG test. Yanchuk and Kiss (1993) also reported virtually no genetic correlation of height and diameter growth with wood relative density, in the PG test. Generally, genetic correlations were all of the same sign as phenotypic correlations, but environmental correlations for H with A V R D and H with A V L W % , were of the opposite sign on both sites. The phenotypic correlation between A V L W % and A V R D was moderate, and genetic correlation was statistically significant for E K progenies. Genetic correlation between A V L W % and A V R D was not significant for PG progenies. When width-weighted average late-wood percentage was used, i.e., the sum of late-wood widths over the total width of rings, both correlations were slightly higher. The site differences in genetic correlations can be seen as an indication of complex patterns of change for the genetic relationships across environments. Certain change can also be expected with tree age, and these patterns will be studied in more detail for growth and wood density in the next section. Table 4.1.5 Genetic (top), and phenotypic correlations (bottom value), for EK(RR) (above the diagonal) and PG(Q) progeny tests (below the diagonal). Traits are height (H) and diameter (D); average ring width (AVRW), average late-wood percentage (AVRLW%), and average relative density (AVRD). H D AVRW AVLW% AVRD H 0.94**±0.05 0.77**±0.02 0.94**±0.05 0.70**±0.03 -0.67**±0.19 -0.25*±0.10 -0.49*±0.13 -0.39**±0.08 D 0.92**±0.06 0.80±0.02 0.98**±0.04 0.73**±0.02 -1.11**±0.18 -0.31*±0.09 -0.67**±0.11 -0.52**+0.07 AVRW 0.97**±0.19 0.72**±0.02 0.88**±0.14 0.82**±0.02 -0.95**±0.13 -0.42**±0.08 -0.68**±0.10 -0.53**±0.07 AVLW% -0.64*±0.20 -0.37**±0.08 -0.60n s±0.26 -0.52**±0.07 -0.50n s±0.36 -0.54**±0.07 1.10**±0.08 0.55**±0.07 AVRD -0.36 n s±0.20 -0.45**±0.08 -0.49 n s±0.23 -0.58**±0.06 -0.54n s±0.27 -0.61**±0.05 0.80*±0.14 0.65**±0.05 * -Significant likelihood deviance for the component of covariance at a=0.05 level; ** -significant likelihood deviance at «=0.01 level; ns-not significant. 70 4.1.1.5 Stability of Family Performance In addition to the tests presented above for significance of SxF variance, stability of family performance over sites was examined in more detail here. Stability of family performances was also examined over growing seasons at a particular site. Site by season interactions were also considered. Genotype by Environment Interaction The genotype by environment interaction effect (GxE) was initially examined by plotting family means at one site against means at another site. For example, E K progenies performed better on average than the local PG progenies on the RR site. There was, however, an obvious GxE interaction in family response. Some families, e.g., 29 and 70, performed excellent in both regions. Some others, such as family 31 was the best in the region of origin, but only slightly above average after transfer. Family 71 was below the average in the region of origin, but among the best after transplantation (Figure 4.1.1). In a similar fashion, family means at all other sites within a particular region were plotted against each other, to visualise family responses on different sites (APPENDIX 3). 120 60 -I . EK-RR E K - J C P C Figure 4.1.1 Family performances for E K progenies at the RR site in the PG region (EK-RR), and at combined JC and PC sites in the region of origin (EK-JCPC). The trait shown is cumulative radial growth (D) 20 years after planting. 71 The GxE interaction was statistically evaluated first through significance of FxS variance component (Tables 4.1.2 and 4.1.3). In most cases, however, the statistical power of this test was low. Moreover the residual variance for some traits at different sites was not always homogenous. In such a situation, data for the families from different sites could be better linked, by calculating genetic type-B correlations across sites Burdon (1977). This was done for the PG and Q sites (Table 4.1.6). It could not be done for the JC and PC sites, since family variance at one of two sites was always equal to zero. Table 4.1.6 Genetic type-B correlations across PG and Q sites. Traits are height (H) and diameter (D); average ring width (AVRW), average late-wood percentage (AVRLW%), and average relative density (AVRD). H D AVRW AVLW% AVRD 0.70**+ 0.17 0.34ns+0.38 0.20 n s±0.49 0.48 n s±0.39 0.98*±0.31 It was also possible to generate three-way interactions from data sets by including the growing seasons as an additional factor, and this problem will be addressed in the following sub-section. Age-Age Correlations Means and ranges of family means for RW and L W % over nine growing seasons are summarised in the APPENDIX 4. The trends are graphically shown in the Figure 4.1.2a and Figure 4.1.2b. Fluctuations related to cambial age were confounded with those related to climate effects, for each growing season. It is obvious that the fluctuations are much more pronounced for annual than for the cumulative increments. However, when family means at different sites were plotted over growing seasons, presence of genotype by year interaction with some rank changes, was observed. This interaction was analyzed through repeated measures analysis, and the results are presented in the Table 4.1.7. Although family by year interaction (FxY) variance component was significant over the entire period, for certain years family responses, i.e. regression coefficients, did not differ significantly. For example, FxY variance for the E K progenies on RR site, for RW in years 1994 and 1995, was not significant (p=0.092), but when year 1993 was included into calculation it became highly significant (p=0.002). 72 E K ) 14 15 16 17 18 19 20 21 22 Tree A g e 55 : 45 17 18 19 20 21 22 23 24 25 Tree A g e Figure 4.1.2a Trends of family means for each year (RW), and cumulative (D). Data are for EK progenies at RR, JC, PC sites, and for PG progenies at RR and Q sites. Cambial ages were 7 to 15 for EK and 9 tol 7 for PG progenies. 73 - • - R R - A - P C RR Q 21 Tree Age Figure 4.1.2b Trends of family means for LW%, for each year and cumulative. Data are for EK progenies at r r , JC, PC sites, and for PG progenies at r r and q sites. Cambial ages were 7 tol5 for EK and 9 tol7 for PG progenies. 74 Estimates of variance components from year to year and resulting individual heritability are given in Figure 4.1.3. They are given only for the sites where individual growth increments had significant genetic variation. It can be seen that heritability was generally determined more by fluctuations in the error component than in the family component. Phenotypic and genetic age-age correlations for RW and L W % at different sites were calculated, for both individual and cumulative increments. The genetic correlations were generally high for the cumulative increments, and a decreasing trend with lag between measurements was observed. This is illustrated in Figure 4.1.4, for the E K progenies at RR site. Table 4.1.7 Variance components for ring width (RW) and late-wood percentage (LW%) with sources of variation family (F), tree within family (T(F)), and family by year (F*Y) interaction, at sites RR and JC for the EK progenies and RR and Q for the PG progenies. EK PG Var iat ion RR j C a RR Q source RW LW% RW LW% RW LW% RW LW% F 0.13** 8.63** 0.08** 12.62** 0.03NS 0.81NS 0.06NS 1.29NS T(F)-Error1 0.40 18.37 0.36 32.49 0.46 19.56 0.73 22.15 F* Y 0.22" 8.41* 0.04* 10.18* 0.04NS 8.15** 0.13** 8.12* Error2 0.42 53.36 0.18 46.41 0.30 47.03 0.39 60.18 there were only 3 replications for this site * -Significant at a=0.05 level; ** -significant at a=0.01 level; n s - no t significant. 75 A ) R R - R W r s H 2 l h 2 1 T r h I T T 1-6 . _ 12 15 16 17 18 19 20 21 22 Tree Age Figure 4.1.3 The change of variance components (VarComp) with tree age (cambial ages 7-15) and resulting individual heritability (h2) for ring width (RW) and latewood percentage (LW%), for EK progenies at RR and JC sites. 76 0.75 -I , , , , , , , 1 2 3 4 5 6 7 8 Lag belvveen rreasurerrents Figure 4.1.4 Genetic age-age correlations for ring width (RW) and latewood percentage (LW%) among E K progenies at RR site, for individual and cumulative increments. ,77 4.1.2 Tracheid Characteristics 4.1.2.1 Traits Related to Fibre Cross-sectional Dimensions Morphological traits related to cross-sectional dimensions of tracheids were analysed first. Data sets were summarised by calculating site means and ranges of family means for one ring at cambial age 14 for the EK, and two rings at cambial ages 12 and 16 for the PG progenies (Table 4.1.8). Traits included were ring width (RW), number of cells per ring (CN), cell radial size (CSr), cell tangential size (CSt), mean double wall thickness (DW), mean ratio DW/CS (R), early-wood percentage (EW%), transition-wood percentage (TW%), late-wood percentage (LW%), modified Mork's index (MI), double wall thickness in early-wood (DWew), cell radial lumen in early-wood (Lew), ratio DW/CS in early-wood (Rew), double wall thickness in transition-wood (DWtw), cell radial lumen in transition-wood (Ltw), ratio DW/CS in transition-wood (Rtw), double wall thickness in late-wood (DWlw), cell radial lumen in late-wood (Llw), ratio DW/CS in late-wood (Rlw), fibre length (FL), and micro-fibril angle (MFA). Variance components for site (S), replication within site (R(S)), family (F), site by family (SxF), error (Err.) were calculated across all sites (ALL). Because of significant SxF interactions for many traits, site-specific heritabilities (h2) were also calculated. For East Kootenay progenies, site-specific heritabilities are presented separately for the Red Rock site in the Prince George region (RR), Jumbo Creek (JC) and Perry Creek (PC), in the East Kootenays combined (EK), and for the JC and PC sites separately (Table 4.1.9). For Prince George progenies, variance components were calculated as a weighted average of two rings. Heritabilities are reported as overall (ALL), and site-specific, at the Red Rock (RR) and at the Quesnel site (Q) (Table 4.1.10 ). The same analyses, but for each of the two rings separately are given in the Table A 5.2. 78 T 3 IT-CD CD Oi o 8 (/) L CD CD « -C — 1—1 .2 . "3 cn CO CD CD ao o o 2 on „ U =? £ £ Q b -2 „ ta CD l ~ 60 „ | 9 § 3 ^- v ^ ° -o o ° 5 S • £ >. •L ~ >> co ~ CD CD .5 w 0 0 O CD CD Q_ o fc O 3 & <" 2 * . 00 3 •S "> CD Q « o 13 CD OH O •2 CD S -o " E ^£ CO rV C w CD O w — W CO CD ^3 so 12 co S CD ' - o X J O o o % * 5 ta O —1 S c cn " - " g oo 2 U -t-» B £ 3 Q CD O s i <" 00 £ .>> — CO _ .S CD )-2 £ co -—' s Q .a o -s 41 £ 1 3 — " CO —1 V^ CD O O O > CD 2 to la == £ 1 ca C ca CD J3 CD C a S S u p..s c CD E 3 Q Q < s c n 3 lo SI VI CD CD CL 00 2 U - a ^ is on CD ca O CD E ,5 I 90 r ; c n s i •a o o • CD ta U 1^ 1 CD 0 0 ca c CD O 1— CD CD T 3 0 0 O c o ca CD ^ !B 3 O T 3 CD O ^J2 jj ( V j 3 o S o ? ? -a S o CD T 3 LL. CU 5 : DC ! a: s^i 1^  5 a:1 5 cu Q I I w 33 o ^ o II L U H co C O CO p (O O n CO JO^-CD CO o IT) CM oo CN c4 r j T— o ob <<f 2.™ csi co cri co 3 CD o f o g co tv fj-[v. co co in cn ST o CD co CM CN CO ^  CO CN i v 5? o m 5J-in co o CO f-j,^ C O co p C O f^ J_ ^ CO CNJ 00 c\i CM CM. co CD CD p ID cd ^ CM OCO U l ffl? ^ oi ^ rv <o (D t\i ~ m in CM m co oi CM CNJ,c\l ^ ^ o in CNJ cd CM C i S °° coSi ^ CD CO 9J " co CO q CM CM <M ^ OL a. 00 O co' co r2. '* S J2 ™ O N ^ CM jf.CN « P fv 5T N t i b vf SJ.CO O CO CD oS co 58s CD S2.cd O c v l ? C M c o CM r ^ c o CO CD cS CD ^ T-' O if) C in CN CN co, CO m d ^ T- ^ CM S? !2 « N CO o CO CNco <9 ^ cri •st Citv 00 d> 5? 01 co s CM CM co IV- ixi, £7 CO ^  CN in p CD CO o « CO •* o sr CD co cji CM : 5t 0 0 5T CO CO co J2.CD in 4 j CO CO o rv cfj CM J^ CN S °> s O co CO CN CO ^ CO 5" iri cri to co cM.Sr ^ ^ n o J CM CM.CN O 6 Ljr O N T 2. CO CD CO ^ ' CM j!; S SS S" CO co co CD 3, CO CD CO to" CO CO LO O »l ^ CM I ^ C M S9  m ST co m p: ^ rv co C M oi rv CO m o cri CM CM CN in oo ^ in co J C O CM ^ CM CD fo Iv CO co co r^co CO c i CM O CO o CO JM CO in O) CN rv tv' cb CM r^co ^ ^ CO CM -ST CO •St G.^ ID co S-CO N ^ CM r^ co f S2.CD CD co cri CM CM CM co a> ST •st d CM JN CN °! CN CM <D csi T t CSI Si CD CO C M . ^ : O a. ID CD CN iri oi ^ CO CM ^ . CO 4 CO CO r- m Csi CM csi rv o CO O ) cd •St C O . CO CD CO co ° ° .» d CO JO T— O o I D T o cb i f i cb ^ cn I D CM cb csi CSI ^ .co rv •* d a> co io" cp rv S -cr co. to p d> to no oq So CM O l ib CM I^CM co <2.-<r CM cb ST oo d o •st <2.f; I- CO co a s ^! - CM-CO co CM rvco " l i s in cb o T t CO^ LO P 6 CM CD CN ib CM Ci-co CO CD ^ CO oi 5 T t CO.LO co 4 p £^  iv CM f iCN CN CD LO CN J^CSI in co CO J o 2 SS <° •st o> o d CM or a: CD CO CO CO CM m csi •st CO rv cb •st ai CM o CN oo in co So" Ci.sT 3 CO O io jicsi o) p cb o G.CO CSI & •sr est I v co S2.iv rv •f sr C- CM p 5T rv cb « C M Tt ' cb.ib CM CO in Csi o CO rv Csi •st o cb £i d r^cN CO fS.'st CD CD CN : co cb i co cb" S CM a * CO csT !S CD O0 CN CO CO cb CM J^CM 2 CO Jo CO O JD •st j ub CD iri CM O co S3 cri CM, CN p 00 o O0 sr r-J co co.^ COO CM p CO S T U CN CN.ro O CD ^ o cb cb T t 2. S T t ; f v TF CO 001 co uo j cb P CO S" O) cb cb t— C CN CO CO o coo ib csi CM CD <o T t CD q T t 52, cb p T T sr co o in O0 co p CM <M, CM !S S" \t O) o CO Csi, f^' C O c q ? CD lb cb CO JM S> co rv io CJ) TT cb CM CM, co O CM csi CO CM ^ CO CO LO* CO S T T t JO 5 P CM P c\i oi s CM r^CM CO TT S" T t CM c-j T t CO.cb 00 rv CM CO o CO T f CM CD CO CD o Csi CM rv LO T t m T t o cb CM cb T t T t T t CO iri Csi tv cb p CO co p rv. Tt ' csi cb CO P. ib T— CM Sicsi CM EM. CM CM CO d> CM in cb co" CM rv' cb co cb d csi CD O) co CO S T CM T ~ m O) ^ CO rv o~ CO oo ^ oo in CO co rv csi CLcb CM Ccb CM O or O rv co C M cb co m S d CO CO n co C M m i " r ^ sr o rv rv C CM C M ! i CD c» CM S2. "o CO CM CO o> co co cb T", CSI o cb p P O CN co cb^: CO co" cb ^ co co co p ib CM CO CO P C7> cb J ^ co rv co* JO, T t CD co O o st CD ib C-CM CO Lo" s ° O CD o co ""- CO ficb OS Cv o O) CD ra I D T O DC o o of T3 CD Qi CD < 2 v - <u = o 2 3 5 o > &o CD 0) \s c S -3 (33 T3 t n <c U v f T3 — C 60 c<3 .ti OH _. CD « 'K m >^ ° S a O 1 £ CD CD> « -S •3 n CD p u CO c CD +-< o o CD CD IH O to O cd X ) W 6 CD 3 _ C *"> ~ <0 CD cd CD s s o o 8 e c <L> I s? is ° cd (D > a 47\ CD • O ^! c •* -c ^ cu D 2 CU 0-0 . co o co o z o 0-CN vr o CO i~-CD vr o o CD o d CD o o CO v t o o d v t o v t CT> o CD d v t LO o cn co o 00 |-^  CO d LO |v-d CM CD O O o 00 v t If) LO o d d o d v ? i O v t O d v f 5 O d S CM o co r» o d d CD d '• CD O O d d ~ |v-CD O d d " v t CO co S-V t CO CO m O d d w u. LO V f OO v t CM 00 O v t CO ro v t CD LO v t CO CM CM *oo o d 'oo 'CD v t 1.52 v t d 2.69" o 0.93" 0.493* 5.56 c CM LO o c r-. CM CD o d 81.4* 805.1 0.18" 0.03"' 0.00 0.08" 0.66 CM LO v t v t ro 00 CM v t L U CM |v- ro o CM O T— CM CN CO . — *— i — +i +i +i +i i +i -1 •ri • ; i |V- O CN q ro CD cn CD LO LO IV. LO vT CO LO LO CD ro oo CM o T CN CO o •<— T— ! i +i +i i +i i • l +i i o CD o CD CN CD LO LO r-: LO CO LO vt ^ _ CM LO CO CO vr CM vt T— T— CN +i +i i +i i +i +i i +i +i v t CM CD vr CO LO CM v t CO CO vr cq vr CD CO o v t ^- vr CN v t +1 +1 +1 , +1 +i +1 LO CO o v t CO CO CO CD CO v t LO vr co o v t vr CN vr i — x-+i 1 +i 1 1 • i +i LO CM CN CO CO CO CM vr LO CO CO LO T— CO T— T +i i 1 1 +i +i 1 i i LO LO LO CO CD vr co 00 CD o CO CO CO 1— CN v t o +i +i 1 +i i +i +i I +i 1 00 CO ro CD CD v t ro CO CO vr vr CO CD ro CD CO CM T— CN CO T— T— T— +i i +i 1 +i +i • +i +i co CD vr CO CO OO CM vt ro vr CM vr LO CM CO T— CM T T— +i +i i 1 i +i i i i o> CO 00 CM CO v t v t v t 00 LO LO CM CO Iv. CO CM CN vr T— O T— T— +i +i +i +i i +i +i +i +i i CO ro v t CD co v t ro v t o> LO CD vr vr vr o 00 vr v t »- vr CM CO i — CN +i +i i +i • +i +i i +i CO |v- ro ro CO ro CM CO CD CO CO vr CO v t CO CO vr CO T— +i 1 i i +i +i 1 i • +i v t CO ro CO IV. vr CO Iv. CM o 00 CN CO T— CM v t CO T T— T— +i +i i +i +i i +i • +i +i LO CO CO o |v. OO 00 LO LO ro LO vr vr LO LO CO o CO LO T CM v t T— .— T— i +i • +i 1 • • i +i i 00 co CO CD cn o CO LO Iv. v r vr vr T _ LO CN T— 1 +i i 1 1 i +i • i 1 LO CO CO LO 00 CO vr CO T — CN CO T — T — ^— +i +i i 1 +i --' +i • i -: v f O LO CO o LO CO v t CO vr vr vr CO CO 00 LO ro CO T - CN CO T— o +i t +i 1 +i +i i +i i +i v t co LO v t i-~ LO CM co CO co co vr f - CD LO i - -T— CM ^— o +i +i i 1 1 • i +i i i i CO O o CO CO ^ vr CO oo CO o CN oo co T— CN vt T— o T— +i +i i +i 1 +i +i i +i i CO co o •* LO CO LO ro cn LO CD vr CO CN I +i i 1 1 +i i ! ro LO LO v t o CM T— i +i i • 1 i +i i i 1 I-. Iv. LO —1 | 0. O o - J ca o o < 0. LU —j a. < LU — 5 0-o d CO (/) tu <" o <° ro £ > c cu £ "D CO •o CL o cu I S "55 .£ ^ c 6 > O) CO 8 CO o c E CT) cn o "cu 0) S -5 LO o ro Is 0 o _ <» 73 C O £ C C D •c o a . o o. • ro .9 0 O) • T3 O O a in = S? ° cu — o * o s in 2 •— [!> o B 3 ro <" o 3 I L W cr — LL. a g co CO ~ o S- D,LL ™ '«> !•= o 00 1 cu 0-LU CO O z o 0. u. Q d o r -o : CN LO CJ T _ o o o c IV. CD O : LO o o d d d o o vr c O O CD LO d CD o t o c O O d d d vr LO O CN CN o o o d d d o o o CM o o d LO d Sr O o c CJ) CO ! LO vr d d d o o o CD o CN o o o d d d c O o c vr vr c o d vt d c O o c co vt : CO LO d d d o o o CO t -o o d d d o o d vr CO o o d c O o CO i vr o d d CN c o o c * | v . d d vr o o r~ CO e Iv . Iv . d d d o o O) CO o o d vr d c O o o c 5) c O O O d d d 0.00 c LO d : LO LO d V. c co 'o CD LO LO CO CD CO o *CN O d d d LO O) CO CO v " ? Q: S i o o o o CO u. CO CD LO o d O) CJ) CN CD CO vr CO vr CN vr CJ) CN d vr co CN CM O d d " CN CN 2 " 1 1 O LO o o o d d ~ I i vr m T — m O d d ~" vr co o d '• 2 i c CO gg I*: in d in 2 ° O q 5 O O o J; - d - ° o o ' d ; O) co LO CN vr d CD CJ) CO CJ) CJ) to LO CD vr T CN CN . — T— . — -•i +i +i •: +i • CJ) r- CD r-- vr vr CN vr vr CN vr vr LO CO vr . CN . — T— • +i 'I • +i O CJ) CO CD CO vr CO vr CM +i CO v t LO r-CN T— 1 i +i i I +i 1— CO CO CO o CO CO CO CO vr CN CN CN T — T— T— +i +i +i +i +i +i CO CD vt vt o CM vt LO CN vr LO IV. CN 1 +i CN LO CJ) CD T— CN +i +i CO O CO LO CO CN 1 +i o LO CO | v - LO CD T— CD CN CN O T— T— +i +i +i +i +i +i CD CJ) CN CJ) LO CN LO co vr vt LO vr r-. | v . V t CO CN * — T— +i +i • +i i vr co CD o CO LO CO LO +1 CO vr +i CO CN +1 o CN +1 CO CO CO CN +i CO CO +1 IV. co | v -CN +i LO LO | v . CM +i CO LO 0-0 . +1 lO vr +l Iv . vr +i CD vr CD vr +i CO CO +1 CO CO +1 vr CO +1 o CO CD +i +i vr a. cc Ring Width For E K progenies, ring width (RW) had a highly significant site variance component. The family component was significant only at the RR site. For a particular ring, smaller genetic differences were expected for RW than for D, since genetic effects accumulate over several growth periods. SxF interaction was highly significant (Table 4.1.9). For the PG progenies, the site variance component was insignificant. Overall genetic variance estimated across sites was statistically significant, but heritablility was low. The interaction SxF was not significant (Table 4.1.10). Early-wood percentage For E K progenies for early-wood percentage (EW%), there was statistically significant site variation. Only at the RR was the site family variance component estimate significant, but heritability was low. There was no significant SxF interaction (Table 4.1.9). For PG progenies, there was no significant site or family variation, and no significant SxF interaction (Table 4.1.10). Transition-wood percentage For E K progenies for transition-wood percentage (TW%), there was no significant site variation. The family component estimated across sites was highly significant, and overall heritability estimate was moderate. Interaction variance component SxF was not significant. Moderate site-specific heritabilities were found at the RR and JC sites (Table 4.1.9). For PG progenies, a similar pattern of variation was found. There was no significant site variance. The family component estimated across sites was highly significant, and overall heritability estimate was moderate. Interaction variance component FxS was not significant. Moderate site-specific heritability was found at the Q site (Table 4.1.10). Late-wood percentage For E K progenies for late-wood percentage (LW%), there was no significant variation among sites. The family component across sites was highly significant, and the overall heritability estimate was moderate. Interaction variance component SxF was not significant. Site-specific heritability estimates were high at JC and PC sites (Table 4.1.9). For PG progenies, patterns of variation were similar. There was no significant variation among sites. The across-sites family component was highly significant, and 82 heritability estimates were moderate. The interaction variance component SxF was not significant (Table 4.1.10). Mork's index For E K progenies, Mork's index (MI) was influenced by site as a source of environmental variation. The overall family component was highly significant, but overall heritability was lower. Site-specific family variance was significant only at the PC site. The interaction variance component SxF was not significant (Table 4.1.9). For P G progenies, there were even more differences between MI and L W % . For MI, only the SxF variance component was statistically significant (Table 4.1.10). Number of cells per ring Cell number (CN) was highly correlated with the ring width, but its pattern of variation was somewhat different. The difference was in significance of site variance components for the two sets of progenies (Tables 4.1.9 and 4.1.10). Mean radial and tangential cell sizes For E K progenies, pattern of variation was similar for mean radial and tangential cell sizes (CSr and CSt). Site differences were pronounced, and variance due to site was significant. Both traits had significant overall family variation. SxF interaction was significant for both traits, and they appeared to be strongly inherited on RR and JC sites. Radial cell lumen sizes in early-wood (Lew), transition-wood (Ltw), and late-wood (Llw) followed the same pattern of variation (Table 4.1.9). For P G progenies, site differences were significant for each of two rings measured separately, but not for the average CSr. Site differences were also not significant for CSt. Overall family variation was significant in both rings for both traits. Heritabilities were generally moderate. SxF interactions were not significant. Radial cell lumen sizes in early-wood (Lew) and transition-wood (Ltw) followed a similar pattern, but in late-wood (Llw), only SxF variance component was significant across sites, and the family variance component was significant only at the RR site (Tables A5.2 and 4.1.10). Mean double wall thickness For both E K and P G progenies for mean double wall thickness (DW), the site variance component was not statistically significant. For E K progenies, the trait had a significant overall family component, but it had low overall heritability. SxF interaction was not 83 statistically significant, but site-specific heritability estimates were moderate at the E K sites combined, and high at the PC site. Double cell wall thickness in early-wood (DWew) and in transition-wood (DWtw) followed the same pattern, but in late-wood family variation was significant only for combined E K sites and the PC site {Table 4.1.9). For PG progenies, there was also no significant variation among sites. The family variance component was statistically significant only at the Q site. The interaction component SxF was not statistically significant. Double cell wall thickness in early-wood (DWew) followed a similar pattern, but in transition-wood (DWtw) and in late-wood (DWlw) none of the variance components were statistically significant (Table 4.1.10). Mean DW/CS ratio The ratio DW/CS (R) is the most closely related to the ring relative density. For E K progenies, there was highly significant site variation. The JC site which had the slowest growth rate had the highest ratio, i.e., highest relative density. The overall genetic variance component was highly significant, but overall heritability was low. No significant SxF interaction was found. The site-specific heritability estimate was high on the PC site. The ratio in early-wood (Rew), transition-wood (Rtw), and in late-wood (Rlw) followed that overall pattern, but for the latter site-specific heritability estimate was moderate at the RR site, and high at the JC site (Table 4.1.9). For PG progenies, there was no significant site variation. The overall family variance component estimate was not significant. Significant SxF interaction was detected. The site-specific family variance component estimate was significant at the RR site, but site-specific heritability was low. The ratio in early-wood (Rew) and transition-wood (Rtw) had the significant family variance component only at the Q site, with a moderate heritability estimate. In late-wood (Rlw) none of the variance components were statistically significant. (Table 4.1.10). 4.1.2.2 Fibre Length and Micro-fibril Angle Fibre length For E K progenies for fibre length (FL), variation among sites was highly significant. Overall family component was also highly significant, and overall heritability for FL was 84 moderate. Interaction variance component SxF was not significant. Site-specific heritability estimate was high at the JC site (Table 4.1.9). For PG progenies, there was no significant variation among sites for FL. The overall family variance component was highly significant, with a moderate heritability estimate. The interaction variance component SxF was not significant (Table 4.1.10). Micro-fibril angle Although microfibril angle (MFA) was not measured directly, it was assumed that its approximation was adequate. For E K progenies, variation among sites was not significant. The overall family component was highly significant, and overall heritability estimate was moderate. Interaction variance component SxF was not significant. Site-specific heritability estimates were high at E K and JC sites (Table 4.1.9). For PG progenies, there was no significant variation among sites. The overall variance component was highly significant, but overall heritability was low. Interaction variance component SxF was not significant. Site-specific heritabilities were moderate (Table 4.1.10). 4.1.2.3 Relationships among the Anatomical Traits Principal component analyses Principal component analysis (PCA) was used to reduce dimensionality of the anatomical data set. The percentage of variation explained individually and cumulatively by the principal components is shown in Table 4.1.11. Approximately half of the variation was accounted for by the first component and approximately two thirds of the variation by the first two components. Remaining components accounted for less than one third of the variation and were not considered. Results were similar for the E K and PG breeding zones. The first principal component (PCI) had negative correlations with traits related to growth (RW, CN) and cell size (CSr, CSt, Lew, Ltw, Llw) (Table 4.1.12). It had positive correlations with cell density (R, Rew, Rtw, Rlw) and ring latewood percentage (LW%, MI). This result implied negative linear relationship between growth rate and wood density components. The second principal component (PC2) had positive correlations with growth (RW, C N , CS), cell double wall thickness (DW, DWew, DWtw, DWlw) and fibre length in transition wood. It had a negative correlation with microfibril angle. This second result implied a positive linear relationship 85 between growth and cell dimensions including wall thickness. It also implied a reduction in M F A with increased growth rate, and cell dimensions. The third principal component was driven by C N , EW, and TW (with opposite signs for the two breeding zones). These relationships among anatomical traits were further studied using correlation analyses between pairs of traits. Phenotypic and genetic correlations As expected, genetic correlations were generally of the same sign as the phenotypic correlations, but higher in magnitude (Tables 4.1.13 and 4.1.14). (Exceptions were the genetic and phenotypic correlations between CS and R.) Interesting results from correlation analyses were those of RW with C N and of RW with CS. It appeared that ring width was more determined by the number of cells than by their average size. The result that RW was negatively correlated with R and all other traits positively correlated to wood density was expected. Remarkable also were the high correlations, both genetic and phenotypic, between RW and measures of percentage of cells in different classes: EW%, TW%, L W % . Positive genetic correlations of RW with CS, DW and FL, and negative genetic correlation between FL and M F A confirmed the results from PCA. More detailed analysis and detection of open-pollinated families with less negative correlation between growth traits and traits related to wood density may be rewarding. Desirable families may be the "correlation breakers" rather than just families with weak within-family correlation. Such an analysis is presented in the next section. 86 Table 4.1.11 Eigenvalues for correlation matrices for the EK (left) and PG (right) progeny tests. E K P G E i g e n v a l u e Proport ion Cumu la t i ve E i g e n v a l u e Proport ion Cumu la t i ve P C 1 11.188 0 .508 0 .509 9.800 0 .446 0 .447 P C 2 4 .737 0 .215 0.724 4 .500 0 .205 0.652 P C 3 2 .114 0 .096 0.820 2 .500 0 .115 0.767 P C 4 1.142 0.051 0.872 1.200 0 .056 0.824 P C 5 1.037 0 .047 0.919 1.000 0 .048 0.872 P C 6 0 .449 0.020 0 .939 0.700 0 .035 0 .907 P C 7 0.381 0 .017 0 .957 0 .600 0.031 0 .938 Table 4.1.12 Eigenvectors for the first four principal components for the EK (left) and PG (right) progeny tests. E K P G Traits P C 1 P C 2 P C 3 P C 4 P C 1 P C 2 P C 3 P C 4 R W -0 .195 0 .184 0.331 0 .309 -0 .217 0 .169 -0 .175 0.489 C N -0 .106 0 .073 0 .536 0.316 -0 .140 0 .077 -0 .328 0 .620 C S r - 0 .213 0 .260 -0 .215 0 .127 -0 .240 0 .230 0 .246 -0 .009 C S t -0 .136 0 .280 0 .068 0 .312 -0 .146 0 .215 0.101 0 .133 D W 0.216 0 .287 -0 .057 0 .093 0 .163 0 .370 0.184 0.032 R 0 .288 0 .080 0 .063 0.021 0 .293 0 .127 -0 .056 0.148 E W 0.134 -0 .043 -0 .467 0 .233 0 .108 • -0 .133 0 .455 0 .143 T W -0 .215 0.134 0 .283 -0 .394 -0 .160 0 .218 -0 .384 -0 .258 L W 0.199 -0.181 0.012 0 .423 0 .168 -0 .169 0 .248 0 .336 Ml 0 .287 0 .030 0.121 0.057 0.294 0 .096 -0 .037 0 .108 M Y 0 .260 - 0 . 065 0 .192 0.088 0.294 0.011 0.031 -0 .043 D W e w 0.164 0.362 0.035 -0 .032 0 .128 0 .388 -0 .016 -0 .100 L e w -0 .230 0 .199 -0.221 0.200 -0 .256 0 .155 0 .279 0.064 R e w 0.264 0 .140 0.190 -0 .099 0 .248 0.221 -0.171 -0 .108 D W t w 0.194 0 .295 0 .040 0 .013 0 .153 0 .355 0 .079 -0 .105 Ltw -0 .248 0.194 -0 .170 0.052 -0 .272 0 .175 0 .170 -0 .006 Rtw 0 .283 0 .045 0.131 0 .035 0 .290 0 .063 -0 .046 -0 .064 DWIw 0 .209 0 .272 -0 .194 0 .080 0.170 0 .248 0 .270 0 .147 L lw -0 .268 0 .050 0 .007 0 .160 -0 .262 0 .123 0.192 -0 .153 R lw 0 .237 0 .170 -0 .154 -0 .067 0 .248 0 .135 -0 .054 0.190 F L - 0 .033 0 .375 0.004 0 .022 -0 .122 0 .280 0 .099 -0 .023 M F A 0.041 -0 .327 0.004 0.436 0.071 -0.231 0.261 0.034 87 c CD > " 53b CD ca cn C o > C S ca L -H -o •c «= CN it, § i n 60 O S © cu 3 "ca > o o X) c — 0 "S 8 I 1 « 60 ° ^ .2 £ 8 o o 'a, 'a  c CD CL, a •tr £ . 5 MFA -0.07 -0.05 -0.06 -0.22 -0.18 -0.14 0.10 -0.14 0.13 0.00 -0.24 -0.03 -0.19 -0.18 -0.01 -0.08 -0.12 0.07 -0.17 -0.31 _ i L L -0.02 -0.03 0.02 0.14 0.05 0.06 0.09 0.02 -0.14 -0.05 0.10 0.00 0.08 0.06 0.00 0.07 0.04 -0.10 0.14 -1.03 Rlw -0.41 -0.32 -0.35 -0.17 0.67 0.87 0.35 -0.33 0.13 0.69 0.60 -0.45 0.78 0.56 -0.48 0.72 0.75 -0.80 - 0.63 -0.28 Llw 0.54 0.27 0.79 0.42 -0.22 -0.71 -0.39 0.36 -0.14 -0.57 -0.18 0.82 -0.64 -0.13 0.82 -0.62 -0.31 --0.74 0.23 -0.28 -0.13 -0.21 0.14 0.04 0.92 0.74 0.26 -0.34 0.27 0.67 0.78 0.01 0.64 0.81 -0.08 0.63 *--0.66 1.05 1.67 -0.54 Rtw -0.34 -0.21 -0.43 -0.23 0.73 0.94 0.09 -0.27 0.38 0.92 0.67 -0.55 0.91 0.69 -0.60 - 0.56 -1.19 0.74 0.15 0.38 Ltw 0.51 0.17 0.94 0.54 -0.03 -0.59 -0.28 0.37 -0.30 -0.54 0.03 0.93 -0.53 0.06 --0.96 0.02 0.83 -0.20 0.70 -0.71 0.01 -0.09 0.22 0.18 0.96 0.71 -0.15 -0.02 0.23 0.68 0.94 0.06 0.74 - 0.30 0.10 1.00 -0.43 0.89 1.67 -0.71 Rew -0.30 -0.17 -0.41 -0.18 0.76 0.92 -0.01 -0.11 0.20 0.84 0.79 -0.57 *- 0.68 -0.54 0.91 0.93 -1.01 0.92 0.71 -0.11 Lew 0.50 0.15 0.97 0.54 0.02 -0.53 -0.17 0.24 -0.20 -0.49 0.03 --0.64 0.40 0.99 -1.00 -0.10 0.87 -0.28 0.66 -0.51 5 s Q C D 0.01 -0.08 0.21 0.18 0.94 0.70 -0.14 0.06 0.06 0.61 - 0.30 0.54 1.07 0.43 0.06 0.96 -0.24 0.75 1.64 -0.67 I -0.37 -0.26 -0.40 -0.23 0.73 0.93 0.18 -0.43 0.58 - 0.01 -1.00 0.90 0.02 -0.98 0.97 0.48 -1.25 0.79 0.05 0.61 _ i -0.32 -0.27 -0.24 -0.22 0.29 0.43 0.28 -0.73 *- 0.33 -0.71 -0.60 -0.06 -0.88 -0.72 0.21 -0.78 -0.33 -0.35 -0.71 CO > v ? 0.49 0.45 0.27 0.30 -0.17 -0.36 -0.84 --0.97 -0.77 0.47 0.75 -0.31 0.44 0.86 -0.71 0.27 0.58 -0.04 0.47 -0.98 L U -0.44 -0.43 -0.20 -0.23 0.00 0.18 --0.96 0.87 1.16 -0.18 -0.85 0.66 0.06 -0.93 1.17 0.28 -0.80 0.45 -0.18 0.69 CL--0.39 -0.26 -0.43 -0.22 0.78 <-0.95 -0.54 0.11 0.98 0.38 -0.73 0.97 0.45 -0.68 0.94 0.74 -1.07 0.89 0.41 0.24 OW -0.06 -0.15 0.17 0.12 - 0.54 0.14 0.33 -0.76 0.15 1.03 0.12 0.74 1.01 0.24 0.21 0.97 -0.49 0.92 1.79 -0.54 CSt 0.49 0.32 0.55 - 0.41 -0.54 -0.85 0.79 -0.67 -0.76 0.58 0.75 -0.18 0.48 0.77 -0.61 0.14 0.74 -0.33 0.97 -0.43 cs 0.50 0.14 - 0.81 0.25 -0.66 -0.87 0.83 -0.74 -0.98 0.43 0.99 -0.53 0.32 1.00 -0.94 0.04 0.80 -0.15 0.87 -0.67 NO 0.92 0.13 0.73 0.39 0.00 -0.75 0.64 -0.49 0.10 0.50 0.02 0.41 0.50 0.14 0.21 0.17 0.15 -0.08 0.54 -0.71 RW 0.91 0.52 0.95 0.38 -0.30 -0.99 0.89 -0.73 -0.36 0.55 0.44 0.08 0.50 0.54 -0.25 0.12 0.46 -0.14 0.83 -0.91 RW NO SO est DW a. L U > VP —i 5 5 3 Q 0) Lew Rew Ltw Rtw 1 ^ Llw Rlw _ i L L MFA .2 .a 'so < •* CN o CO p SO CO o o CD O 0 O Tt 0 CO 0 00 0 0 Tt O CM O T t CO d O d d d o d d d d d d d d d d d d d _1 o CO O CM CM Tt CO CSI CO CO p m o o CM 3 CO CM CD T t CM T t CO CM LO Tt CD CO | v T— CO CO to CM 10 0 LL d d d d 9 d d d 9 9 d 9 9 d d d d d 3 CN CO CO CM CO CO o CO cn CO CO Cs| CO T t CM CO | v CM | v CO LO CO LO | v CO LO CM CO T t CO LO CO CC d 9 9 d d d d d d d p d d 9 d d 9 3 in CO CO LO r-. 0 •* T— CO LO CO cn CO IO CO 0 Iv CM T t CM CO 0 Iv 3 CO IO IV. T -IO CM P CO ao _j o d d o 9 9 d 9 O 9 d d 9 0 9 9 T _ p LO o s f o |v O CO cn T— CO o CO cn CM CM CM CO Iv. CO CO 0 CO •0 Iv 10 CO LO CO LO Iv Tt rv CD CO |v 0 P d d d d d 9 d d d 9 d d 9 0 d 1— •~ Rtw CSI CO d o CM 9 T— CO 9 CM CO d T t cn d L  o d Iv T -d IO CO d CO cn d CO rv d D CO d CM cn d M CO d CO Iv 9 0 d CD 00 d 00 CO d CO O s o IO CO CM T— Ol CM CO o IV IO CM o T t rv T t CD CO CO CM CO CD CO to to CM |v CD to CO 0 CD 0 •st CO ZJ o d d d d 9 d p d 9 d d d cp *7 , _ d Lo CO o o | v CT) CO CO Iv O o IO CM CO T t CD | v CM LO CO CO St CD CO CD CS| CO CM p CD IO p d d d d d d d 0 O d d d d 7 d Rew CM rv *— CM TT Iv CO T t CD CO o o o CM T t O) IV 0 rv to St CO LO i v CD 00 iv CO 00 CM O CD O Rew 9 9 d d d d d d d d d d d d •7 d •"-Lew T t LO T— CO CO o CO 0 CO IO to CM CO CO rv CO cn CM rv CO CO Tt CD Tt IV Tt CO CM 0 CM CO CD P Lew d d d d 9 d d d d d d d d d d •"- d TT o s t O CO o IO cn cn Iv o o o CO o CO rv 0 CM CO CO CM CM CO 0 CM •st CO CM O O p Tt CD Q OJ d p d d d d d d d d d T— d d "7 O d I o CO T— CO st s t CO IO cn LO o CM CM CO T t CO CM p CO CD Tt Tt rv 0 p CO CO co IV. CD CM d 1 9 o d d d d I d d •7 d d -7 •~ d *7 CO CO d • CO CM 9 3 9 CM d CO T t d •p. CO d IO IV 9 CM d 3 P 0 CM p CM CM d CD O d CO CO d 00 d p IV 0 d CD d d I-. co o CO CM CO CM o CO CO oo o CO CO rv CD LO Tt CD Tt Tt CO eo 0 LO 00 0 0 CO CO LO 0 ? 6 d d d d 9 -7 d d d d d d d d d d d LU CO CO o CO IO T— CO o rv CJ) |v St CO CM 10 IO Tt CO rv 00 3 |v IO CO 0 CO CO 0 0 rv CO 0 d 9 9 d d --- d •7 d d -7 d d d d -~ d DC IO CO CO CM CT) •st to CO o CM r-Tt CM CO 0 LO CD CN O Tt CO rv CD CM LO p p d 9 d d d d •~ d *7 d d 7 , ~ -7 7 d CO o IO o •st o Csi LO Tt rv CJ) ST rv 0 Tt rv CD 0 LO CO CM CD O CO Tt m CO 0 ° ° CO CO CO Q d d d d "7 d d d d d d ,— d d 7 d Csi CO CO io CM CO LO |v CD |v CO CN CO s 0 CM p 0 rv CM O LO CD |v IV 00 0 |v CSj CO CO o 6 d d 7 d d d 0 d d d d d • r - •"- d z LO cn CO CO | v CO S"t O |v CJ) CO CO CO CO m CM CO CM CO CO •st T— 10 CO CM CO LO CM Tt Tf co 0 O Tt rv co o o d d d d d d d d d O d d d T _ d d d g rv CD CO LO o CO CO CO CSI p Tt CO CM St rv Tt CD CM rv CO CD CM CO CM LO 0 T t 00 0 10 CM CD CM CD Tt s d d d d •"7 d d d 1 d d d d d d d d d 5 Oct z o CO O 5 Q CC UJ s ° _ i I 5 3 Q CP 3 cu _ i 3 cu ce _ l 3 1 ^ 3 _ i 3 CC _ i u. < LL ON 00 4.2 RELATIONSHIP BETWEEN G R O W T H R A T E AND W O O D DENSITY AND A N A T O M Y 4.2.1 Growth Rate and Wood Density: Overall and Ring-by-Ring Relationship Overall relationship Average height growth rate (AVH) and average ring width (AVRW) had negative linear relationships with average relative density (AVRD). The relationship between A V H and A V R D was less negative than that between A V R W and A V R D . Height growth, however, was not considered as a reliable trait to correlate with corresponding wood density because of the high incidence of weevil attacks reducing tree height by two years growth in some particular years. Significant differences in RD were detected among half-sib families, even when A V R W was included as a covariate. Phenotypic and genetic correlations at each site are presented in Table 4.2.1. There were differences among sites in the magnitude of phenotypic and genetic correlations. The results should be taken with caution, since standard errors of the estimates were high. Table 4.2.1 Genetic and phenotypic correlations of average relative density (AVRD) with average height (AVH) (top) and average ring width (AVRW) (bottom). Genetic correlation is given only for the sites at which genetic variances for both traits were significantly different from zero. Region and site abbreviations are explained in the text. Region EK PG Sites ALL RR JC PC ALL PG Q AVRD-AVH -0.46 n s±0.20 -0.36ns+0.20 _ -0.32" s±0.22 -0.26 n s±0.26 -0.41 n s±0.19 rg AVRD-AVH -0.68**±0.10 -0.54 n s±0.34 - ~ -0.85**±0.22 -0.54 n s±0.27 rP AVRD-AVH -0.28**±0.05 -0.19**+0.09 -0.39**±0.11 -0.30**±0.13 0.20**±0.06 -0.15**+0.10 -0.29**+0.08 AVRD-AVH -0.55"+0.05 -0.53"±0.08 -0.57"±0.10 -0.52**±0.10 - -0.42**±0.09 -0.61**±0.04 0.60**±0.04 Ring-by-ring relationship On an individual growth ring basis latewood percentage (LW%), a trait well correlated with relative density, showed generally negative phenotypic and genetic correlations with ring width (RW). The correlation coefficients were highly variable from year to year. But 90 cumulative increments showed a consistent pattern. The evolution of this relationship from pith to bark for cumulative increments is plotted in the Figure 4.2.1. Somewhat lower negative phenotypic correlation at early cambial ages may be due to the presence of the compression wood. The trend seemed to become more stable as the cambium matured. 91 17 18 19 20 21 22 23 24 25 Tree age Figure 4.2 .1 Pith to bark change of phenotypic (rp) and genetic (rg) correlations between average latewood percentage LW% and cumulative radial increments RW. Sites: RR and JC for the EK progenies (above), RR and Q for the PG progenies (below). Corresponding cambial ages were 7-15 (above) and 9-17 (below). 92 4.2.2 Growth Rate and Within-Ring Anatomy 4.2.2.1 Within Ring Distribution of Cell Wall Thickness and Cell Size The parameters of within-ring cell frequency distribution: mean (JJ), standard deviation (cr), skewness (co), and kurtosis (K) were first summarized across regions and sites for double wall thickness (DW), cell size (CS), and their ratio (R) (Table 4.2.2). DW was slightly skewed to the left (co slightly positive), CS was slightly skewed to the right (co slightly negative), resulting in R prominently skewed to the left (co highly positive). The statistics /u and a were mutually positively correlated, while co and /cwere highly positively correlated for R and DW, but not correlated for CS. The mean frequency in ten R classes for rings with approximately average and high growth rates are graphed in Figure 4.2.2. It was obvious that higher growth rate increases skewness to the left. The figure represents data for all sites and breeding zones combined, but the pattern for correlation of growth rate with the statistics describing within-ring frequency distribution for CS, DW, and R was generally similar across all sites (Table 4.2.2). (There were some minor differences between sites in the shapes of these distributions, i.e., on less productive sites with narrower rings cell frequency increased for approximately the same amount in three lowest classes.). A negative relationship between the RW and the standard deviation (SD) of R (or cell densities) was found interesting. It is obvious from Figure 4.2.2 that, with increased growth rate, there is an increase in low-density classes, but no significant increase in high-density classes. Wide rings had an increase in the number of earlywood cells, and less overall within-ring density variation, (i.e., more within-ring uniformity). 93 Table 4.2.2 Mean statistics describing within ring distribution of duble wall (DW), cell size (CS), and their ratio (R): mean (//), standard deviation(a), skewness(&>), kurtosis(/c) - top value; Correlation between ring width and those statistics - bottom value. Region and site abbreviations are explained in the text. Reg -ion Cell Size (CS) CS/DW Double wall Thickness (DW) Site RW M a CO K M C T CO K M cr CO AT EK RR 2.84 24.6 0.46 6.7 0.34 -0.23 0.02 0.16 0.03 0.22 -0.41 0.16 -0.44 2.28 0.36 6.05 0.37 5.51 -0.09 1.52 -0.26 0.91 0.06 0.46 0.79 JC 2.25 21.9 0.54 5.77 0.43 -0.12 0.06 0.05 0.12 0.26 -0.44 0.18 -0.44 1.92 0.39 3.92 0.36 4.86 -0.20 1.58 -0.27 0.93 0.23 0.68 0.27 PC 3.13 24.3 0.77 6.46 0.65 -0.22 -0.11 0.22 0.17 0.23 -0.58 0.16 -0.26 2.50 0.52 8.06 0.43 4.62 -0.31 1.47 -0.20 0.75 0.32 0.62 0.19 PG RR 1.53 22.1 0.70 6.35 0.43 -0.27 0.05 -0.15 0.29 0.27 -0.43 0.19 -0.43 2.06 0.48 4.88 0.46 4.92 -0.06 1.45 -0.04 0.71 0.24 -0.05 0.27 Q 2.96 24.7 0.59 6.47 0.40 -0.32 -0.08 0.28 0.75 0.25 -0.36 0.18 -0.31 1.93 0.32 4.07 0.28 4.81 -0.03 1.54 -0.19 0.84 0.10 0.37 0.22 PG R90 RR 2.79 21.8 0.53 5.69 0.51 -0.19 0.08 0.29 -0.00 0.25 -0.28 0.16 -0.12 2.17 0.24 5.86 0.18 4.72 -0.06 1.46 -0.05 0.64 0.10 0.19 0.70 Q 2.64 23.3 0.65 6.31 0.51 -0.42 -0.31 0.34 0.30 0.25 -0.55 0.17 -0.32 2.16 0.63 5.77 0.59 4.46 -0.27 1.30 -0.39 0.67 0.76 0.79 0.22 Note: Data are based on 240 data points for EKRR, 120 points for EKJC and EKPC, and on 240 ponts for both sites used to test PG progenies. 120 100 o c cu cr CD CD O Average RW Max RW Figure 4.2.2 Distributions of ratio of cell wall to size (R) for rings with approximately average ±10% and maximum 2 0 % growth rates. 94 After within-ring parameters were adjusted, using growth rate (RW) as a covariate in an analysis of covariance (ANCOVA), the relative contribution of family differences to the overall variation in distribution position and shape was tested for significance. Results of the analysis are given in the Table 4.2.3. Table 4.2.3 Analysis of covariance on mean statistics describing within ring distribution of of double wall thickness (DW), cell size (CS), and their ratio (R): mean (ju), standard deviation(cr), skewness(cy), kurtosis(Ar) with ring width (RW) as a covariate (Cov). Significance of F-test for Cov (top value), significance variance component for family intercept (a) (bottom value). None of the tests for slope differences were significant mostly due to insufficient DF. Region and site abbreviations explained in the text. Region Cell Size (CS) CS/DW Double wall Thickness (DW) Site So M a CO IC M a CO K M cr CO K EK RR Cov a ** ** ** ** ns ns ns ns ** ** ** ns ** ** ** * ns ns ns ns ns ns JC Cov a ** ** ** ns ns ns ns ns ** ** * ** ** ** ns * ns * ** ** ns ns PC Cov a ns ** ns ns ns * ns ** ** ** ns ** ** ** ns ** * ns ns ns PG RR Cov ** ** ns ** ** ** ** ** ns ns ** ** a ns ns ns ns ns ns ns * ns ns ns ns Q Cov a ** ns ** ns ns ns * ns ns ** ns ** ns ns * ns ** ns ns ns ns PG 20 RR Cov ** ** ns ns .. ** ns ns ns a * ns ns ns ns ns ns ns ns ns ns ns Q Cov a ns ** ns ns ns * ns ** ns ** ns ns ** ns ns ns ns ns Note: Data are based on 288 data points for EKRR, and 144 points for EKJC and EKPC (48 families). And on 240 points in PGRR, PGQ, PG20RR, PG20Q (40 families). At E K sites, family differences in mean ratio R were still significant after RW was accounted for. Family differences in R existed at all levels of RW. This was paralleled with family differences in both CS and DW. 4.2.2.2 Linear Prediction of Ratio of Cell Wall Thickness to Cell Size Multiple regression analyses (MAXR) were performed after mean ring R was partitioned into component traits according to Vargas-Hernandez and Adams (1991). At any given site only 95 three variables were sufficient to describe more than 9 5 % of variation in transformed R. The results of the M A X R regressions are presented in the Table 4.2 A Table 4.2.4 Results of multiple regression (MAXR) analyses for predicting average ring ratio R of double cell wall to size from different component traits. Late-wood percentage (LW%), double wall thickness in early-wood (DWew), cell size in early-wood (CSew), double wall thickness in transition-wood (DWtw), cell size in transition-wood (CStw), double wall thickness in late-wood (DWlw), cell Predictor V a r i a b l e s R 2 E K R R D W l w 0.551 C S e w , D W l w 0.860 L W % , D W e w , C S I w 0.957 E K J C C S I w 0.687 C S t w , D W t w 0.921 CSIw, DWtw, C S t w 0.967 E K P C C S I w 0 .692 DWtw, C S I w 0 .955 DWtw, C S t w , C S I w 0.966 P G R R D W t w 0 .705 C S t w , D W t w 0.932 CSIw, C S t w , D W t w 0 .965 P G Q C S I w 0 .615 DWtw, C S t w 0.916 L W % , D W e w , C S I w 0 .969 P G 2 0 R R D W t w 0.769 DWtw, C S I w 0.939 C S t w , DWtw, C S I w 0.964 P G 2 0 Q C S I w 0.706 DWtw, C S t w 0 .935 CSIw, DWtw, C S t w 0.960 Although there were some site specific relationships, variables DWtw, CStw, and CSIw were included in most of the trait combinations that best describe mean ring R. The trait CSIw was an overall good single predictor, among both E K and PG progenies (Q site). And DWtw was particularly a good single predictor among PG progenies at the RR site. 96 4.2.2.3 Avoiding Negative Genetic Correlation between Growth Rate and Wood Density Possibilities for overcoming the negative genetic correlation between RW and R were examined by including component traits in selection. Component traits were selected for instead of traits of interest directly. A selection intensity (z=l) was placed on component traits that had positive genetic correlation with either RW or R. Correlated responses in RW and R were compared with direct selection response in those traits. The estimated response was based on genetic parameter estimates and their dispersion as obtained by both jackknife and delta techniques. Confidence limits were obtained through Monte Carlo simulations with 1000 repetitions each. Only traits with statistically significant genetic variance were included in the calculations (Table 4.2.5). Table 4.2.5 Direct and correlated response in ring width (RW) and ratio of wall thickness to cell size (R) as proportion of the present mean and 95% confidence limits in brackets. Component traits used were average cell size (CS), transitionwood percentage (TW%), cell lumen size in earlywood (Lew), cell lumen size in transition-wood (Ltw), cell lumen size in latewood (Ltw), double wall thickness in earlywood (DWew), earlywood percentage (EW%) and latewood percentage (LW%) Data are for EK progenies at the RR site, and for PG progenies at the RR site. SELECTED TRAITS EK-RR RW, R CS, R TW%, R Lew, R Ltw, R Llw, R RW, DWe„ RW, EW% RW, LW% RW 0.252 (0.069 to 0.425) 0.110 (-0.113 to 0.413) 0.294 (0.125 to 0.442) 0.257 (0.105 to 0.438) 0.283 (0.022 to 0.462) 0.246 (0.013 to 0.482) 0.326 (0.009 to 0.496) 0.108 (-0.181 to 0.377) 0.116 (-0.148 to 0.313) R 0.067 (-0.045 to 0.160) -0.016 (-0.258 to 0.123) -0.041 (-0.124 to 0.028) -0.052 (-0.167 to 0.041) -0.065 (-0.196 to 0.090) -0.072 (-0.232 to 0.099) -0.061 (-0.153 to 0.005) 0.094 (-0.038 to 0.251) 0.002 (-0.097 to 0.099) PG-RR RW 0.126 (0.016 to 0.204) 0.047 (-0.065 to 0.124) -0.005 (-0.069 to 0.037) -0.019 (-0.089 to 0.057) 0.090 (0.026 to 0.149) R 0.074 (-0.294 to 0.416) -0.047 (-0.320 to 0.381) 0.018 (-0.222 to 0.227) 0.048 (-0.222 to 0.388) 0.052 (-0. 198 to 0.251) It was possible to obtain positive response in both RW and R. The relative importance of component traits were evaluated. Only among EK progenies at the RR site was indirect selection for RW and E W % estimated to be more efficient than direct selection for RW and R. 9 7 This would depend on relative values (economic weights) for the latter traits. More results regarding opportunities for multiple trait improvement are given in the next section. 4.3 OPTIMIZATION OF SELECTION 4.3.1 Optimization of Single Objective Functions Results of the optimization process, which gives the maximum response in a non-linear objective function based on a linear selection index under a given selection intensity, are presented below. Maximisation was done using the maximum possible genetic response as constraint. 4.3.1.1 Volume, Dry Weight, and Wood Density Expected genetic response (A) after one generation of truncation selection (i=2) was calculated based on multiple traits for the following selection objectives. Separate selection indices that maximize response in volume ( I V O L ) , that maximize response in volume and place restrictions on changes in relative density (IVOL_RD), and that maximize response in dry-weight (IDWT) were used. The expected genetic response was presented both as an absolute value and as percentage of the present mean in the base population. Selection was simulated among the progenies at sites were all traits of interest had significant family variance, i.e., E K progenies at RR site, and PG progenies at Q site (Table 4.3.1). It can be seen that significant improvement could be achieved by selection for volume alone, using the index I VOL- Taking DWT for the selection objective in IDWT did not bring significant improvement. The achievable change in RD is negligible relative to the achievable change in V O L , considering improvement of their joint function DWT. For the E K progeny at RR, the same sets of parents were obtained from selection using I V O L and IDWT at the selection intensity i=2. For the PG progeny at Q, different sets of parents were obtained from selection using I V O L and IDWT (Table A6.1). Constraining change of RD to zero, using the index IVOL_RD, leads to a significant reduction in volume improvement compared to IVOL- Since there was generally a negative genetic relationship between growth rate and wood density in spruce, it may desirable to breed for high V O L and keep RD constant, depending on the desired end product. When 98 restricted selection was applied by using the index V O L _ R D , there was approximately 50% less expected gain in both V O L and DWT, in PG progenies at the Q site, relative to selection without the constraint I VOL- However, the reduction was considerably smaller less than 30% in the E K progeny at the RR site {Table 4.3.1). Optimization using linear approximations To test i f simplifying the optimization procedure was possible, linear approximations based on Taylor series of objective functions V O L and DWT were used. Index weights for IVOL and IDWT changed, resulting in different sets of parents being selected. The differences in sets of selected parents increased with increasing selection intensity. The simplified selection using linear approximations was also less sensitive to small changes in parameters of objective functions, such as those included in calculation of stem volume (Table A6.1). Sensitivity analyses Changes in correlated response depends on multiple factors, some of which were the optimization parameters and were examined in the sensitivity analyses. Critical factors affecting the selection outcomes were examined by varying the optimization input parameters, including trait means, variance-covarince matrices (G), or site by genotype interaction. Reduction in magnitude of all elements of the matrix G by one standard deviation, at the E K R R site for example, lead to a small change in the expected genetic response in V O L (from 0.69% to 0.66% ) and DWT (from 0.60% to 0.54%). Genetic covariances of H and D with R D were negative, while environmental covariances were positive. And reduction in G also reduced the negative synergism of the traits. 99 Table 4.3.1 Genetic response (A) after one generation of truncation selection (i=2) based on multiple trait selection. Selection indices: I VOL that maximizes response in volume, IVOL.RD that maximizes response in volume and places restriction on change in relative density, and I DWT that maximizes response in dry-weight. Genetic responses as a percentage of the present mean are given in brackets. Index Optimal Weighting Expected genetic responses H D RD AH (m) AD (cm) zlRD zlVOL (dm3) zlDWT (kg) EK-RR 1 0.887 1.00 1.17 1.63 -0.006 10.29 3.64* •VOL — (22.7%) (18.8%) (-1.5%) (69.4%) (66.8%) IvOL.RD 0.192 0.216 1.00 0.95 (18.4%) 1.16 (13.4%) 0.000 (0.0%) 7.40 (49.9%) 2.73 (49.9%) 'DWT 0.558 0.628 1.00 1.17 (22.9%) 1.59 (18.4%) -0.021 (-5.6%) 10.32 (69.6%) 3.28 (60.1%) PG -Q • 0.974 1.00 1.05 1.04 0.001 8.382 2.93* IVOL — (21.1%) (10.6%) (0.21%) (46.3%) (46.6%) 'VOL.RD -1.405 1.050 -0.000 0.75 (15.1%) 0.25 (2.6%) 0.000 (0.0%) 3.773 (20.9%) 1.31 (20.9%) 'DWT -1.405 1.050 1.00 1.05 (21.1%) 1.02 (10.4%) 0.006 (1.83%) 8.304 (45.9%) 2.71 (43.3%) T h e re sponse e x c e e d s the one from se l ec t ion that targets D W T due to s a m p l i n g errors 4.3.1.2 Volume Growth, Dry-weight, and Pulp and Paper Properties The formulae that describe pulp and paper properties were based on fibre characteristics and relationships were examined within-ring. Anatomical data for the ring of cambial age 14 at the (EK)RR, and two rings combined, of cambial ages 12 and 16, at the (PG)RR site were used. Correlated genetic response (A) in pulp and paper properties after one generation of truncation selection (i=2) is presented in the Table 4.3.2. Selection was based on the previously defined indices IVOL, IVOL_RD, and IDWT- Expected genetic responses were calculated as absolute values or as percentages of the present population mean. Pulp and paper traits considered were tensile strength of wet-webs (T T O ) , tensile strength of paper (Tp), tear of weakly (TRi) and well bonded (TR2) paper, burst of weakly (BRi) and well bonded (BR2) paper, and tensile strength of mechanical pulp (Tm). 100 Selection on IVOL resulted in a positive correlated response in T^, and T m , and T p was also positively influenced, but to a lesser degree. It is unclear why tear strength TR- was negatively correlated with TR2, and positively correlated T w , T p , and T m . Seth and Page (1988) obtained a similar result and concluded that empirical formulas, such as Clark's (1985), have only limited validity and should be considered with caution. There was no obvious response in burst strength of paper. Tearing resistance of paper was the only quality that decreased with selection for increased volume growth. Selection on I D W T resulted in predicted changes generally similar to those from selection for V O L , but the estimated responses differed slightly in magnitude. For selection on IVOL_RD there were differences in the expected response between the two sets of progenies. Negative response in Tp, BF-, and TR2 was obtained for E K progenies, while response in those paper properties was positive for PG progenies. 101 Table 4.3.2 Correlated genetic response (A) in pulp and paper properties from one generation of truncation selection (/—2). Selection was based on three indices: I V O L that maximizes response in volume (VOL), IVOL.RD that maximizes response in volume and places restriction on change in RD, and IDWT that maximizes response in dry-weight (DWT).Genetic responses as a percentage of the present mean are given in brackets. Pulp and paper traits are tensile of wet-webs ( T w ) , tensile of paper (Tp), tear of weakly (TR-) and well bonded (TR 2) paper, burst of weakly (BR-) and well bonded (BR 2) paper, and tensile strength of mechanical pulp (Tm). A) Selection among E K progenies at RR site, B) Index Cor re la ted genet ic r e s p o n s e s (Nm/g) AJP (Nm/g) zlTR, zlTR2 ABFZ Alm E K - R R 'VOL 0.017 (2.3%) 0.45 (0.7%) (-1.2%) (0.6%) (0.0%) (0.0%) (2.3%) IvOL.RD -0.015 (-2.1%) -0.41 (-0.6%) (2.3%) (-1.1%) (-2.3 %) (-0.0%) (-2.1%) IDWT 0.048 (6.8%) 1.23 (2.0%) (-4.6 %) (2.3%) (4.8%) (-0.1%) (6.8%) PG-R R Q 'VOL 0.043 (6.9%) 1.29 (2.2%) (-8.2%) (6.1%) (12.6%) (1.2%) (6,8%) 'vOL.RD 0.006 (-0.8%) 0.53 (0.8%) (-3.9%) (4.2%) (6.6%) (0.8%) (-0.8%) IDWT 0.050 (7.9%) 1.50 (2.5%) (-6.9%) (3.6%) (6.9%) (0.4%) (7.9%) Sensitivity analyses Beside the above mentioned critical factors affecting selection outcomes, changes in objective function coefficients can also be considered. For example, when a change in sheet structure, i.e., relative bonded area (RBA) was considered, there was change in the expected selection response in tensile strength of paper (Tp). While R B A is altered upon pulp treatment, such as beating or wet pressing, bond strength (b) stays unchanged, according to Page (1969). The same author found a linear relationship between 1/Tp and 1/RBA. Here it was found that the relationship between zlRBA and zlTp was also approximately linear over a range of R B A values, varying from zlTp =2.2% at 0.5 R B A to ZlTp =1.6% at 0.9 R B A . 102 Similarly, other parameters in the semi-empirical formulae that relate fibre to properties of pulp and paper can be varied accordingly to the type of the end-product we are interested in, and the expected response to selection can be monitored. 4.3.2 Multiple Objective Functions Direct and correlated responses after selection optimised for single objective functions were presented above. If more than one selection objective is optimised at the same time, multiple objective optimization of selection is needed. 4.3.2.1 Non-inferior Solutions and Sensitivity Analyses Comparison of simultaneous response in different objective functions is obtained for different scenarios by using the NIMBUS multi-objective programming system. The data used are from rings 20 and 25 combined for the PG progenies. The ideal criterion vector (ICV) and the worst case (Nadir) vectors are given, together with the trade-off trend within the solution space. Solution space includes all Pareto optimal or non-inferior solutions. Points of maximum relative improvement in either trait (MaxiMax), maximized average value (MaxiAvr), maximized minimum improvement (MaxiMin), minimised maximum loss (MiniMax) are also presented, assuming that all objective functions have an equal value. The selection scenarios were the following: Volume and tensile strength of pulp wet web The ICV and Nadir vectors representing the expected genetic change as a percentage of the present population mean at selection intensities of i=2 and z'=l, respectively, were as follows: Function ICV/'=2 NADIR i=2 ICV/=1 NADIR/=1 V O L 0.685 0.614 0.315 0.289 T w w 0.087 0.069 0.039 0.033 The trade-off within the Pareto-optimal solution space can be visualized from the Figure 4.3.1 103 0.09 0.085 | 0.08 0.075 0.07 0.58 0.6 0.62 0.64 0.66 0.68 0.7 ^ V O L Figure 4.3.1 The approximate nature of trade off between improvement in volume (zlVOL) growth and simultaneous response in tensile strength of produced pulp wet web (zlTww), based on multiple objective optimization. • MaxiMax, • MiniMax, • MaxAvr, and • MaxiMin solutions. The improvement expressed as proportion of the present population mean, at the selection intensity of z-2. It can be seen that positive responses could be obtained in both traits, but a significant trade off existed across the optimal solution space, with a pronounced curvilinear relationship. The range of improvement, where the trade off existed, was larger for V O L , approximately 7%, and smaller for Tww, less than 2%. The trade off became increasingly stronger at high values of A\OL. Points for MaxiMax, MiniMax, MaxiAvr, and MaxiMin solutions were separated, especially on the zlVOL axis. However, there was no difference in the sets of parents selected based on the index weights derived for those points (Table A6.2). Volume and tensile strength of paper The ICV and Nadir for V O L and Tp, as a proportion of the present population mean for selection intensity of i=2 and i=\, respectively, were as follows: Funct ion I C V / = 2 N A D I R i=2 I C V ;=1 N A D I R ;'=1 V O L 0 .672 0.546 0 .319 0 .315 T p 0 .028 0.022 0.011 0.011 The approximate trade-off within the Pareto-optimal solution space can be visualized from the Figure 4.3.2. 104 0.028 0.027 0.026 ^ 0.025 0.024 0.023 0.022 0.55 0.6 0.65 0.7 ^ V O L Figure 4.3.2 The approximate trade off between improvement in volume (zlVOL) growth and simultaneous response in tensile strength of produced paper (^ ITp) based on multiple objective optimization. • MaxiMax, MiniMax, MaxAvr (approximately same), and • MaxiMin solutions. The improvement expressed as proportion of the present population mean, at the selection intensity of i=2. Again, positive responses can be obtained for both traits. A trade off exists, but its magnitude is insignificant. The trade off again becomes increasingly stronger towards high values of zlVOL. MaxiMax, MiniMax, and MaxAvr solutions were all approximately the same. Points of MaxiMax, and MaxiMin solutions were separated mostly on the zTVOL axis, but there was no difference in the sets of parents selected based on the index weights derived for the extreme points (Table A6 .2 ) . Volume and tear strength of paper A significant trade-off was necessary for simultaneously improving both volume and tear strength of paper. The ICV and Nadir vectors, respectively, were: Funct ion I C V ;'=2 N A D I R i=2 ICV/=1 N A D I R /=1 V O L 0.685 -0.460 0.315 -0.249 TR1 0.092 -0.082 0.039 -0.040 This relationship can also visualized on the following graph (Figure 4.3.3 ). 105 0.081 -0.08 J Figure 4.3.3 The approximate trade off between improvement in volume (zlVOL) growth and simultaneous response in tear strength of produced paper (zlTRl), based on multiple objective optimization. • MaxiMax, MiniMax, MaxAvr (approximately same), and • MaxiMin solutions. The improvement is expressed as a proportion of the present population mean, at selection intensity i=2. It can be seen that there was approximately linear decrease in tear strength of paper with the improvement in volume of wood. The range of improvement in relation to the present population mean is again much wider for V O L . Allocation of objectives according to some particular criteria, or different risk-reduction strategies, was examined more closely. Choosing different selection objectives would give different sets of selected parents. Assumptions could be made as to what would be the future value of objective functions. The points of MaxiMax, MiniMax, and MaxAvr solutions were all approximately the same, i f equal value was assumed for TRi and V O L functions. Only when significantly higher weight was placed on TRi , were different solutions obtained. The most conservative MaxiMin solution was separated on both axes from the other solutions, and it would require a trade-off with V O L , which was greater in magnitude (approximately 10 times). There were differences between the sets of parents selected based on the two extreme points (Table A6.2). Opting for a conservative option, such as MaxiMin, would therefore pay off only if 1% improvement in TR] corresponds in the future value to approximately 10% improvement in V O L , assuming additive utility of the two value functions. 106 Wood Density and Pulp and Paper properties There was significant reduction in expected genetic response of some pulp and paper traits when ratio of double all thickness to cell size (R) (as a surrogate for wood density) was introduced and held constant, while selecting for high volume growth (Table 4.3.2). The trade-off within the solution space was most pronounced for tensile strength of pulp wet webs. The ICV and Nadir vectors representing the genetic change as a percentage of the present population mean at selection intensities of i=2 and i=l, respectively, were as follows: Function ICV/'=2 NADIR ;'=2 ICV/=1 NADIR /'=1 V O L R 0.549 -0.082 0.260 -0.042 T~ 0.048 -0.008 0.024 -0.004 A WW The trade-off within the solution space, obtained by the multiobjective optimization at the selection intensity i=2, can be visualized from the Figure 4.3.4. ^e_ei-j ^ V O L _ R Figure 4.3.4 The approximate trade-off line between improvement in volume growth with density (R) held constant (zlVOLR) and simultaneous response in tensile strength of produced pulp wet web (zlTww) based on multiple objective optimization. • MaxiMax, MiniMax, MaxAvr (approximately same), and • MaxiMin solutions. The improvement is expressed as proportion of the present population mean at the selection intensity of i=2. It can be seen that by restricting the change in R and selecting for V O L the solution space has been shifted and extended towards the negative response in V O L and Tww, in comparison to selection without constraint on R. MaxiMax, MiniMax, and MaxAvr solutions were all approximately the same. Points of MaxiMax, and MaxiMin solutions were separated on both 107 axes, and there were differences between the sets of parents selected based on the index weights derived for the two extreme points (Table A6 .2 ) . Volume, tensile strength of pulp wet web, and tear strength of paper For simultaneous change in these three objectives the ideal criterion vector ICV and worst-case Nadir solutions for selection intensities of i=2 and /=1 respectively, were the following: Function ICV i=2 Nadir i=2 ICV/'=1 Nadir ;=1 VOL 0.685 -0.460 0.315 -0.249 Tww 0.087 -0.047 0.039 -0.038 TR1 0.092 -0.082 0.039 -0.041 Multiple intermediate solutions were generated for the range between alternative function maxima by decreasing the previously maximized functions. The solutions represented the range of necessary trade-offs between improvement in different objective functions, depending on how much emphasis was placed on each particular objective. For improving the three traits V O L , Tww, and TR-, simultaneously within one breeding population, multiple trade-offs were necessary. Although positive response was possible in all three functions at the same time, there was only a narrow range of values that gave positive responses in all three traits (Figure 4.3.5). At-the point where Tww and TR- intersected (MaxiMin solution), the V O L function was at less than a quarter of the total achievable improvement. Intersects of the function V O L and the two functions related to pulp and paper properties were relatively close to the point of zero improvement. MaxiMax, MiniMax, and MaxAvr solutions were all approximately the same when selection intensity i=2. Points of MaxiMin solution was separated, however, and there were differences between the sets of parents selected based on the index weights derived for the two extreme points (Table A6 .2 ) . Again it is necessary to place significantly higher weight on Tww and/or TR- before opting for some conservative option, such as MaxMin. This may seem unrealistic, especially regarding TR- for which 1 % improvement over present mean would have to be equal to approximately 10% improvement in V O L . A more realistic approach would perhaps be with 0% change in TR, and with 30% and 5% improvement in V O L and Tww respectively. The value of such a population, however, will be only approximately 50% of the MaxiMax solution, if there is no knowledge about the future worth of any of these potential objectives. 108 cn I -• 5 O > •o O > c o C O 15 E -4—• Q. O o CO D_ CD CD T3 CD E L _ O > CO c o m N i n c o i n i n u i ^ i o c o i n c M U ) t o c N i o c o i n ' ^ r i o u o i o c D BS1AI )U3S8Jd p u o i y o d o j d •3 C3 c _o o '0 M C N P O . cn ? o « P U T J 8 § ? & 3^ J3 *-3 2 u 0 S > > c £ 5 ° B — TJ g 6 i a s - ° P 9 b 1 i 2 a, H B 13 C e % • a 2 c2 0 C JS, £ Q) cn C o ! u 1 g "2 w cd cn S C cn •H CU C t u cn cn 3 2 o p . a cn -a ° § 3 " CU 60 o s CU N 0 ;~ cn Ci - i 1 ^ s •-IS cn 1.1 • C I I O w C+H o t u IT) CU 3 D, <*5 'ft £ -3 -4—» 1 S I P •a & 1 s 2 § o D-O O The simultaneous change in the functions can be also visualized relative to the maximum possible improvement in the following petal diagrams were the relative losses in the three functions are presented. A) P E T A L D I A G R A M O T T H E A L T E R N A T I V E S Figure 4.3.6 Selection alternatives and trade-offs followed from an intermediate solution to the maximum improvement in A) volume (VOL) growth, B) tensile strength of pulp wet web (Tww), and C) in tear strength of produced paper (TR1). Alternatives are Pareto-optimal solutions obtained by multiobjective optimization. Losses in objective functions are expressed relative to Nadir (worst case) solutions and represented by different colours. If the values of objective functions were known or could have been better estimated, further optimization could have been obtained through iteration process using the multiobjective method. Here a range of alternatives is presented and choosing between alternatives by classification of functions would depend on the decision-maker's preferences (Figures 4.3.5 and 4.3.6). Sensitivity between the two points of interest could be checked by simply inserting the number of new alternatives. 110 4.3.3 Multiple Breeding Populations Under uncertainty about future value of objective functions, introduction of additional breeding populations was considered. Multiple populations would make it possible to cover a wider range of future options and include some more risky alternatives. General solutions to the problem of optimal choice of population, assuming normal probabilities, was given by Namkoong (1976) and Roberds and Namkoong (1989). For example, it could be assumed that some future optimal economic weights on TR-relative to V O L would have discrete probabilities as presented in the Figure 4.3.7. For single population breeding, the population choice A would be optimal. This choice would have, however, an associated expected loss represented by the sum of probabilities of other possible outcomes (or choices) multiplied by their distance from A on the horizontal axis. If there is a possibility to have two populations, then populations at B and C would be the ones that reduce the probability of expected future losses the most, i.e., by 40%. A 0.1875 p 0.125 0.0625 B 4 < 1 • 1 4 C k 9 f i F A A G o 4 4.5 5 5.5 6 6.5 7 7.5 8 Weight on TR1 Figure 4.3.7 Economic weights on objective TR| relative to objective V O L , their probabilities, and population choices A - G . The higher the uncertainty about future values of objective functions, i.e., the range on the horizontal axis in Figure 4.3.7, the more divergent the optimal populations would be. For two optimal divergent populations (B and C) two selection indices could be derived with two different economic weights on TRi . Aggregated expected value of such a set of populations would always have an expected value greater than an optimal single population. In the chosen i l l example, however, because of the much higher genetic response in V O L , values for the two optimal populations did not yield different sets of parents. Only when a much wider range for expected economic weights was considered, would such differences occurred. 112 V DISCUSSION 5.1 HERITABLE VARIATION IN GROWTH, M A C R O WOOD PROPERTIES AND TRACHEID CHARACTERISTICS 5.1.1 Heritable Variation in Tree Growth and Macro Wood Properties Patterns of variation were examined for cumulative growth as height (H) and diameter (D). The portion of accumulated growth with cambial ages from 12 to 17 was examined separately. It is assumed that transition to mature wood usually occurs in spruce at cambial ages greater than approximately 17 years (Zobel and Sprague 1998, Yang and Hazenberg 1994). Variation in those growth rings in older juvenile wood portion was examined as average ring width (AVRW), average latewood percentage (AVLW%), and average relative density (AVRD). A detailed discussion of age trends and relationship between growth rate and wood traits follows in the section 5.1.1.5. The outer portion of accumulated growth was of importance because it represented the larger portion of volume of the examined trees. 5.1.1.1 Height Growth There was more variation in H among sites for the E K progeny tests, probably because the test sites were more diverse, covering a much larger area than sites for the PG tests. Site differences did not appear statistically significant in the PG tests, despite different site preparation treatments, which had a major influence on the early height growth (Kiss and Yeh 1988, Yanchuk and Kiss 1993). Significant differences among families resulted in high heritability estimates for height, which in turn led to predictions of a good response to either individual (mass) or family selection. However, high heritabilities at the earlier age decrease gradually with age in 113 shade tolerant species such as spruce, when planted in open field (Kiss and Yeh 1988, Dr. John King personal communication1). The magnitude of differences in H among sites, and among families, was significantly influenced by weevil damage. Occasionally, gall aphid (Pineus and Adelges spp.) damage, perhaps in combination with frost damage, can contribute to stunting and deformity of young spruce trees, and damage can be locally severe (Coates et al. 1994). The frequency of attacks also varied among blocks within plantation sites, which reflects local concentrations of pest activity. Mechanisms of resistance to weevil attacks are not fully understood. The genetic basis of resistance was analysed quantitatively by Kiss and Yanchuk (1991). High family heritablility (h2f= 0.11+ 0.11), and only moderate individual heritability (h2,= 0.18 ± 0.03) was found for weevil attacks. They also suggest that families more vigorous at age 10 were less susceptible to initial weevil attacks. However, the opposite phenotypic correlation between tree height and incidence of initial weevil attacks was reported, for young Sitka-spruce plantations (Xu 1998). The influence of family effects on the severity of attacks, and influence of those attacks height growth, is a complex question (King et al. 1997). The available data on weevil damage from the BC Ministry of Forests score attacks in tree categories: no, initial, and recurrent attacks. The damage score is a categorical variable, but it can be considered to have an underlying normally distributed variate of genetic effects (Lynch and Walsh 1997). Score transformations that have a joint distribution with the underlying normal variable of genetic effects have been developed (Gianola and Norton 1981). Categorical responses translated into the normal score maximize heritability and accuracy of its estimation. If more than three categories were used, transformed scores could be considered a continuous co-variable, and used for estimating the relative contribution of weevil resistance to differences in height, among families. Since existing data do not lend themselves to an analysis of covariance, height growth is confounded with weevil damage. It is known that weevil attacks affect bud phenology in spruce (Seaby and Mowat 1993). One could perhaps ask some more specific questions related to the topic of this thesis: How does the pattern of height, or volume growth, i f affected by 1 Research Sciencist, BC Ministry of Forests, Victoria. 114 pest attacks, eventually reflect on wood anatomy? Can simultaneous breeding for height growth, insect resistance, and wood quality be useful? These problems are, however, beyond the scope of the present study. 5.1.1.2 Radial Growth Perhaps radial growth is not influenced by the weevil attacks to the same extent as height growth. Heritability estimates for D were generally lower than for H. Among PG progenies, the family variance component for D was not statistically significant, while for H it was significant. The same result was reported by Yanchuk and Kiss (1993). Interestingly, partition of variation according to sources for A V R W did not always follow the pattern of D. This might be merely a consequence of using ring widths instead of their basal areas in the analyses (see section 3.2.1.2), but could also mean that more mature wood production followed different patterns of regulation compared to the total radial growth. For the PG sites, e.g., there were no significant site differences in D, but a highly significant site difference for A W R V occurred. RR site produced narrower growth rings in the last three growing seasons than the Q site, which can be considered a sign of maturing of wood. If the maturation of cambium is taken as a major factor, it seems that it occurred earlier on the RR site. Another possible factor is different climates on the two sites. Effects of combinations of these two major factors are discussed in more detail later in the section 5.1.1.5. 5.1.1.3 Wood Macro Properties From a practical standpoint, A V L W % is relatively easy to measure using photometry. It correlates well with wood density and other properties, and it can be used to draw some conclusions regarding the cambial activity. Therefore, it is an important trait to be used for evaluation of progeny tests (Larson 1969). The trait showed more phenotypic and genetic variation among the E K than among the PG progenies. The locations of tests including the E K progenies are both in the E K and PG region, latter being outside the region of origin of parent trees. Test sites for the PG progenies are only in the PG region. The two regions have different climates. The PG region is located more than 3° latitude north from the E K region. It generally receives more precipitation per year (614mm vs. 384mm). It has lower annual mean daily temperatures (3.7 115 vs. 5.6), and lower number of degree-days with temperature above 5°C (1641 vs. 1238) (Environment Canada 2000). The E K family differences in A V L W % were especially pronounced on the site outside the region of origin, implying possible changes in their wood structure as consequence of the transfer. Within-region variation in site conditions could have also played a role in the expression of family differences, since some sites exhibited significant family variance, while other did not. Although for black spruce from eastern Canada latewood percentage had generally low heritability (Zhang and Morgenstern 1995). Since A V L W % is correlated with relative density, i f considered together with A V R D , it may provide more information about the performance of a particular family, especially about the relationships among growth, phenology, and resulting wood quality. For example, duration of cambial activity may be similar for different genotypes, but the ones with slow, prolonged, terminal growth would be expected to produce more earlywood. The others with more rapid terminal growth would produce more latewood (Worrall 1970). Although the test sites were more diverse in the EK, than in the PG progeny tests, only the latter exhibited significant S variation for A V R D . Contrary to this finding, Yanchuk and Kiss (1993) reported no site variation in this trait for 40 PG families evaluated at age 15. This could be explained by considering the fact that A V R D was strongly related to radial growth rate, and there was a significant difference between PG sites in the radial growth rate during the most recent years (age 22-25), for which A V R D was measured. When A V R W was included as a covariate the site effect disappeared. At the same time, family variance was statistically significant only for A V R D and not for A V R W . 5.1.1.4 Relationships among Traits For multi trait selection, single trait heritabilities have only limited use for predicting selection response. Multiple traits are usually inter-correleted, and after including various traits as covariates, the overall picture usually changes. For example, heritability for wood density may not have much influence if density is negatively correlated with growth rate, and the trait of interest is total dry weight. When genetic gain is predicted based on an index, the relative economic weights of traits depend on the targeted value function. Estimates of variance and covariance are needed as well. Nevertheless, phenotypic correlations are more dependent on genetic correlations, i f heritabilities are high, then on environmental correlations. Otherwise, 116 environmental correlations are the determining factor. In other words, phenotypic and genetic correlations between traits are similar i f heritabilities of the two traits are high (Falconer 1989, Roff 1997, Koots and Gibson 1996). Differences among populations were observed in the extent to which traits were genetically correlated. For the PG progenies, a low genetic correlation between growth characteristics and wood density was found. This finding is in agreement with the results reported by Yanchuk and Kiss (1993), who reported virtually no genetic correlation. For the E K progeny, however, a significant negative correlation was found in this study. There are genetic differences between the E K and PG progenies, the former being composed mainly of Engelmann spruce and the latter mainly of white spruce. The contrast between the E K and PG progenies could be due to the genetic differences between the two spruce species, or dictated by the climatic and other environmental differences between the test locations. The separability among causes of phenotypic variation can be difficult since a significant genotype by environment (GxE) interaction was found for certain traits. This problem was examined in more detail for growth and wood density, and it is discussed in the section 5.2.1. 5.1.1.5 Stability of Family Performances The stability of family performance was evaluated over sites as family by site (FxS) interaction, and over growing seasons at a particular site as family by age (FxY) interaction. This interaction was evaluated also jointly for several growing seasons over sites. There was an obvious genotype-by-environment interaction, when family performances were compared at different sites. The FxS interaction was strong for some families, but not for others. Within regions, the FxS interaction was statistically significant for H. Similar result was obtained by Kiss and Yeh (1988). For other traits, the test had generally low statistical power, and was therefore uninformative. The type B genetic correlations across sites (Burdon 1977) also had substantial estimation errors and were not always particularly useful, especially for traits with lower heritabilities. To get equal genetic gains on several sites from selection among the plus trees, it may be sufficient to use an aggregated value found by weighting the breeding value on a site by the mean of that site (BLP or BLUP). The type B genetic correlation should indicate which sites are more appropriate for progeny screening. 117 Alternatively, one may want to further analyse the opportunities for target selections as described by White and Hodge (1989). Trends with age for RW and L W % were followed over 9 growing seasons. The trend for L W % was similar to the one previously reported for wood density of interior-spruce by Jozsa and Middleton (1994). For each growing season, fluctuations in RW and L W % related to cambial age were confounded with those related to climatic effects. Trends of the site and family means are much more pronounced for the cumulative than for individual annual increments. In the juvenile wood, age trends for L W % varied with site productivity as indicated by ring width. At similar cambial ages, the rate of decrease in L W % was more rapid on productive sites than on poorer sites. For some families, genotype-by-year interaction with rank change was found for both RW and L W % . Although the FxY variance component was significant over the entire period, in certain years family responses (regression coefficients) did not differ significantly. It would be interesting to pinpoint which climatic factors in certain years had caused significant Fx Y interaction, and how those climatic factors affected phenology in different families. The examination of those relationships is, however, beyond the scope of the present study. There was no clear pattern of variability and resulting heritability with cambial age, for RW or for L W % . Heritability was generally determined to a greater extent by fluctuations in the error component than by annual variation in the family component. Based on these results, it does not seem that there is a decrease in heritability with age for the above traits. Such a decrease in heritability was reported for height growth by Kiss and Yeh (1988) and for volume by Namkoong et al. (1972). Phenotypic and genetic age-age correlations for RW and L W % differed somewhat on different sites for both individual and cumulative increments. Genetic correlations were generally high for the cumulative increments, and decreased with the lag between measurements. This trend has been observed in other conifers (Namkoong et al. 1972, Lambeth 1980, Kremer 1992, Magnussen and Yanchuk 1994). Age-age correlations were modeled as the repeatability of performance over age by using the /? distribution. Bootstrap-generated distributions of age-age correlations supported assumption of the distribution as a realistic model in small populations where a normal approximation is inappropriate (Magnussen 1991). Those correlations certainly have implications for early selection and 118 breeding, and especially for the optimization of selection age. It appears that early selection for D would be more successful than for cumulative L W % . Generally, the results discussed above indicate that there is enough variation to provide opportunities for genetic improvement of growth and wood properties in the two populations. The most suitable ways for taking advantage of the variation that is present for simultaneously improving economically significant traits are discussed in section 5.3. In the next section, special attention will be given to the quantitative variation of within-ring wood anatomy, i.e., tracheid (fibre) characteristics. 5.1.2 Heritable Variation in Tracheid Characteristics 5.1.2.1 Cross Sectional Dimensions Compared to a single ring, genetic differences were generally better expressed over cumulative growth periods. Since the variable ring width (RW) was correlated with various anatomical traits and only certain sites or site combinations showed a significant family variance for RW, all subsequent analyses were limited to those sites. RW was closely correlated with the number of cells per ring (CN). This is expected, since more vigorous trees have more layers of dividing cells in the cambial zone (Gregory and Wilson 1967), and cell production is positively influenced by favourable environmental conditions such as daily solar radiation (Ford et al. 1978). Although RW was highly correlated with C N , their variation across sites was not identical. Both traits, nevertheless, had generally low heritability. RW and C N were both positively correlated with mean cell size (CS). CS is strongly controlled by hormonal activity (Larson 1968, Savidge 1996). CS had overall moderate heritability, but the magnitude of heritability was site-specific. Patterns of variation among different sources of variation, for both radial and tangential CS, were similar. Although, at the same position within a growth ring, larger cells have thicker walls (Ford and Robards 1976), ring averages for double wall thickness (DW) and CS showed little correlation. The trait DW, which is assumed to depend mostly on the availability of photosynthates. The trait had moderate to high heritability on some sites. Those sites were not always the ones for which CS had significant heritability. 119 The result of principal component analyses that RW was negatively correlated with cell density (R) was confirmed by the pairwise correlation analysis. This relationship is characteristic for Picea sp. and some other conifers (Zobel and vanBuijtenen 1989, Nyakuengama et al. 1998, Wang and Aitken 2000). The genetic improvement of the ratio of DW/CS, as a ratio of two normally distributed random variables, could be obtained through a linear or a non-linear index approximation, or through direct selection (Lin 1980, Famula 1990). The ratio R was taken here as a determining factor for selection, since it is the most closely related to the ring relative density. Its relation to growth rate is discussed in more detail in section 5.2. Cells in different regions of annual rings are usually classified in different density classes such as earlywood (EW) and latewood (LW), delimited according to some rule such as e.g., Mork's index (MI). There was little similarity in the variation pattern between latewood percentage defined according to MI (modified) and that defined according to Larson's phyto-hormonal theory of cambial activity (LW%). The transition wood (TW) comprised a large proportion of each ring. Similar results were obtained for Sitka spruce, where tracheid diameter declined steadily from the beginning of the earlywood (Denne 1976). It is generally accepted that the end of earlywood production coincides with the cessation of shoot elongation, but this transition is gradual for interior spruce (Larson 1968, Worrall 1970, Savidge 1996). Percentages of cells in different density classes (EW%, TW%, LW%) had remarkably high phenotypic and genetic correlations with radial growth rate (RW). The EW and TW portions of a growth ring are most important in spruce simply because they represent a large portion of the ring (Worrall 1970). Double cell-wall thickness, cell lumen, and their ratio in the early- and transition-wood portions appear to be under weak to moderate genetic control, but on some particular sites, heritabliity estimates were high. Since for both sets of progeny L W % had relatively high estimates of heritability and low among-site variance, it could also be significant for tree breeding. Cell morphological traits within different density classes exhibited statistically significant family variation, which was high on some particular sites. These component traits are considered as a vehicle for the improvement of overall ring density in section 5.2.3. 120 5.1.2.2 Fibre Length and Micro-fibril Angle The length of cambial initials is one of the primary factors controlling tracheid length in conifers (Bannan 1967). It is usually assumed that when growth is fast, the initials divide before they have a chance to reach their potential length. The rate of development of fusiform cambial initials can also play an important role. Therefore, it is usually assumed that the correlation of fibre length (FL) with growth rate is negative in conifers. However, either insignificant or positive correlations were found in this study, possibly because averages for the fibres in the transition wood were considered instead of ring averages. Nevertheless, some other studies also reported positive correlations between FL and growth for Picea sp. (Zobel and vanBuijtenen 1989, Dutilleul et al. 1998). As it was found previously here for some other anatomical traits, transplantation of populations to a new region may also have affected FL. It could have reduced genetic differences at the site outside the region of origin. Because FL had highly significant genetic variance at some sites, it was one of the anatomical traits considered in most of the value functions used for subsequent development of selection indices. There were generally negative genetic and phenotypic correlations for microfibril angle (MFA) with RW, and especially with FL. The negative correlation with growth rate was also found in Norway spruce by Lindstrom et al. (1998). M F A followed a similar pattern of variation as previously described for FL, and overall heritability appeared to be moderate, but higher on some particular sites. Significant differences in M F A among genetic groups were found in other conifers (Zobel and Jett 1995, Donaldson and Burdon 1996). 4.1.2.3 Trends Across Sites and with Age The above genetic and phenotypic correlations are discussed in more detail when the influence of trait correlations on genetic response is discussed below. Identification of sites with less negative correlation between growth and traits related to cell density would be desirable. The genetic correlations of anatomical traits with growth rate were generally similar among sites and for wood of different cambial age, at least in their sign. Genetic correlations were, as expected, generally of the same sign as the phenotypic correlations, but higher in magnitude. 121 Across the juvenile wood, L W % decreased with ring number from the pith. This is normally associated with an increase in the average tracheid radial-diameter. These overall trends are considered inherent to tree growth, presumably associated with cambial ageing or height of live crown, and they occurred on all sites. In the juvenile wood, L W % decreased with site productivity, as indicated by ring width, and wider rings were expected to have a more rapid rate of decrease in the ratio of tracheid wall thickness to tracheid diameter with ring age. However, the extent of the juvenile wood as delimited by the inner core of wide growth rings does not necessarily correspond to the region of varying tracheid dimensions (Denne 1997). 5.2 RELATIONSHIP BETWEEN GROWTH RATE AND WOOD DENSITY AND ANATOMY 5.2.1 Growth Rate and Wood Density: Overall and Ring-by-Ring Relationship Radial growth had generally stronger negative correlation with wood density than height growth, in the present study and the result agrees with the previous findings for spruce species (Yanchuk and Kiss 1993, Zobel and Jett 1995, Zang and Morgenstern 1995). Corrivaeu (1991) viewed selection based on height instead of diameter as an opportunity to avoid general negative correlation between volume growth and wood density. Site differences in the magnitude of the phenotypic and genetic correlations were also significant. It has been suggested that the negative relationship between growth and wood density may be weaker in families growing in a more favourable environment (Zhang and Morgenstern 1996). There was no clear evidence for such a trend in the present study. Nevertheless, some open-pollinated families not only grew fast, but also maintained relatively high wood density. On a ring-by-ring basis, phenotypic and genetic correlations of latewood percentage (trait well correlated with relative density) with growth were also consistently negative. The correlation coefficients varied from year to year. Nevertheless, correlation for cumulative increments showed a consistent pattern. A clear trend with age for genetic correlations is difficult to predict, considering the large estimation errors (Hodge and Purnell 1993). On a site having trees with high growth rates, density usually decreases more rapidly across the 122 juvenile wood to a lower minimum value than on sites with a slower growth rate (Denne 1997). Less negative correlations of certain families and at some sites could be seen as an opportunity for breaking this relationship through selection and deployment. 5.2.2 Growth Rate and Within-Ring Anatomy 5.2.2.1 Within-Ring Distribution of Cell Wall Thickness and Cell Size Within-ring data for cell densities (R) were more informative than looking at just two, low-and high-density classes, or early- and late-wood components. Changes in the relative proportion of these two classes do not describe adequately all within-ring changes in the density distribution. Their boundary also is usually arbitrarily defined. From analyses of within-ring cell-frequency distribution it was obvious that high growth rate (RW) increases skewness of R. There was also a negative relationship between RW and standard deviation of R. It was obvious that there was a frequency increase in low-density cell classes with increased growth rate, without a significant increase in high-density classes. Since wide rings had an extension of relatively uniform cells in earlywood, there was less within-ring density variation. That effectively made wood more uniform at the within ring level. This can be of great importance later, when quality of end-products, such as pulp and paper is considered (Zobel and van Buijtenen 1989). It has been reported in Sitka spruce, that increase in mean radial tracheid diameter with ring width was associated with greater increase in earlywood, than in latewood. At the same time, variations in cell wall thickness with ring width were associated with wall thickness in latewood (Denne 1997). These results depend, however, on how the demarcation between early- and latewood zone was determined. If the EW/LW boundary is shifted, the frequency of cell types falling in the two classes is affected, without changing the actual cell dimensions. When boundary of fixed density value was used in western hemlock, there was an increase in EW width, no change in L W width, but a decrease in L W proportion with an increased growth rate. There were also no or small changes in EW and L W density (De Bell etal. 1994). In the present study, the demarcation was done according to Larson (1969). There were positive correlations between growth rate and lumen size in all three density-classes (Lew, Ltw, and Llw). And there were no strong negative phenotypic correlations with the wall 123 thickness in all three density-classes (DWew, DWtw, and DWlw), in this study. In the present study, there was an increase in the TW% (Larson's definition) with increased growth rate. This also can be explained by looking at the distribution of R and its components DW and CS. Because mean DW did not change much with increase in RW, while mean CS increased, many additional cells become classified as TW. Worrall (1970) claimed that, in Norway spruce, earlywood width was correlated with height growth, while latewood width varied independently from it, and was determined largely by the cessation of radial growth. He further claimed that the extra component of larger increments could be either earlywood or latewood. The latter case may correspond to some particular families with high-density wood. Detection of such families, with both high growth rate and R, is possible according to the present results using A N C O V A with RW as a covariate. It would be also interesting to compare those "special" families with the average families, and see how they differ in their "strategy" of wood formation. Those families with less adverse combination of growth and wood density could have a genetically determined difference in the way they achieve the same ratio of double cell wall to cell size. It was not possible, however, to detect family differences in CS and DW after introducing the ratio R as a covariate. This question requires an experimental design with extra statistical power. The above observations connect biologically explained relationships and the observed data well, and give us an opportunity to look at the within ring density development and its biological meaning through component traits. 5.2.2.2 Linear Prediction of Ratio of Cell Wall Thickness to Cell Size The average ring ratio of DW/CS (R) was decomposed into component traits, according to Vargas-Hernandez et al. (1994), and the M A X R multiple regressions were performed. This was done more as an exploratory model building exercise than with an objective of predicting ring density from anatomical traits. Only three component traits were sufficient to describe more than 95% of variation in R. Double wall thickness in transition-wood (DWtw), cell lumen in transition-wood (Ltw), and cell lumen in latewood (Llw) were particularly good predictors. Ltw or Llw as single predictors described more than 70% variation in mean ring R. 124 5.2.2.3 Possibilities for Avoiding Negative Genetic Correlation between Growth Rate and Wood Density Height growth had a less negative genetic correlation with wood density than did radial growth. This indicates that putting more weight on height than diameter in selection could reduce strong negative correlations between wood density and volume. However, height growth in this study was not considered a reliable trait to correlate with corresponding wood density because of the high incidence of weevil attacks in some particular years. If wood dry-weight, which has been also suggested for selection in the simultaneous improvement of growth and density (Zhang and Morgensten 1995, Rozenberg and Cahalan 1997), is used as a major selection criterion, density may rapidly decrease. This is because of the shape of the weight yield function, which is usually non-linear for volume growth and wood density. Here the relationship between radial growth and density was examined in more detail using anatomical measures instead of density itself. Ford (1993) also indicated this possibility for selection in Sitka spruce. In certain cases, indirect selection on anatomical components of density and growth may be more efficient than direct selection on R W and R. The usefulness of anatomical component traits depends on how those traits relate to value of final product. Some particular value functions are considered and effect of selection based on wood anatomy is discussed in the next section. 5.3 OPTIMIZATION O F S E L E C T I O N A N D D E V E L O P M E N T O F B R E E D I N G S T R A T E G Y Defining a single "best" direction for a breeding population in tree breeding is usually difficult. Given also the statistical difficulties of estimating the best set of economic weights for selection indices, alternative strategies need to be considered. Breeding for volume and controlling wood density are two major objectives of many tree improvement programs. It is usually not known how these two objectives relate to other traits that may be of interest at present or in the future. Various breeding alternatives exist for improving some pulp and paper properties. Beside single population breeding, the multiple population breeding (MPB) (Namkoong et al. 1988) emerges as a viable option. 125 5.3.1 Relative Gains from Different Selection Scenarios 5.3.1.1 Single Breeding Population Single objective function Based on the optimization of selection for single objective functions, gain can be assessed and comparisons made for value functions of interest within a single breeding population. Results of this study show that selection for volume (VOL) as a single objective could bring significant improvements in this trait, and it would lead to an increase in wood dry weight (DWT). Taking V O L and DWT objectives as separate value functions would therefore not be advantageous. This result is in conflict with the suggestion of Zhang and Morgenstern (1995) for black spruce that direct selection for dry weight would result in considerably higher genetic gain in this trait than conventional selection for volume. The discrepancy could be attributed to the difference in species and environments, or significant estimation errors associated with field experiments. Since RD was not changed much by selection for V O L , it may not be profitable to form a specialised population bred for V O L with restriction of no change in RD. The restriction would significantly reduce the expected gain in V O L itself and in DWT. When correlated response was examined for value of pulp and paper properties, there was a difference between the two sets of progeny in the sign and magnitude of response. Multiple objective functions Accommodating multiple objective functions within a single breeding population requires multiobjective optimization. Allocation of objectives according to some particular criteria, i.e., different risk-reduction strategies, was examined more closely. Positive response in multiple value functions, with minor trade-offs, appears to be possible in some cases. For example, improvement of V O L would bring about positive correlated response in both tensile strength of pulp wet-webs (Tww) and tensile strength of paper (Tp). To include these two properties in a selection strategy as separate selection objectives, based on the predicted range of improvement, would not be advantageous (Kibblewhite et all 1997). Only i f significantly higher weight was placed on Tww and Tp would different solutions that warrant compromises be obtained. Without such an assumption, even for the most conservative strategy (MaxiMin), there was no difference in the sets of selected parents. 126 On the other hand, in the case of simultaneous improvement of V O L and tearing resistance of weakly bonded paper (TRi), a significant trade-off would be required. The trade-off was, more or less, linear within the range of improvement. The most conservative MaxiMin option was separated on both axes from the other options, and it resulted in different sets of selected parents. Opting for a conservative alternative, such as MaxiMin, would pay off only i f the value ratio of unit improvement in TRi and V O L was approximately 10:1, assuming an additive relationship between the two value functions. This was again due to much higher potential of V O L for improvement. Generally, when significantly higher weight is placed on pulp and paper properties, different solutions for MaxiMax, MiniMax, and MaxAvr are obtained. Trade-offs were obtained for the simultaneous improvement of the three value functions for V O L , Tww, and TRi , within one breeding population. Although positive responses were possible in all three functions at the same time, intersections of the two functions related to pulp and paper properties with the function V O L were relatively close to zero improvement. It would be necessary that significantly higher weight be placed on Tww and/or TRi before opting for some conservative option, such as MaxiMin or MaxiAvr. Generally, gains from more conservative strategies, such as MaxiMin, MiniMax, or MaxAvr, could be obtained only i f the expected value of a particular objective and its genetic response are not extreme. Assuming equal variances, product of expected value and expected genetic gain for one function should not be much larger than for the others (Namkoong 1976). In the present case, the relative response in V O L was by far the most superior, and any additional objective function would have to have a much higher relative value to justify any risk reduction strategy. If the relative values of objective functions are estimated with more precision, further optimization can be obtained through an iterative process using the interactive multiobjective method. Here a range of alternatives is presented and choosing between alternatives by classification of functions would depend on a decision-maker's preferences. Sensitivity between the two points of interest could be checked by inserting a number of new alternatives. Optimization of a breeding system would depend on both the relative value assigned to each function, and on decision-maker's attitude toward risk. If some expert knowledge is available, the value of each function can be predicted and scaled to, for 127 example, the dollar value at the time of harvesting. If this can be done with some certainty, probabilities can be assigned. Then methods such as value-variance (E-V) analyses or multiobjective stochastic optimization can be used for optimal allocation of selection intensity within one, or multiple breeding populations (Bunn 1984, Stancu-Minasian 1984, Harwood 1999). If goals, constraints, and consequences of possible actions are not precisely known, decision-making techniques in fuzzy environments may be considered (Kacprzyk and Fedrizzi 1988). If uncertainty about the relative importance of particular objectives stays high, then assignment of the objective functions to diversified breeding populations could be advantageous (Harwood 1999). 5.3.3.2 Multiple Breeding Populations Under high uncertainty about future values, which is characteristic of long rotation forestry, diversification as a strategy can be particularly useful. For example, diversification can be applied by forest managers to form mixed stands of varieties bred for different objectives. This would make possible to cover a wider range of future options and include some more risky alternatives. In that same sense, the multiple population breeding (MPB) could be advantageous at the breeding stage (Namkoong et al. 1988). The utility of having extra populations in a breeding program would depend, however, on the relationship between value functions. If there are a high number of breeding objectives with low expected values, the multiple objective optimization within one population will be very ineffective. A better solution is to form multiple populations aiming at different objectives. In the present case, when the relative value of having additional populations was examined simply as a sum of potential improvement in all objectives, the splitting of breeding population was not justified. The M P B strategy would pay off only i f the value of TR] was high relative to V O L . If assumptions were made as to what would be relative expected value for the two objective functions, then it would be reasonable to form two populations with different objectives. Aggregated expected value of the set of populations would be higher than the value of a single optimal population. The uncertainty about the value of TRi , however would have to be large before different sets of parents are obtained for different populations. The advantage of the M P B system is that trade-offs can be reduced, by reassigning breeding objectives among each of several breeding populations. In order to conclude that 128 including additional populations would be beneficial for a tree improvement program, however, it is necessary to examine selection strategies in more detail. Under a more realistic analysis, opting for two populations would depend on the breeding and maintenance costs, possible loss in selection intensity, dangers of loosing useful alleles and inbreeding (McKeand and Svensson 1997). Risk Reduction by MPB Strategy When uncertainty about future value increases, diversification becomes increasingly more useful (Roberds and Namkoong 1989). The above discussion has shown that for intrinsically risky tree breeding, any compromises with improvement of volume growth may deviate from an unknown real value. Volume is likely to continue to be valuable compared to other traits, considering the fast development of pulping technology for dealing with other traits, and the high expected response of volume to selection. Other value functions would have to have, in adition to high expected value, a high level certainty, i.e., low variance, to be included as additional objectives. If risk is defined as the variance in potential gain, a range of additional factors that can cause variation in genetic response to selection can be considered. For example, although errors in estimating economic weights appear to have only limited effects on index weighting (James 1982 in Lynch and Walsh 1999), the value functions used for their estimation might not perfectly fit the empirical data (Seth 1995). Changes in certain function parameters can cause changes in index weighting. Moreover, the formulae relating tracheid properties to pulp properties are based only on average of individual fibre characteristics (Karenlampi 1995a and 1995b). It follows that parameters that describe the effects of within-ring distribution of fibre properties on product quality could be included in simulations to predict genetic gain. The accuracy of estimation of genetic parameters is also crucial for determining the economic weights. It can be assumed that, for some growth traits, heritabilities are decreasing with age (Kiss and Yeh 1988). Inbreeding in the natural populations, where plus-trees were selected, can also have an influence on heritability estimates (Park et al. 1984). In some cases where a strong GxE interaction is present, genetic gain would be different for overall, averaged, or site-specific variance covariance matrices (Hanson and Johnson 1957, Cladwell and Weber 1965). Additional variables can also be added to those included in an index to 129 increase the accuracy of selection, but significant variation can result from using different matrices, especially if the principal traits have low heritabilities (Bouchez and Gofiinet 1990). After all these factors are considered, a question arises: How much improvement in wood quality is still possible? This study is an attempt to tackle this question by using the knowledge of site and family variation, choosing the right traits for selection and multiple selection indexes for multiple value functions. In addition, genotype by environment interactions for selection indices as composite traits can be incorporated into the models for finding the best allocation among the populations. Based on sensitivity analyses, risk reduction through diversification, and by applying M P B selection strategies, can be achieved. In order to optimise gain and associated risk, preference and risk attitude are determining factors. Therefore, this final task has to be left to the decision-makers. 130 VI CONCLUSIONS AND RECOMMENDATIONS 6.1 CONCLUSIONS REGARDING MAJOR GOALS AND SPECIFIC OBJECTIVES OF THE STUDY The following are conclusions and recommendations regarding each of major goals of this study from section 1.4. These conclusions and recommendations pertain to predicaments such as specific populations, growing conditions, tree (cambial) ages, and end-product processing conditions. However, the approach used is still valid, even i f the predicaments change. 6.1.1 Genetic Control over Growth and Wood Macro- and Micro- Characteristics 6.1.1.1 Growth and Wood Macro Characteristics For both growth and wood macro characteristics, some sites had nonsignificant family differences, while others showed moderate to high heritabilities. Since genetic variance and heritabilities differed greatly across sites, especially when trees are transplanted outside the region of origin, it would be desirable to consider breeding for particular sites or site types, if sites can be accurately classified prior to planting. There was genotype by environment interaction detected in the family response, but more extensive sampling of the existing progeny tests is needed to precisely quantify the existing interactions. Although the family by growth-season variance component was significant for the entire period, in certain years family responses did not differ significantly. It would be interesting to pinpoint which climatic and related physiological factors in certain years had caused significant interaction. Precise meteorological data and appropriate statistical tests are needed. There was also no clear pattern in the change of variance components and resulting heritability with cambial age, either for ring width, or for late-wood percentage. Estimates of genetic age-age correlations, for ring width and late-wood percentage, were generally high for 131 the cumulative increments, and tended to decrease with an increase in lag time. Those correlations could have implications for selection age optimization, and it appears that early selection for cumulative radial growth would be more successful than for cumulative latewood percentage. 6.1.1.2 Tracheid Characteristics For anatomical traits, genetic variance ranged from nonsignificant to highly significant, with moderate to high heritabilities. Heritability estimates for some of the traits were different among sites within and outside breeding zones. The wood anatomy may change more in some families than in others after transplanting to different sites, either within or outside their region of origin. Ring width was closely correlated, both phenotypically and genetically, with the number of cells per ring. Less pronounced was its positive correlation with mean cell size. On average, rings with larger cells did not have significantly thicker walls. Ring width was negatively correlated with ring density and its within-ring components. Faster growth rate did not result in shorter fibres, and positive genetic correlations were found between these traits. Fibre length had a generally negative correlation with microfibril angle. The transition from earlywood to latewood was gradual, and transition-wood comprised a large proportion of most growth rings. Remarkable were the high correlations, both genetic and phenotypic, between the ring width and measures of percentage of cells in different density classes. The cell characteristics within different ring portions showed significant family variation, and these component traits can be considered as a vehicle for the improvement of wood quality in interior spruce. 6.1.2 Wood Density and Wood Anatomy Wood density is particularly interesting, because it is, at present, the only wood quality trait that might be considered for including in selection indices for the interior spruce tree-improvement-program. Differences were observed in the extent to which growth traits were genetically correlated with wood density among populations. It would be possible to successfully breed simultaneously for the two traits among the Prince George progenies. Site 132 differences in the magnitude of the phenotypic and genetic correlations were also significant. Some half-sib families not only grew fast, but also maintained relatively higher wood density. Wide rings had a greater number of cells classified as earlywood and transition wood, without an increase in number of cells classified as latewood. The extension of relatively uniform cells in low density classes reduced overall within-ring density variation, i.e., more uniform wood at the within ring level was formed. This can be of importance for quality of end-products, such as pulp and paper. Only three other anatomical variables were sufficient to describe more than 95% of variation in ring density. Double wall thickness in transition-wood, cell lumen in transition-wood, and cell lumen in latewood were particularly good predictors. This could provide an opportunity to look at the within ring density and its biology of development, by looking at its component traits. There were no particular benefits, however, from considering the component traits for breaking the negative genetic correlation between growth and density. Only in certain cases will indirect selection on density-component traits be more efficient than direct selection. 6.1.3 Effects of Selection on Value of Final Products Based on the existing knowledge of relationships between fibre properties and paper quality, effects of selection on value of final products from improved trees can be examined. 6.1.3.1 Single Breeding Population Selection within single breeding population for volume, as a single objective, could result in significant improvements, positively influencing dry weight and some pulp and paper characteristics, and resulting either no or a slightly negative changes in wood density. Selection for volume, with a restriction of no change in wood density, would significantly reduce genetic gain in both volume and dry-weight. In order to put the confidence limits on the expected genetic response, the sensitivity analyses clearly require more examination, and this can be suggested as a topic for future research. Simultaneous improvement of volume growth and pulp and paper properties would always require trade-offs, and multiobjective optimization would be beneficial. Improvement without much trade-off was possible for volume and tensile strength of pulp and paper. 133 Significant trade-offs would be required to improve both volume and tear factor of paper, however. Due to high genetic responses in volume relative to other objectives, solutions for risk-reducing options were all approximately the same. Estimated solution parameters may deviate critically from an unknown real value. Only i f significantly higher weight is placed on objectives other than volume would different solutions be obtained. The conservative options nevertheless seem unrealistic because compromises with volume growth may not be warranted. 6.1.3.2 Benefits from Multiple-Population Breeding Strategy Under uncertainty about future values of breeding objectives, introduction of additional populations could be considered. The utility of having extra populations in a breeding program would depend on the relationship between value functions. Theoretically, total expected loss, would be reduced if two or more populations could be formed, and two selection indices derived with two different weights on objectives. Aggregated expected value of such a set of populations would always exceed the value of one population at a single optimum. Expected genetic response in volume was by far the most superior, and any additional objective function would need to have a high relative value to justify any of the risk reduction strategies, including diversification through multiple breeding population system. 6.2 EVALUATION OF HYPOTHESES Four major hypotheses from section 1.3 of this thesis are addressed here. The following are major conclusions and recommendations regarding simultaneous improvement of growth and tracheid properties in interior spruce: 1) There is substantial genetic variation for growth and wood properties, including tracheid characteristics, in the spruce populations studied. Therefore, genetic gains can be achieved through simultaneous selection for those traits. 2) Only in certain cases, and only at particular sites, may indirect selection on density-component traits be more efficient than direct selection. 134 3) Simultaneous selection is significantly influenced by the degree of inheritance, genetic and phenotypic correlations, and the character of the objective value functions that relate the traits to product quality. Analyses based on multiobjective optimization are needed to provide suitable ways for taking advantage of the variation that is present. 4) When complexity and uncertainty regarding the value functions are considered, multiple index selection, as a part of a multiple-population breeding strategy is more efficient than traditional index selection within a single population. It is a viable option to be considered, but its usefulness will depend on one's aversion towards the risk of failure in adopting wood quality as a selection objective. 135 VII REFERENCES Anderson V L , McLean R A 1974 Design of Experiments: a realistic approach. Marcel Dekker Inc., N Y , 418pp Anderson JR, Dillon JL, Hardaker B 1997 Agricultural Decision Analysis. Iowa State university Press, Iowa. Aubry C A , Adams WT, Fahey TD 1998 Determination of relative economic weights for multitrait selection in coastal Douglas-fir. Can J For Res 28(8): 1164-1170 Baker RJ 1986 Selection Indices in Plant Breeding. CRC Press, Florida, 218pp Barman M W 1956 Some aspects of the elongation of fusiform cambial cells in Thuja occidentalis. Can Jour Bot 33:175-196 Barnes RD 1994 The seedling orchard in the multiple population breeding strategy. Silvae Genet 44(2-3):81-88 Borralho N M G , Cotterill PP, Kanowski PJ 1993 Breeding objectives for pulp production of Eucalyptus globulus under different industrial cost structures. Can J For Res 23(4):648-656 Bouchez A , Gofiinet B 1990 Evaluation of selection index: application the choice of an indirect multitrait selection index for soyabean breeding. Theor Appl Genet 79:261-267 Boyle TJ, Balatinecz JJ, McCaw P M 1987 Genetic control of some wood properties of black spruce. 21 s t Can Tree Impr Assoc, Truro, Nova Scotia, 198 pp British Columbia Ministry of Forests 1976 Whole stem standing cubic meter volume equations and tables: centimeter diameter class merchantible volume. Brown C 1970 Physiology of wood formation in conifers. Wood Science 3(1): 8-22 Brown G H 1969 An empirical study of the distribution of the sample genetic correlation coefficient. Biometrics 22:63-72 Bunn DW 1984 Applied Decision Analysis. McGraw-Hill Inc, 247 pp 136 Burdon RD 1977 Genetic correlation as a concept for studying genotype by environment interactions in forest tree breeding. Sylvae Genetica 26(5-6): 168-175 Cladwell Weber 1965. Selection response in soyabean populations Crop Sci 5:223 Clark Jd'A 1985 Pulp Technology and Treatment of Paper. 2 n d ed., Miller Freeman Publicatoins, San Francisco. Coates DK, Haesussler S, Lindenburgh S, Pojar R, Stock A J 1994 Ecology and Silviculture of Interior Spruce in British Columbia. FRDA Report 220, Forestry Canada and BC Ministry of Forests, 182pp Cockerham CC, Wier BS 1977 Quadratic analysis of reciprocal crosses. Biometrics 33:187-203 Cockrell R A 1974 A comparison of latewood pits, fibril orientation, and shrinkage of normal and compression wood of giant sequoia. Wood Sci Tech 8:197-206 Cotterill PP, Jackson N 1984 On index selection I. Methods of determining economic weight. Silvae Genet 34:2-3 Cotterill PP, James JW 1984 Number of offspring and plot sizes required for progeny testing. Sylvae Genetica 33:202-209 ' Corriveau A , Beaulieu J, Doust G 1991 Heritability and genetic correlations of wood characteristics of Upper Ottawa Valley white spruce populations grown in Quebec. For Chron 67(6): 698-705 Corriveau A, Beaulieu J, Mothe F, Poliquin J, Doucet J 1990 Densite et largeur des cernes des populations d'epinettes blanches de la region forestiere des Grands lacs et du Saint-Laurent. Can J For Res 20:121-129 Corson SR 1999 Tree and fibre selection for optimal TMP quality. Appita J 52(5):351-357 Danell O 1991 Surway of past, current and future tree breeding in Sweden. Silva Fennica 24:241-247 DeBell JD, Tappeiner JC, Krahmer R L 1994 Wood density of western hemlock: effect of ring width. Can J For Res 24 (3) 638-641 Denne M P 1976 Wood production and structure in relation to bud activity in some softwood and hardwood species. Leiden Botanical Series 3:204-211 Denne MP 1988 Definition of latewood according to Mork (1928). IAWA Bull ns 10(l):59-62 137 Donaldson, L A Burdon RD 1996 Clonal variation and repeatability of microfibril angle in Pinus radiata. NZJFor Sci 25(2): 164-174 Dutilleul P, Herman M , Avella-Shaw T 1998 Growth rate effects on correlations among ring width, wood density, and mean tracheid length in Norway spruce (Picea abies). Can J For Res 28 (l):56-68 Dykstra DP 1984 Mathematical Programing for Natural Resource Management. McGraw Hi l l , N Y , 309pp Evans R, Kibellwhite PR, Stringer S 1996 Kraft pulp property from wood properties in eleven radiata pine clones. CSIRO Forestry and Forest Products. Report No. B203, March 1996, and Proceedings of 50 t h Appita annual general conference, Auckland, May 1996. Environment Canada 2000 Canadian Climate Normalsl961-1990 http://www.cmc.ec.gc.ca/climate /normals/BCC026.HTM(15/9/2000) Falconer DS 1989 Introduction to Quantitative Genetics. 3 r d ed., Longman, London, 340pp Famula TR 1990 The equivalence of two linear methods for the improvement of traits expressed as ratios. Theor Appl Genet 79:853-856 Ford ED 1993 Potential measures for use in developing selection procedures for high timber density in juvenile wood of Picea sitchensis: II. Cell dimensions and wood histology. Journal of Sustainable Forestry 1(2): 43-67 Ford ED, Robards A W 1976 Short term variation in tracheid development in the early wood of Picea sitchensis. Leiden Botanical Series 3:212-221 Ford ED, Robards A W , Piney M D 1978 Influence of environmental factors on cells production and differentiation in the early wood of Picea sitchensis. Ann Bot 42:683-692 Fowler DP, Roche X 1976 Genetics of Engelmann spruce. US Dep Agric, For Serv, Washington DC, Res Pap WO-30 Franklin C L 1945 Preparing thin section of synthetic resin and wood-resin composites, and a new maceration method for wood. Nature 155:51 Fritts HC 1976 Tree Rings and Climate. Academic Press, 567 pp Gianola D, Norton HW 1981 Scaling threshold characters. Genetics 99:357-364 Gibson JP, Kennedy B W 1990 The use of constrained selection indexes in breeding for economic merit. Theor Appl Genet 80:801-805 138 Goddard M E 1983 Selection indices for non-linear profit functions. Theor Appl Genet 64:339-344 Gordon A 1996 The sweep of the boreal in time and space, from forest formations to genes, and implications for management. For Chron 72(1): 19-30 Gregory RA, Wilson BF 1968 A comparison of cambial activity of white spruce in Alaska and New England. Can J Bot 46: 733 Hall JP 1984 The Relationship between Wood Density and Growth Rate and the Implication for the Selection of Black Spruce Trees. Information Report N-X-224, Newfoundland Forest Research Centre, Canadian Forestry Service, 22pp Hannrup B 1999 Genetic Parameters of Wood Properties in Pinus Silvestris (L.). Doctoral Thesis, Swedish University of Agricultural Sciences, Uppsala. Hansen HP 1955 Postglacial forests in south central and central British Columbia. American Journal of Sciences 253:640-658 Hansen K H 1999 Selection strategy in four 13-year old clonal trial with Sitka spruce (Picea sitchensis (Bong.) Carr.). Submitted to the Can J For Res. Hanson WD, Johnson HW 1957 Methods for calculating and evaluating a general selection index obtained by pooling information from two or more experiments. Genetics 42:421-432 Harris DL 1970 Breeding for efficiency in livestock production: defining the economic objectives. JAnim Sci 30:860-865 Harville D A 1975. Index selection with proportionality constrains. Biometrics 31:223-225 Harwood Joy L 1999 Managing risk in farming: concepts, research, and analysis. USDA Economic Research Service, Agricultural economic report No774, 125 pp Hirakava Y , Fujisawa Y1995 The relationships between microfibril angles of the S2 layer and latewood tracheid lengths in elite sugi tree (Cryptomeria japonica) clones. J Japan Wood Res Soc 41(2):123-131 Horgan GW and Hunter E A 1993 Introduction to R E M L for scientists. Biomathematics and Statistics Scottland, Scottish Agricultural Statistics Service, University of Edinburgh. Horn R A 1974 Morphology of wood and pulp fibre from softwoods and influence on paper strength. USDA Forest Service Res Pap 242 139 Hodge GR, Purnell RC 1993 Genetic parameter estimates for wood density, transition age, and radial growth in slash pine. Can J For Res 23(9): 1881-1891. Hui Liu B, Knapp SJ, Birkes D 1997 Sampling distributions, biases, variances, and confidence intervals for genetic correlations. Theor Appl Genet 94:8-19 Itoh Y , Yamada Y 1988 Linear selection indices for non-linear profitfiinctions. Theor Appl Genet 75:553-560 Ivkovich M , Koshy MP 1997 Wood density measurement: Comparison of X-ray, Photometric, and Morphometric Methods. Proceedings of The 26th Biannual Meeting of the Canadian Tree Improvement Association (CTIA/IUFRO), International Workshop on Wood Quality, Quebec City. Jansson G, Danell O 1993 Needs and benefits of empirical power transformations for production and quality traits in forest tree breeding. Theor Appl Genet 87:487-497 Jagels R, Telewski FW 1990 Computer-Aided Image Analysis of Tree Rings. In: Cook ER and Kairiukstis L A Eds. Methods of Dendochronology: Applications in the Environmental Sciences. Kluwer Academic Publishers, Boston, 394pp Jandel Corporation 1995 SigmaScanPro© Automated Image Analysis Software, User's Manual. Jaquish B 1982 Variation studies in interior spruces (E.P. 646, 672) In Forest Research Review 1981-82. BC M i n For Res Br, Victoria B C , p61-62 Josza L A , Middleton GR 1994 A Discussion of Wood Quality Atributes and Their Practical Implications. Forintek Canada Corp. Special Publication No.SP-34, 42pp Kacprzyk Janusz, Fedrizzi Mario 1988 Combining fuzzy imprecision with probabilistic uncertainty in decision making. Springer-Verlag, 399pp Kang H, Nienstaedt H 1986 Managing long-term tree breeding stock. Silvae Genet 36(1):30-Karenlampi P 1995a Effect of distributions of fibre properties on tensile strength of paper: A closed form theory. Pulp Paper Sci 21(4):J138-143 Karenlampi P 1995b Tensile strength of paper: A simulation study. Pulp Paper Sci 21(6): J209-13 Karenlampi P 1995c Tensile strength of mixture of two pulps. Pulp Paper Sci 21(12):J432-436 140 Karnis A J 1994 The mechanism of fibre development in mechanical pulping. Pulp Paper Sci 20(10):J280-288 Kempthorne O, Nordskog A W 1959. Restricted selection index. Biometrics 15:10-19 Kennedy RW 1961 Variation and periodicity of summerwood in some second-growth Douglas-fir. TAPPIJM:\6\-\65 Kennedy RW 1995 Coniferous wood quality in the future: concerns and strategies. Wood Sci Tech 29:321-338 Khalil M A K 1985 Genetics of wood characters of black spruce (Picea mariana (Mill.) B.S.P.) in Newfoundland, Canada. Silvae Genet 34(6):221-230 Kibbelwhite RP, Shelboune CJA 1997 Genetic selection of trees with designer fibres for different paper and pulp grades. Proceedings of 11 t h fundamental research symposium. Cambridge, September 1997. Kibellwhite RP, Evans R, Riddell M K 1997 Handsheet property prediction from kraft-fibre and wood-tracheid properties in eleven radiata pine clones. Appita J 50(2): 131-138 King J N , Yanchuk A D , Kiss GK, Alfaro RI 1998 Genetic and phenotypic relationship between weevil (Pissodes strobi) resistance and height growth in spruce populations of British Columbia. Can J For Res 27(5):732-739. King JN, Yeh FC, Heaman JCH, Dancik BP 1988 Selection of wood density and diameter in controlled crosses of coastal Douglas-fir. Silvae Genet 37:152-157 King J N , Cartwright C, Hatton J, Yanchuk A D 1998 The potential of improving western hemlock pulp and paper quality. I. Genetic control and interrelationships of wood and fibre traits. Can J For Res 28(6):863-870. Kiss G, Yeh FC 1988 Heritability estimates for height for young interior spruce in British Columbia. Can J For Res 18:158-162 Kiss G, Yanchuk A D 1991 Preliminary evaluation of weevil resistance of interior spruce in British Columbia. Can J For Res 21:230-234 Klee FWP 1996 Optimal Risk Management Strategies for a Cattle Backgrounding Operation in the Peace River Area. MSc Thesis: The University of British Columbia. Koots KR, Gibson JP 1996 Realized sampling variances of estimates of genetic parameters and the difference between genetic and phenotypic correlations. Genetics 143:1409-1416 141 Krajina V J , Klinka K, Worall J 1982 Distribution and ecological characteristics of trees and shrubs of British Columbia. Fac For, Univ BC, Vancouver BC. Krasowski MJ , Letchford T, Eastham A M 1993 Growth of short day treated spruce seedlings planted throughout British Columbia. FRDA Report 209, Forestry Canada, 39pp Kremer A 1992 Predictions of age-age correlations of total height based on serial correlations between height increments in Maritime pine (Pinus pinaster Ait.). Theor App Genet 85:152-158 Kuehl RO 1994 Statistical Principles of Research Design and Analysis. Duxbury Press. Kyrkjeeide PA 1990 A wood quality study of suppressed, intermediate, and dominate trees of plantation grown Picea abies. Forest Products laboratory, Madison, WI, USA. Land SBJr, Mattheiss T H 1983 A multiple-objective Forest Tree Breeding Strategy Proc-South-Conf-For-Tree-Iprovm. USDA, 146-155 Lambeth CC 1980 Juvenile-mature correlations in Pinaceae and implications for early selection. For Sci 26:571-580 Langlet O 1963 Patterns and terms of intra-specific ecological variability. Nature 200:347-348 Larson PR 1969 Wood formation and the concept of wood quality. Yale University: School of Forestry, Bulletin 74, 53pp Larson PR 1994 The Vascular Cambium: Development and Structure. Springer-Verlag N Y . Lasdon LS, Waren A , Jain A , Ratner M 1978 Design and Testing of a Generalized Reduced Gradient Code for Nonlinear Programming. ACM Transactions on Mathematical Software 4 ( l):34-50. Lasdon LS, Smith S 1992 Solving Sparse Nonlinear Programs Using GRG. ORSA Journal on Computing 4(1):2-15. Ledell 1970 Quality criteria in black spruce: an examination of the relationships between paper strength and the characteristics of the wood. Department of Land and Forests, Wood Ontario Research Foundation. Lester DT 1993 Utilizing genetic resources of conifers in British Columbia. BC Min For, Victoria, BC Research Report 93001-HQ Lester DT, Ying CC, Konishi JD 1990 Genetic control and improvement of planting stock. In: Regenerating British Columbia's Forests. Lavander DP, Parish R, Johnson C M , 142 Montgomery G, Vyse A , Willis RA, Winston D (editors), Univ BC Press, Vancouver, BC 180-192 Lindgren D, Gea L, Jefferson P 1996 Loss of genetic diversity monitored by status number. Silvae Gewef 45(l):52-59 Lindstrom H 1997 Fiber length, tracheid diameter, and latewood percentage in Norway spruce: development from pith outwards. Wood Fiber Sci 29(l):21-34 Lindstrom H, Evans JW, Verrill SP 1998 Influence of cambial age and growth conditions on microfibril angle in young Norway spruce (Picea abies [L.] Karst). Holzfoschung 52(6):573-581 Littell RC, Milliken GA, Stroup WW, Russell Wolfinger 1996 SAS System for Mixed Models. SAS Institute Inc. Cary, NC, USA. Loo-Dinkins J 1992 Field test design. In Fins L et al. (eds.) Handbook of Quantitative Genetics, Kluwer Academic Publ, 96-139 Loo-Dinkins JA, Gonzales JS 1991 Genetic control of wood density profile in young Douglas-fir. Can J For Res 21:935-939 Lowe WJ, vanBuijtenen JP 1986 The development of a sub-lining system in an operational tree improvement program. In: Proceedings IUFRO conference-Joint meeting of working parties on breeding theory, progeny testing and seed orchards. Williamsburg, Virginia. October 13 to 17, 1986 N C State University-INdustry Cooperative Tree improvement Programme, 89-106 Lin C Y 1980 Relative efficiency of selection methods for improvement of feed efficiency. J Dairy Sci 63:491-494 Lynch M , Walsh B 1997a Genetics and Analysis of Quantitative Traits. Vol . 1 Sinauer Associates Inc, USA, 980pp Lynch M , Walsh B 1999 Genetics and Analysis of Quantitative Traits. Vol . 2 h t t p : / / n i t r o . b i o s c i . a r i z o n a . e d u / z b o o k / v o l u m e 2 / v o l 2 . html(20/10/1999) MacLeod J M 1986 Kraft pulps from Canadian wood species. Pulp and Paper Canada 87(1):76-81 Magnussen S 1990 Selection index: economic weights for maximum simultaneous genetic gain. Theor Appl Genet 79:289-293 Magnussen S 1991 A distribution model for heritability. Genome 3 5 (6): 931-93 8 143 Magnussen S 1991 A probability distribution model for age-age correlations and its application in early selection. Can J For 21(10): 1550-1558 Magnussen S, Yanchuk A 1994 Time trends of predicted breeding values in selected crosses of coastal Duglas fir in British Columbia: A methodological study. For Sci 40(6):663-685 Matolcsy G A 1975 Correlation of fibre dimensions and wood properties with the physical properties of kraft pulp of Abies balsamea L. (Mill.). TAPPIJ 58(4): 136-141 McKnead S, Svensson J 1995 Loblolly pine: Sustainable management of genetic resources. J For 95(3):4-9 Miettinen K 1999 Nonlinear Multiobjective Optimization. Kluwer Academic Publishers, 298pp Miettinen K, Makela M M 1995 Interactive bundle-based method for nondifferentiable multiobjective optimization: NIMBUS . Optimization 34:231-246 Miettinen K, Makela M M 1996 Comparing Two Versions of NIMBUS Optimization System, Report 23/1996, University of Jyvaskyla, Department of Mathematics, Laboratory of Scientific Computing, ht tp : //nimbus • math . j yu . f i / i n f o . h t m l Mitchell M D , Denne MP 1997 Variation in density of Picea sitchensis in relation to within-tree trends in tracheid diameter and wall thickness. Forestry 70(l):47-60 Morgenstern K E 1996 Geographic Variation in Forest Trees: Genetic Basis and Application of Knowledge in Silviculture. U B C Press. Nyakuengama JG, Matheson C, Spencer D, Evans R, Vinden P 1998 Correlation among growth, density, fibre diameter and heartwood in radiata pine. Appita J51(l):35-38 Namkoong G 1966 Inbreeding effects on estimation of additive genetic variance. For Sci 12(1):8-13 Namkoong G 1976 A multiple-index selection strategy. Silvae Genet 25:5-6 Namkoong G 1985 The influence of composite traits on genotype by environment relations. Theor Appl Genet 70:315-17 Namkoong G1997 A gene conservation plan for loblloly pine. Can J For Res 27:433-437 Namkoong G, Jhonson 1986 Influence of the value function on genotype-by-environment relations. Silvae Genet 36(2):92-94 144 Namkoong G, Kang HC, Brouard JS 1988. Tree Breeding: Principles and Strategies. Springer-Verlag, N Y , 180pp Namkoong G, Usianis RH, Silen RR 1972 Age-related variation in genetic control of height growth in Douglas Fir. Theor Appl Genet 42:151-159 C A M C O R E Central America & Mexico Coniferous Resources Cooperative. North Carolina State University 2000 < h t t p : //www2.ncsu.edu/ c a m c o r e > Nepveau G 1984 Variabilite genetique de la qualite du bois chez l'epcea commun et le douglas. Revue Forestiere Frangaise 36(4):303-312 Nienstaedt H , Teich A 1972 Genetics of white spruce. US Dep Agric For Serv, Washington DC, Res Pap WO-15 NIMBUS Nondifferentiable Interactive Multiobjective Bundle-based optimization System, University of Jyvaskyla, Department of Mathematical Information Technology, Finland, 2000. < h t t p : / / n i m b u s . m a t h . j y u . f i / > Oglivie RT, von Rudolff E 1968 Chemosystematic studies in the genus Picea (Pinaceae). IV The introgression of white and Engelmann spruce as found along the Bow River. Can J f l o f 46:901-908 Olesen PO 1977 The variation of the basic density level and tracheid width within the juvenile and mature wood of Norway spruce. Forest Tree Improvement, Arboretet Hoersholm, 12,21 pp O'Neill P, Mohiddin S, Michell A J 1999 Exploring data for the relationships between wood, fibre and paper properties. Appita /52(5):358-62, 382 Page D H 1969 A theory for the tensile strength of paper. TAPPI J 52(4):674-681 Page D H 1993 A quantitative theory of the strength of wet webs. J Pulp Paper Sci 19(4): J175-176 Page DH, Seth RS 1988 A note on the effect of fibre strength on the tensile strength of paper. L4mJ71(10):182-183 Page DH, El-Hosseiny F, Winkler K, Lancaster APS 1977 Elastic modulus of single wood fibres. TAPPIJ60(4):\\4-W Page DH, Seth RS, DeGrace JH 1979 Elastic modulus of paper: I. The controling mechanisms. TAPPI J 62(9):99-102 Park YS , Adams GW 1993 Breeding strategies of important tree species in Canada. Fredricton, N B , Nat Res Can Report M - X 1861 145 Park YS, Fowler DP, Coles JF 1984 Population studies of white spruce. II. Natural inbreeding and relatedness among neighboring trees. Can J For Res 14:909-913 Parker M L , Jozsa L A 1973 X-ray scanning machine for tree ring width and density analyses. Wood Fiber 5:192-197 Patterson HD, Thompson R 1971 Recovery of inter-block information when block sizes are unequal. Biometrica 58:545-554 Pesek J, Baker RJ 1969 Desired improvement in relation to selection indices. Can J Plant Sci 49:773-783 Raymond CA, Greaves B L 1997 Developing breeding objectives for kraft and cold caustic soak (CCS) pulping of eucaliptus. CTIA/IUFRO International Wood Quality Workshop Proceedings, Quebec City. Regent Instruments 1999 WinDendro and WinCell user manuals. Regent Instruments, Quebec. Revel J 1969 Effects of altitudinal and latitudinal displacement of white and Engelmann spruce provenances, Prince George plantations. In For. Res Rev BC For Serv, 39pp Roberds JH, Namkoong G 1989 Population selection to maximize value in an environmental gradient. Theor Appl Genet 77:128-134 Robertson A 1959 Experimental design in the evaluation of genetic parameters. Biometrics 15:219-226 Roche L 1967 Geographic Variation in Picea glauca in British Columbia. Ph.D. thesis: The University of British Columbia, 207pp Roff D A 1997 Evolutionary Quantitative Genetics. Chapman&Hall, N Y , 493pp Roff DA, Preziosi R 1994 The estimation of the genetic correlation: the use of jackknife. Heredity 73:544-548 Rolland C, Schueller JF 1994 Relationship between mountain pine and climate in the French Pyrenees (Font-Romeu) studied using radio densitometrical method. Pirineos 143/144:55-70. Rozenberg P, Cahalan C 1997 Spruce and Wood Quality: Genetic Aspects (A Review). Silvae Genet. 46(5):270-279 Rudie A W 1998 Wood and how it relates to paper products. TAPPIJ%\(5): 223-228 146 Rudie A W , Morra J, St Laurent J, Hickey K 1994 The influence of wood and finber propertieson mechanical pulping. TAPPI11'(6):86-89 SAS Institute Inc 1987 SAS® Guide to Macro Processing, Version 6, First Edition, SAS Institute Inc, Cary, NC, 233pp SAS Institute Inc 1988 SAS/IML™ User's Guide, Release 6.03 Edition, SAS Institute Inc, Cary, N C , 357pp SAS Institute Inc 1996 SAS/STAT® User's Guide, Version 6, Fourth Edition, Volumes 1 and 2, SAS Institute Inc, Cary, NC, 1686pp SAS Institute Inc 1997 SAS/STAT® Software: Changes and Enhancements through Release 6.12. SAS Institute Inc, Carry, N C , 1162pp Savidge R A 1996 Xylogenesis, genetic and environmental regulation: a review. IAWA 17(3): 269-310 Seaby DA, Mowat DJ 1993 Growth changes in 20-year-old Sitka spruce {Picea sitchensis) after attack by the green spruce aphid (Abietinum elatobium). Forestry 66(4):371-379 Senft JF, Bendtsen A B 1985 Measuring microfibrilar angles using light microscopy. Wood Fibre Sci 17(4):564-567 Seth RS 1995 The effect of fiber length and coarseness on tensile strength of wet webs: a statistical geometry explanation. TAPPIJ798(3):99-102 Seth RS 1996 Optimizing reinforcement pulps by fracture toughness. TAPPIJ 19 (1):170-177 Shaw R 1987 Maximum likelihood approaches applied to quantitative genetics of natural populations. Evolution 41(4):812-826 Shaw R 1991 The comparison of quantitative genetic parameters between populations. £vo/Kf/o»45(l):143-151 Shelbourne T, Evans R, Kibblewhite P, Low C 1997 Inheritance of tracheid transverse dimensions and wood density in radiata pine. Appita J50(l):47-50, 67. Sorensson CT, Cown DJ, Ridoutt BG, Tian X 1997 The significance of wood quality in tree breeding: A case of radiata pine in New Zeland. Proceedings: The 26th Biannual Meeting of the Canadian Tree Improvement Association (CTIA/IUFRO), International Workshop on Wood Quality, Quebec City. Stancu-Minasian IM 1984 Stochastic programming with multiple objective functions. Editura Academiei, Bucuresti, Romania, Distributors for the U.S.A. and Canada, Kluwer Academic, 334pp 147 Stoehr M U , El-Kassaby Y A 1997 Levels of genetic diversity at different stages of domestication cycle of interior spruce in British Columbia. Theor Appl Genet 94(l):83-90 Sutton BCS, Flanagan DJ, Gawley JR, Newton C H , Lester DT, El-Kassaby Y A 1991 Inheritance of chloroplast and mitochondrial D N A in Picea and composition of hybrids from introgression zones. Theor Appl Genet 82:242-248 Talbert CB 1984 An analysis of several approaches to multiple-trait index selection in loblolly pine {Pinus taeda L). PhD Thesis. North Carolina State University at Raleigh, 106pp Tallis GW 1962 A selection index for optimum genotype. Biometrics 18:120-122 Taylor FW, Wang EI, Yanchuk A , Micko M M 1982 Specific gravity and tracheid length variation of white spruce in Alberta. Can J For Res 12:561-566 VanVleck LD, Henderson CR 1961 Empirical sampling estimates of genetic correlations. Biometrics 34:123-127 Vargas-Hernandez J, Adams WT 1991 Genetic variation of wood density components in young coastal Douglas-fir: implications for tree breeding. Can J For Res 21(12): 1801-1807 Vargas-Hernandes J, Adams TW 1994 Genetic relationship between wood density components and cambial growth rhythm in young coastal Douglas-fir. Can J For Res 24 (9):1871-1876. Vargas-Hernandez J, Adams WT, Krahmer R L 1994 Family variation in age trends of wood density traits in young coastal Douglas-fir. Wood-Fiber-Sci 26 (2):229-236 Villaneuva B, Kennedy B W 1990 Effects of selection on genetic parameters of correlated traits. Theor Appl Genet 80:706-712 Vose D 1996 Quantitative Risk Analysis: A Guide to Monte Carlo Simulation Modeling Wiley, Toronto, 328pp Wang T, Aitken S 2000 Selection for improved growth and wood quality in lodgepole pine: effects on xylem anatomy of tree stems. Wood and Fibre Sci (in press) Wearden S 1959 The use of power function to determine an adequate number of progeny per sire in a genetic experiment involving half-sibs. Biometrics 15:417-423 Westin J, Sundblad L G , Hallgren JE 1995 Seasonal variation in photochemical activity and hardiness in clones of Norway spruce (Picea abies). Tree Physiol 15(10):685-689 148 White TL and Hodge GR 1989 Predicting Breeding Values with Applications in Forest Tree Improvement. Kluwer Academic, Dordrecht, The Netherlands, 369pp Whiteman PH, Cameron JN, Farrington A 1996 Breeding trees for improved pulp and paper production. Appita J49(l):50-53 Williams D G 1983 A fiber network model theory for the wet web strength of paper. TAPPIJ 66(3): 159-162 Williams M F 1994 Matching wood fibre characteristics to pulp and paper processes and products. TAPPI ./77(6):86-90 Williams ER, Matheson A C 1996 Experimental Design and Analysis for the Use in Tree Improvement, CISRO Australia, 174pp Williams CG, Hamrick JL, Lewis PO 1995 Multiple-population versus hierarchical conifer breeding programs: a comparison of genetic diversity levels. Theor Appl Genet 90 (3/4):584-594 Worrall J 1970 Interrelationships among some phenological and wood property variables in Norway Spruce. TAPPI J53(l):58-63 Worrall J 1975 Provenance and clonal variation in phenology and wood properties of Norway Spruce. Silvae Genet 24:2-5 Xu P 1998 Long-Term Growth Responses In Sitka Spruce Populations To Geoclimatic Changes From IUFRO Provenance Trials In British Columbia. MSc Thesis, The University Of British Columbia, Vancouver. Xie C Y , Yanchuk A D , Kiss G K 1998 Genetics of Interior Spruce in British Columbia: Performance and Variability of Open-pollinated Families in the East Kootenays. BC Ministry of Forests, Research report 07. Yang K C , Hazenberg G1994 Impact of spacing on tracheid length, relative density, and growth rate of juvenile wood and mature wood in Picea mariana. Can J For Res 24(5):996-1007. Yamada Y 1995 Are economic selection indices always superior to a desired gains index? Theor Appl Genet 91:655-658 Yanchuk A D , Kiss G 1993 Genetic variation in growth and wood specific gravity and its utility in the improvement of interior spruce. Silvae Genet A2(2/3):14\-\41 Zhang SY, Morgenstern E K 1995 Genetic variation and inheritance of wood density in black spruce (Picea mariana) and its relationship with growth: implications for tree breeding. Wood Sci Tech 30:63-75 149 Zhang SY, Simpson D, Morgenstern E K 1996 Variation in the relationshop of wood density with growth in 40 black spruce (Picea mariana) families grown in New Brunswick. Wood Fiber Sci 28(l):91-99 Zobel BJ, vanBuijtenen JP 1989 Wood Variation: Its Causes and Control. Springer-Vearlag, N Y , 363pp Zobel BJ, Jett JB 1995 Genetics of Wood Production. Springer-Verlag, N Y , 337pp Zobel BJ, Sprague JR 1998 Juvenile Wood in Forest Trees. Springer-Verlag, N Y , 300pp 150 VIII APPENDICES 151 APPENDIX 1: LOCATION OF SELECTED TREES FOR EAST KOOTENAY AND PRINCE GEORGE REGIONS L e g e n d E K E a s t Koo tenay M G R M a c G r e g o r N C H N e c h a k o Q L Q u e s n e l L a k e C T C a r i b o o Trans i t ion T r e e No. P lan z o n e E levat ion Latitude Longi tude D B H ( c m ) H ( m ) 2 E K 1372 50 03 1 1 5 2 8 69 .6 40 3 E K 1311 50 04 115 31 50 35 4 E K 1585 50 08 115 06 50.3 32 5 E K 1570 50 09 115 07 68.1 40 6 E K 1509 50 10 1 1 5 0 9 52.8 34 7 E K 1463 50 10 115 11 80.8 4 6 8 E K 1418 50 10 115 11 49 .3 38 9 E K 1372 50 11 115 14 60 .2 38 10 E K 1311 50 11 115 15 60 .5 35 11 E K 1311 50 13 115 15 40 .9 29 12 E K 1250 50 12 115 16 39.4 24 13 E K 1524 50 24 1 1 5 2 1 48 .5 30 14 E K 1524 50 23 1 1 5 2 0 55.4 35 15 E K 1524 50 22 1 1 5 2 0 78 40 16 E K 1494 50 21 1 1 5 2 0 6 5 37 17 E K 1483 s o 21 115 19 57.7 30 18 E K 1463 50 20 115 18 47.8 30 19 E K 1433 50 18 115 18 59.9 40 20 E K 1402 50 15 115 16 59.7 30 21 E K 1433 50 14 115 17 6 3 37 22 E K 1433 50 14 115 17 78.7 41 23 E K 1158 50 10 1 1 5 2 7 47 .8 35 24 E K 1128 50 13 1 1 8 2 7 53.6 35 2 5 E K 1158 50 09 1 1 5 2 7 60 .5 34 26 E K 1402 50 07 1 1 5 2 5 51.8 34 27 E K 1220 50 09 1 1 5 2 6 59.4 38 28 E K 854 50 17 1 1 5 4 1 4 5 30 29 E K 1189 49 37 115 33 42 .4 27 30 E K 1189 50 37 115 33 50.5 35 31 E K 1189 50 36 115 33 49 .3 29 32 E K 1524 49 52 116 05 77 4 3 33 E K 1402 49 56 116 05 52 .3 30 152 T r e e No. P l a n z o n e E levat ion Latitude Longi tude D B H ( c m ) H ( m ) ~| 35 E K 1372 50 22 1 1 6 2 9 39 .9 1-—I 1 29 36 E K 1372 50 22 1 1 6 2 7 31.2 18 37 E K 1372 50 22 116 27 46 .2 27 38 E K 1342 50 21 116 25 54.1 4 0 39 E K 1311 50 33 116 31 42 .2 30 40 E K 1280 50 33 116 30 35.3 18 41 E K 1250 50 33 1 1 6 2 9 43 .7 21 4 2 E K 1250 50 34 116 27 4 5 32 4 3 E K 2012 50 30 116 26 49 .3 32 44 E K 1860 50 31 116 27 58.4 29 4 5 E K 1768 50 31 116 Z 7 47 .2 30 46 E K 1677 50 32 116 26 33.8 18 4 7 E K 1616 50 33 1 1 6 2 6 34 21 48 E K 1159 50 34 116 2 3 63.8 34 49 E K 1128 50 34 1 1 6 2 0 41.1 24 50 E K 1128 50 34 116 18 43 .2 23 51 E K 1433 50 2 3 116 06 67.1 38 52 E K 1280 50 21 116 04 47 .8 32 53 E K 1524 50 22 116 02 40 .4 29 54 E K 1463 50 52 116 14 40 .6 24 55 E K 1311 50 50 116 11 44 .5 30 56 E K 1311 50 46 116 07 39.4 29 57 E K 1433 50 44 116 03 39 .9 32 58 E K 1341 50 4 5 116 05 41 .4 26 59 E K 1738 50 30 116 17 64.8 40 60 E K 1616 50 31 116 16 69 .3 37 61 E K 1494 50 32 116 16 55.1 29 62 E K 1006 50 32 116 08 50.3 32 6 3 E K 1646 50 41 116 36 68.1 32 64 E K 1585 50 42 116 32 64 .5 30 6 5 E K 1463 50 4 3 1 1 6 2 9 43 .7 27 66 E K 1341 50 4 3 1 1 6 2 7 59.7 27 67 E K 1341 50 4 3 116 27 64.8 32 68 E K 1311 50 44 1 1 6 2 5 42 .4 23 69 E K 1280 50 44 116 24 49 .3 30 70 E K 1220 50 4 5 1 1 6 2 3 51.1 29 71 E K 1189 50 44 1 1 6 2 1 46 .7 34 72 E K 1494 50 38 116 26 41 .4 27 73 E K 1433 50 39 116 2 5 48 .5 32 74 E K 1402 50 38 1 1 6 2 0 43 .9 29 75 E K 1311 50 37 116 16 43 .2 29 76 E K 1128 50 37 116 14 505 30 77 E K 1067 50 35 116 12 41 .4 34 78 E K 1280 50 28 115 50 63 .2 37 79 E K 1189 50 29 115 52 51.6 37 80 E K 1250 50 57 116 09 53.6 38 81 E K 1250 50 57 116 10 50.3 31 82 E K 1250 50 58 116 11 62 .2 26 83 E K 1250 49 46 116 32 54.9 35 84 E K 1341 49 49 116 36 75.2 34 85 E K 1280 49 4 5 1 1 5 3 1 98.8 49 153 e e No. P l a n z o n e E levat ion Lat itude Longi tude D B H ( c m ) H ( m 86 E K 1677 49 48 115 36 57.7 38 87 E K 1616 4 9 49 115 36 104.6 4 4 88 E K 1372 49 37 116 34 40 .5 37 89 E K 1402 49 40 1 1 6 2 7 43 .7 23 90 E K 1250 4 9 39 116 24 50.8 35 91 E K 1372 49 34 116 11 44 .2 27 92 E K 1220 49 3 S 116 11 53.6 38 93 E K 1250 4 0 35 116 09 41 .2 32 94 E K 1555 49 22 116 06 63 .2 38 95 E K 1555 49 21 116 06 78.7 4 3 96 E K 1555 49 22 116 07 54.1 30 97 E K 1877 4 9 21 116 03 58.7 23 98 E K 1738 49 17 116 04 83.3 35 99 E K 1677 49 17 116 04 101.6 41 100 E K 1341 49 25 115 58 58.9 34 101 E K 1067 4 9 20 115 53 77.7 40 102 E K 1494 4 9 18 115 59 64 .3 41 103 E K 1189 49 18 115 55 54.9 27 104 E K 1128 49 20 115 53 54.1 37 105 E K 1524 49 26 116 08 46 .5 32 106 E K 1463 49 28 116 06 45 .7 29 107 E K 1402 49 29 116 06 62 35 108 E K 1341 49 30 116 04 65 .3 30 109 E K 1280 49 32 116 03 87.9 44 110 E K 1616 4 9 10 1 1 5 4 6 81 44 111 E K 1402 49 14 115 42 45.7 30 112 E K 1280 49 16 115 39 41 .9 32 113 E K 1524 49 47 1 1 5 2 1 0.2 46 114 E K 1707 4 9 4 6 1 1 5 2 6 84.6 41 115 E K 1677 4 9 55 115 17 74.7 46 116 E K 1585 49 55 115 17 78.2 40 117 E K 1524 49 54 115 18 74.2 40 118 E K 1463 49 53 115 17 49 .5 34 119 E K 1585 4 9 41 1 1 5 2 7 74.4 46 120 E K 1585 49 16 114 38 55.4 34 121 E K 1585 49 18 114 4 5 61 .5 34 122 E K 1494 49 17 114 46 85 .9 38 123 E K 1220 49 20 114 56 64 .8 34 124 E K 1524 49 01 1 1 5 2 8 49 .5 29 125 E K 1250 49 05 1 1 5 2 9 59.9 27 126 E K 1098 49 25 115 10 46 .2 23 127 E K 1646 49 04 114 36 77 .5 38 128 E K 1707 49 05 114 37 90.2 41 129 E K 1524 49 06 114 37 82 47 130 E K 1494 49 09 114 18 69.6 35 131 E K 1524 49 09 114 18 101.1 40 132 E K 1311 4 9 08 114 30 67 .3 30 5 3 - A E K 1189 50 21 115 58 40 .6 26 154 T r e e No. P l a n z o n e E levat ion Lat itude Longi tude D B H ( c m ) H (m) 1 MGR 610 54 2 MGR 610 54 3 MGR 610 54 4 MGR 610 54 5 MGR 610 54 6 MGR 610 54 7 MGR 610 54 8 MGR 610 54 9 MGR 610 54 10 MGR 808 53 11 MGR 793 53 12 MGR 747 53 13 MGR 732 53 14 MGR 762 53 15 MGR 777 53 16 MGR 686 53 17 MGR 808 53 18 MGR 793 53 19 MGR 762 53 20 MGR 762 53 21 MGR 732 53 22 MGR 763 53 23 MGR 763 53 24 MGR 763 53 25 MGR 763 53 26 MGR 763 53 27 MGR 763 53 28 MGR 762 53 29 MGR 762 53 30 MGR 762 53 31 MGR 762 53 32 MGR 823 53 33 MGR 976 53 34 MGR 1006 S3 35 MGR 976 53 36 MGR 884 53 37 MGR 945 53 38 MGR 915 53 39 MGR 915 53 40 MGR 869 53 41 MGR 869 53 42 MGR 1006 53 43 MGR 899 53 44 MGR 884 53 45 MGR 808 53 46 MGR 793 53 47 MGR 808 53 48 MGR 808 53 49 MGR 838 53 50 MGR 623 53 50 A MGR 823 53 04 122 02 66 41 03 122 04 54.6 32 04 122 05 65 40 05 122 05 61 32 05 122 08 70.2 37 04 122 07 61 34 06 122 04 55.9 36 06 122 03 61 40 06 122 05 72.4 38 52 121 56 60.9 34 52 121 57 76.2 40 53 121 59 48.3 35 59 122 08 57.2 38 58 122 08 76.2 40 57 122 08 71.1 37 56 122 06 73.7 40 56 122 04 63.5 38 56 122 02 92.7 41 55 122 00 50.8 34 54 121 59 79.5 38 54 122 02 79.2 37 54 122 04 60.9 40 53 122 06 68.6 43 53 122 08 69.9 43 53 122 12 54.6 34 53 12Z 16 41.1 29 53 122 18 673 34 52 122 16 47.8 30 5Z 122 15 50.8 34 50 122 14 69.9 34 48 122 14 59.2 32 46 122 15 41.4 24 43 122 08 67.3 35 43 122 09 55.1 35 44 122 10 67.8 37 45 122 11 71.1 34 44 122 04 66 38 45 122 06 66.5 43 46 122 09 65 38 45 122 11 51.3 26 45 122 21 71.4 37 47 122 06 52.6 32 46 122 06 63.8 38 46 122 07 69.3 40 49 121 55 686 38 51 121 59 40.6 29 51 122 00 50.8 34 51 122 02 62 34 51 122 04 49.8 34 50 122 05 59.4 35 50 122 05 88.4 43 155 ;e No. P l a n z o n e E levat ion Latitude Longi tude D B H ( c m ) H ( m ) 51 M G R 869 S 3 50 122 07 58.4 37 52 M G R 838 53 4 9 122 08 63 .5 41 53 M G R 838 53 4 7 122 09 57.9 34 54 M G R 869 53 47 122 09 72.6 4 0 55 M G R 915 53 44 122 22 69 .9 36 56 M G R 1052 53 4 3 122 03 80 40 57 M G R 1082 53 4 2 122 02 6 3 31 58 M G R 1006 53 4 2 122 01 66 37 59 M G R 1006 53 4 0 122 01 94.7 4 0 60 M G R 960 53 37 122 02 62 .2 34 61 M G R 945 53 34 121 59 63 37 62 M G R 945 53 36 122 01 74.4 38 6 3 M G R 960 53 37 122 02 88.9 4 3 64 M G R 976 53 38 122 03 80 4 0 6 5 M G R 976 53 38 122 05 55.9 34 66 M G R 953 53 39 122 07 58.9 34 67 M G R 915 53 4 0 122 08 67.6 38 68 M G R 854 53 40 122 11 87.1 4 3 69 M G R 1006 53 28 121 56 71 .9 32 70 M G R 1037 53 32 121 59 55.9 34 71 M G R 1006 53 32 122 00 77.5 40 72 M G R $60 53 34 122 03 71.1 40 73 M G R 915 53 34 122 04 69 .3 35 74 M G R 915 53 36 122 07 48 .8 35 75 M G R 915 53 37 122 08 58.4 38 76 M G R 915 53 37 122 09 73.7 41 77 M G R 915 53 38 122 10 75.1 40 78 M G R 945 53 38 122 10 74.9 40 79 M G R 960 53 39 122 11 61 .5 37 80 M G R 869 53 4 3 12Z 15 45 .2 35 81 M G R 1143 53 4 0 122 24 45 .7 34 82 M G R 1067 53 39 122 25 56.9 30 83 M G R 1037 53 39 122 26 61.7 41 84 M G R 915 53 38 122 32 43.7 27 85 M G R 915 53 37 122 35 36.8 30 86 M G R 869 53 37 122 3 S 68 .9 34 87 M G R 930 53 37 122 31 57.2 26 88 M G R 991 53 36 122 30 49 .5 30 89 M G R 1021 53 35 122 29 54.6 38 90 N C H 1098 53 34 122 30 56.6 32 91 N C H 1006 53 34 122 30 56.9 34 92 N C H 991 53 33 122 31 66 34 93 N C H 930 53 32 122 31 59 .9 32 94 N C H 915 53 18 122 05 56.7 37 95 Q L 915 53 19 122 04 54.9 29 96 Q L 915 53 20 122 05 57.4 37 97 Q L 976 53 26 121 56 73.7 40 98 Q L 1021 53 26 121 57 70.1 38 99 M G R 960 53 28 121 58 58.7 37 100 M G R 976 53 27 121 59 56.6 34 101 M G R 976 53 27 121 59 66 40 156 ree No. P l a n z o n e E levat ion Latitude Longi tude D B H ( c m ) H ( m ) 102 E L 976 53 26 121 59 59.7 32 103 Q L 976 53 26 122 01 61 .7 37 104 Q L 1021 53 2 5 ' 122 01 71.1 38 105 E L 1006 53 24 122 04 6 5 34 106 Q L 991 53 2 3 122 04 54.6 34 107 E L 1006 53 2 3 122 04 72.1 37 108 Q L 1021 53 2 3 122 03 59.7 0 109 E L 1006 53 2 2 122 04 49 .5 0 110 Q L 1006 53 21 122 05 72.1 0 111 Q L 1037 53 21 122 05 69.1 0 112 N C H 960 53 06 122 18 59.9 0 113 N C H 884 53 09 122 11 61 .7 0 114 N C H 960 53 11 122 10 37.6 0 115 N C H 1067 53 12 122 08 40 .6 0 116 N C H 976 53 15 122 05 71.9 0 111 N C H 960 53 16 122 05 70.1 0 118 Q L 976 53 19 122 05 77 .5 0 119 N C H 1006 53 18 122 07 107.4 0 120 N C H 1052 53 18 122 07 62 .2 0 121 N C H 1143 53 18 122 08 56.4 0 122 N C H 1159 53 18 122 11 90.4 0 123 N C H 991 53 18 122 22 55.6 0 124 N C H 1087 53 15 122 17 71.1 0 125 N C H 1037 53 17 122 16 76.5 0 126 N C H 1067 53 17 122 16 58.9 0 127 N C H 1113 53 18 122 13 67.6 0 128 N C H 1189 53 19 122 12 80.3 0 129 N C H 1143 53 19 122 13 75.2 0 130 N C H 1006 53 30 122 11 64 .8 0 131 N C H 1052 53 28 122 13 81.5 0 132 N C H 1006 53 27 122 13 90 .9 0 133 N C H 1006 53 2 5 122 13 99 .3 0 134 N C H 1006 53 2 5 12214 81.3 0 135 N C H 945 53 22 122 15 72.6 41 136 h t C H 960 53 18 122 22 50.3 34 137 N C H 930 53 18 122 2 3 66 38 138 N C H 899 53 18 122 24 61 38 139 N C H 777 53 18 122 25 43 .2 30 140 c T 1082 52 4 7 122 01 66 27 141 c T 1113 52 4 7 122 02 67 .3 27 1 4 1 A c T 1113 52 48 122 02 60 .2 29 142 C T 1159 52 48 122 03 53.3 27 143 C T 1113 52 49 122 05 4 5 29 144 C T 1052 52 50 122 07 73.2 32 145 c T 1052 52 51 122 08 53.3 30 146 C T 1037 52 53 122 09 52.8 30 147 C T 945 52 56 122 11 59.4 37 148 c T 1006 52 57 122 12 68 41 149 C T 1113 52 53 121 44 59.4 41 150 c T 1159 52 54 121 4 5 4 3 9 32 151 c T 1159 52 54 12146 63 .2 . 32 157 T r e e No. P l a n z o n e E levat ion Latitude Longi tude D B H ( c m ) H (m) 152 c T 1159 52 56 121 48 69 .9 35 153 c T 1143 52 58 121 50 49 .3 29 154 c T 1159 52 58 121 51 55,1 37 155 c T 1159 52 59 121 51 67,6 38 156 c T 1067 52 59 121 53 47 34 157 c T 1067 52 59 121 55 54.1 37 158 c T 1067 52 59 121 56 54.6 4 0 159 c T 1006 53 00 121 59 56.4 34 160 c T 930 53 01 122 00 64 .3 41 161 c T 854 53 01 122 04 57 7 34 162 Q L 976 53 13 121 26 58.4 35 163 Q L 1128 53 16 121 30 744 38 164 Q L 1159 53 15 121 30 81.8 38 165 Q L 1159 53 15 121 30 84.6 37 166 Q L 1204 53 15 121 30 64 .3 34 167 Q L 1220 53 14 121 30 41 .9 24 168 Q L 1159 53 13 121 29 43 .2 21 169 Q L 1098 53 12 121 27 32.5 23 170 E L 1098 53 12 121 27 59.2 40 171 Q L 1159 53 11 121 28 56.6 37 172 E L 1143 53 04 121 4 5 55.1 26 173 Q L 1128 53 04 121 47 51.8 29 174 Q L 1021 53 04 121 57 66.8 40 175 c T 915 53 02 122 17 58.4 30 158 APPENDIX 2: WOOD DENSITY MEASUREMENT: COMPARISON OF X-RAY, PHOTOMETRIC, AND MORPHOMETRIC METHODS 159 Wood Density Measurement: Comparison of X-ray, Photometric, and Morphometric Methods Milosh Ivkovich Department of Forest Sciences, The University of British Columbia, 270-2357 Main Mall, Vancouver, B.C., V6T1Z4 E-mail: ivkovich@unixg. ubc. ca Mathew P. Koshy Department of Forest Sciences, The University of British Columbia, 270-2357 Main Mall, Vancouver, B.C., V6T1Z4 E-mail: koshy@unixg.ubc.ca ABSTRACT For evaluating genetic control and estimating genetic worth of the parents, large scale evaluation of wood density from parents and progeny trials are essential. While pilodyn and gravimetric methods can give estimates of wood density, they are not capable of describing the density gradient within the trees and within the rings. X-ray measurements have been used to assess this density gradient. Recently, various other methods using computerized image analysis systems are being used. These methods generally are of two types: photometric and morphometric. In this paper we show that the use of photometric measurement based on reflected light from the polished surface of spruce wood is not an adequate substitute for the X-ray densitometry, for the purpose of comparison among trees. While an universal calibration of reflected light intensity according to x-ray transmission seemed impossible, within single rings, the correlation of the reflected light intensity and wood density determined by x-ray measurement was typically high. Consequently this enables one to determine the relative amount of late-wood within each ring based on the reflected light intensity profile. These relative amounts can indeed be compared across different trees. Also, we emphasize the usefulness of the morphometric analysis of the microscopic wood sections for the use in genetic studies: - Correlation between morphometric and x-ray density measurement is typically high (R2>0.87); - Morphometry provides profiles of cell size and cell wall thickness which are more informative than density profiles; - Various statistics can be used to describe distributions of cell wall thickness and lumen size across the rings; - The late-wood region can be precisely determined considering two characteristics: cell wall thickening and radial enlargement of tracheids. INTRODUCTION Assessment of wood quality traits, especially wood density, is of primary importance in tree breeding. Various methods for collection and evaluation of wood quality traits on small samples were comprehensively described by Kellog et al. (1982), and summarized by Zobel and Jett (1995). Pilodyn needle penetration is used for density assessment in the field. Density of cross-section disks can be determined by the volumetric method given in TAPPI standards (T258 om-94). For small pieces the "maximum moisture content" method gives the best results (Smith 1954). While these methods could assess wood density of whole samples, they are not capable of describing the density gradient within trees and within rings. To obtain within ring density profiles X-ray densitometry has been used (Parker and Jozsa 1973, Josza and Myronuk 1986). This method estimates density of thin wood samples based on their transmitance of X-ray beams. Although density values obtained by the X-ray measurements (or any other indirect measurements) can be biased either upwards or downwards, the relative whole-ring values and the values of within-ring profiles for different trees are useful for wood quality comparisons. Recently, various other methods using computerized image 160 analysis systems are being used. These methods generally are of two types: photometric and morphometric. In the first method, gray scale of images captured by a reflected light scanner is used to estimate density values. Calibration based upon reference values from X-ray densitometry is often used in this method. In the second, morphometric method, cell wall area is measured from wood cross sections, and density is estimated using a constant value of specific gravity of cell wall substance and cell wall to lumen area ratio. METHODS AND MATERIALS Wood samples collected from several 25 year old interior spruce progeny tests in British Columbia were used in this study. A large 10 mm increment borer was used to extract cores at the approximately breast height. Each core was split into two halves: one half was used for X-ray sample preparation and the other for the gravimetric and morphometric analyses. Four outermost rings from each core were used for comparison of density estimates obtained by different methods. X-ray Densitometry Sample cores were air dried, and sawn to uniform 1.57mm thickness. These strips were extracted with alcohol-benzene solution (2:1) and water prior to measurement. X-ray beam was 0.25 x 1.00 mm, operating at 15 to 20 KV and about 2 mA. Density profiles were represented with one hundred points across each ring. Graphical and digital output, as well as summary statistics were obtained. The measurements were done using a direct-reading X-ray densitometer at the FORTNTEK Corp. Vancouver (Jozsa and Myronuk 1986). Photometry Wood surface was first roughly sanded on a portable machine and then polished by hand using fine (280 grit) sand paper. A high resolution (up to 1200 pixels per inch) optical scanner was used to obtain images of polished surfaces. Images were analysed using WinDendro 6.0.5.™ computer package. Pixel based measurements of light intensity were obtained for 30 cores, and 4 outermost rings from each core, were The method utilizes a transmitted light microscope and an image analysis system to measure cell wall and lumen dimensions across the annual rings. (Jagels and Dyer 1983, Park and Telewski 1993) This study looks into these methods, compares the density estimates, discusses the merits and demerits of each system, and their suitability in tree breeding programs. used for comparison with the X-ray transmission data. The pixel size is related to the resolution at which the image was scanned, and it was adjusted for each core by varying the resolution according to the core size. This was done until within each core there were approximately 100 pixels across one ring. Relative location of these 100 pixels corresponded to the location of 100 points obtained by the direct x-ray densitometry. This made a comparison of these two techniques feasible. Morphometry The increment cores were divided into two halves prior to measurements, so that one half could be used for the X-ray, and the other for the morphometric analysis. The microtome sample preparation of four rings long and 18mm thick sections, was done using a sliding microtome. Sections were stained with aniline safranin, and mounted on microscopic slides with Cytoseal™. Monochrome images were captured with a Hitachi™ video camera and SigmaScan™ system was used for the image processing. Frame size of the camera vas ? x ? pixels, with a resolution of ? pixels/inch(cm). The average pixel intensity was recorded along several 11 pixels (approx. 4.7 mm) wide lines across the annual rings. The lines were drawn in the radial direction that was perpendicular to cell walls. A threshold value and a transformation were chosen to separate low and high intensity pixels along the lines, such that cell lumen was assigned value of 0, and the cell wall value of 1. The column of 0's and l's was averaged, which yielded the cell wall to lumen ratio for each cell. The values of double wall thickness and lumen radial diameter for each cell were also obtained. 161 RESULTS AND DISCUSSION: Photometry vs. X-ray densitometry The relationship between photometric and X-ray densitometric methods were explored firstly for the data from each of all 120 rings from all 30 cores. The relationship was approximately linear with a high average correlation. The average R2 for 120 rings that were examined was 0.87 ranging from 0.59 to 0.96. Although the light intensity and the X-ray density were highly correlated within single rings, the relationship could not be generalized across different rings. The regression coefficients and intercepts were characteristically inconstant. Coefficient of variation of these regression coefficients around their mean (a mean regression coefficient that could have been used for the light intensity calibration) was 22% , with 95% confidence limits from 0.0047 to 0.0052. Consequently, density values obtained from calibration of light intensity profiles (gray levels) was not found precise enough for comparison among rings from different cores. At the same time, coefficient of variation in the linear regression coefficients for the rings within individual cores was around 16%. One of the reasons for more variation when rings across cores are compared is probably because there were resolution differences in the images of different cores. An universal calibration was not feasible, neither within nor among cores, for our purpose of wood density comparison among trees from different families. The variation in the relationship between light and x-ray data is perhaps caused by the variation in the late-wood coloration among rings from tree to tree and from year to year within a core. However, as it was shown by Clauson and Wilson (1991) photometry can be more useful in determining the within-ring density distribution, i.e., the location of early- and late-wood boundary, as determined by low and high density values and given as the percentage of high density fraction. Our results of comparison between ring profiles obtained from the two methods show that reflected light gives a good estimation of late- and early-wood proportion within growth rings, since within-ring correlation of two types of measurement was high (Figure 1.). The relative amount of late-wood is an interesting parameter from the standpoint of tree improvement. For example, provenances moved north usually grow longer in the season than the local material and may produce a higher proportion of late-wood. Potentially the high interdependence between late-wood percentage and wood density could give a tree breeder a general idea about the changes in wood quality caused by the provenance manipulation. Figure 1. Typical relationship between the photometric and X-ray density measurements if an universal calibration is used. In this case reflective light values are overestimated, especially in the late-wood zone. 162 Morphometry vs. X-ray Densitometry Morphometric analysis can be suitable not only for density determination, but also for comparison of the cell morphology among different tree families. Double wall thickness to lumen ratio, or "Runkel ratio", of wood fibres is highly correlated to the to burst and tensile strength of paper sheets. Tensile index is considered extremely important because the new high-speed paper rolling machines require a high tensile strength. Runkel ratio is thus considered one of the most important single predictor of pulp properties (Zobel and Jett 1995), and it can also be easily translated into number of fibres/cm3, or number of fibres/g, etc. This, and other indices related to the tracheid dimensions can be found by measuring cell lumen diameters and double wall thickness on projected microtome cross-sections, mounted on the microscopic slides, by using an image analysis system. An example of the graphical output that shows the trend in cell lumen and double wall thickness variation within an annual increment is given in Figure 2. Our preliminary results indicate that the morphometric measurement has several advantages over X-ray densitometry. Most X-ray density measurement systems introduce a slight smoothing in the density, due to internal reflections and limitations of detector sensitivity (Varem-Sanders and Campbell 1996). The resolution of the morphometric measurements is usually higher and this type of smoothing is minimized. Nevertheless, the correlation between morphometric and x-ray density based on measurements of 30 annual rings was typically high, with an average R2 of 0.89, and ranging from 0.77 to 0.92 (Figure 3). Beside the precision of measurement there are several other advantages to morphometric analysis. Morphometry provides profiles of cell size and cell wall thickness which are more informative than density profiles (Zobel and van Buijtenen 1989). Mean, maximum, and minimum cell size, size of the zones with cell dimension grater or less than a specified value, number of cells in a radial file, and various statistics describing distributions of cell wall thickness and lumen size across the rings can be obtained (Vaganovl990). Based on that information, the formation dynamics of late-wood can be precisely defined using the Larson's (1969) definition which considers two independent events: cell wall thickening and cessation of radial enlargement of tracheids during growing season. These parameters can be compared among trees and different tree families and genetic differences determined. By and large, morphometric method, if used with a proper calibration, can be more than an adequate replacement for the x-ray densitometry. However, the morphometric method has so far been used mostly in dendrochronological studies (Vaganov 1990, Park and Telewski 1993), but there are only a few genetic studies using this method. Information on inheritance of cell morphology in forest trees is sparse (Zobel and van Bujtenen 1989). New technology makes this type of studies quite feasible. 40 35 30 25 zL 20 15 10 c \ i e > T i - - ^ - i o c D r « - c o c o a > C E L L N U M B E R L A T E W O O D T O E A R L Y W O O D L U M E N - C E L L W A L L T H I C K N E S S Figure 2. Typical patterns of variation in the cell lumen diameter and double wall thickness across an annual increment obtained by image analysis. 163 Figure 3. Typical relationship between the morphometric and X-ray density measurements. REFERENCES: Jagels, R.; Dyer, M.V. 1983: Morphometric analysis applied to wood structure I: cross-sectional cell shape and area change in red spruce. Wood Fiber Sci. 15:376-386. Josza, L.A.; Myronuk, R.M. 1986: Direct reading X-ray densitometer. FORTNTEK Can. Corp., Vancouver, British Columbia. Intern. Rep. Kellogg, R.M.; van Buijtenen, J.P.; Hatton, J.V.; Wahlgren, H.E.; Brink, D.L. 1982: New methods of measuring wood and fiber properties in small samples. TAPPI Res. and Dev. Div. Conf. Asheville, NC, 1-2. Larson, P.R. 1969: Wood formation and concept of wood quality. Yale University, School of Forestry, Bulletin No. 74. Park, W.K.; Telewski, F.W. 1993: Measuring Maximum latewood density by image analysis at the cellular level. Wood Fiber Sci. 25(4): 326-332. Parker, M.L.; Jozsa, L.A. 1973: X-ray scanning machine for tree ring width and density analyses. Wood Fiber 5:192-197. Smith, D.M. 1954: Maximum moisture content method for determining specific gravity of small wood samples. US Dep. Agric. For. Serv. Rep. No 2014. Madison. WI. T.A.P.P.I. 1994: Basic density and moisture content determination of pulpwood. T258 om-94. TAPPI Standards. Tech. assoc. Pulp and Paper Industry. N.Y. Vaganov, E.A. 1990: The tracheidogram method in tree ring analysis and its application, pp. 63-76. In: Cook, E.R.; Kairiukstis, L.A. (Ed.) Methods of Dendrochronology: Applications in the Environmental Science. Kluwer Academic Publishers. Dordrecht. 394p. Varem-Sanders, T.M.L.; Campbell, I.D. 1996. DendroScan: A tree ring width and density measurement system. Special Report 10. Canadian Forest Service, Northern Forestry Centre. Regent Insruments Inc. 1995: WinDendro V6.0. User's Guide. Regent Insruments Inc. Quebec, Qc. Zobel, B.J.; Jett, J.B. 1995: Genetics of Wood Production. Springer-Verlag, New York. 337p. Zobel, J.B. and van Buijtenen, J.P. 1989: Wood Variation: Its Causes and Control. Springer-Verlag. NY. 362p. 164 APPENDIX 3: FAMILY PERFORMANCES ON DIFFERENT SITES 165 12 J Figure A3.la Plots of family performance in height (H) (top) and diameter (D) (bottom) at two site locations Jumbo Creek (JC) and Perry Creek (PC) within the EK region. Family numbers are scattered relative to the line of equal performance. 166 102 2 3 A V R W at site J C (mm) 0.30 0.35 0.40 0.45 R D at site J C Figure A3.lb Plots of family performance in average ring wdth (AVRW) (top) and relative density (RD) (bottom) at two site locations Jumbo Creek (JC) and Perry Creek (PC) within the EK region. Family numbers are scattered relative to the line of equal performance. 167 Figure A3.1c Plot of family performance in average latewoo percentage (AVLW%) at two site locations Jumbo Creek (JC) and Perry Creek (PC) within the EK region. Family numbers are scattered relative to the line of equal performance. 168 Figure A3.2a Plots of family performance in height (H) (top) and diameter (D) (bottom) at two site locations Prince George (PG) and Quesnel (Q) within the PG region. Family numbers are scattered relative to the line of equal performance. 169 3 18 66 20 . 22 24 26 28 30 32 34 AVLWP at site PG Figure A3.2b Plots of family performance in average ring width (AVRW) (top) and average letewood percentage (AVLWP) (bottom) at two site locations Prince George (PG) and Quesnel (Q) within the PG region. Family numbers are scattered relative to the line of equal performance. 170 Figure A3.2cPlots of family performance in relative density (RD) at two site locations Prince George (PG) and Quesnel (Q) within the PG region. Family numbers are scattered relative to the line of equal performance. 171 APPENDIX 4: MEANS AND RANGES OF FAMILY MEANS FOR RING WIDTH AND LATEWOOD PERCENTAGE FOR NINE GROWING SEASONS 172 Table A4 Means and ranges of family means for RW (mm), and LW%, for 9 growing seasons. Site 1987 1988 1989 1990 1991 1992 1993 1994 1995 EK RR 2.0 (1.2-3.1) 38 (27-50) 3.6 (2.1-5.0) 34 (20-53) 4.1 (2.6-5.5) 22 (10-36) 3.6 (2.2-5.1) 26 (13-41) 3.4 (2.1-4.8) 22 (12-39) 2.7 (1.7-3.8) 22 (13-29) 3.7 (2.0-5.8) 25 (15-38) 3.4 (1.9-4.8) 21 (14-33) 2.3 (1.1-3.99) 29 (21-40) JC 2.5 (1.3-3.2) 35 (21-57) 2.5 (1.5-3.4) 30 (18-46) 2.5 (1.3-3.6) 31 (18-54) 2.6 (1.5-3.5) 30 (15-46) 2.6 (1.6-3.7) 28 (16-44) 2.6 (1.5-3.8) 29 (15-43) 3.0 (1.5-4.3) 29 (15-47) 2.8 (1.7-4.2) 22 (13-40) 2.5 (1.5-3.6) 26 (15-46) PC 3.1 (2.1-4.3) 30 0-52) 2.9 (1.9-4.3) 21 (11-37) 2.9 (1.9-4.6) 20 (8-34) 3.4 (2.1-5.2) 21 (9-37) 4.0 (2.7-5.5) 22 (10-39) 3.8 (2.6-5.6) 19 (11-33) 3.8 (2.6-5.6) 21 (12-31) 3.8 (2.5-5.1) 16 (6-29) 2.4 (1.1-3.7) 33 (22-46) PG RR 3.1 (2.3-3.8) 30 (21-40) 3.9 (2.7-5.0) 25 (19-31) 3.7 (2.4-4.8) 25 (13-39) 3.3 (1.9-4.3) 28 (20-40) 2.7 (1.6-3.5) 26 (19-34) 1.7 (1.2-2.4) 25 (18-35) 2.1 (1.2-3.0) 36 (24-46) 1.9 (1.1-2.7) 31 (22-39) 1.6 (0.9-2.5) 29 (20-39) Q 2.5 (1.4-3.6) 33 (24-39) 3.4 (2.1-4.5) 34 (26-42) 4.0 (2.2-5.1) 25 (18-36) 3.3 (1.8-4.6) 32 (25-42) 4.0 (2.4-5.6) 25 (19-33) 2.5 (1.5-3.4) 22 (16-34) 3.8 (1.9-5.5) 32 (21-44) 3.4 (1.7-4.7) 27 (20-35) 2.8 (1.5-4.0) 25 (17-34) 173 APPENDIX 5: TABLES OF VARIANCE COMPONENTS AND HERITABILITIES FOR INDIVIDUAL SITES 174 PS1 ?? CL) o 3 _ O —1 U J _ <u 13 .S > E o o o •c u JS ^ ^ OH 8 ^ oo C O — ^4 0 0 *7« 0 0 O = O H o "Ea .HfJ § C o ^ cL o £ CJ CD +J o s CO X> c o oo 15 c o too L H o CD 1 ° S g > -c 0 0 C- l & g <u 'lo CL) CD c o 5 DC CD 5 CD Q DC < CO O CO O CO o c O ^_ ! •sT O CO •sr d co •sr d ° : CM t: •sr : o •>* t>- o o o o o o o d d d d t r-CM 'oo •sr O ) <o" CO CO o — . GO O 'c- ' C M l»- CO CO CM *- d d d CD %- p . CM o " •sr <J> LO io 00 1^ O o o o . o d d d d ! o CO ! 00 s in CO CM CD •sr • ^ d c co c O) r-- CM ST o CO CM CN o o O o o d d d d CD d CO •sr d CM co o LO CM •sT co cd 2. r-•sr 1 o •sr o d CD O) 00 ^ •* co •ST CO co m o d L O •ST 00 •sr o cn •sT CM 00 O ' co c CD ' co 1 o o> 1^ T— I CO d d o o o I O o ' L O CO o CD o o 3.11) L O o L O d d d d d : CM c CM ' L O c CM ay cn 00 CO CO CO d d o LO 00 cn CO co cb : c CM f~ O CM CM CO CO o CO CD CM L O 00 d 'oo 'CD CM L O o d 2.69" d 0.93" 0.493' 5.56 c CN LO O CM CD o d 81.4' 805.1 0.18" 0.03"' 0.00 0.08" 0.66 CM L O •sr cn CO CM CO R(S) SxF Err. CM cn O CN T— o T CM CN CO T- '—  T— • ! +i +i +i • +i - i -I +i i o CN o c n CO O) CD l O in N . in •sr CO in LO CD cn CO CN o ^ — CM CO o i— •i 1 +i +i i +i r - i +i • o CD o CO CN co in in CO LO CM m CO CO •sr '— CM •sr i — T— CM T— - j +i i +i i : +i i +i +i CM co •sr CO m CM CO CO •sr CD •sr CD CO o •sr CN •<r +l +1 +i , +i +i LO CO o •sr CO CO CO co CO •sr in •sr CD o •sr •sr CN •sr 1 1 +i t +i i , +i i +i LO CN CM CO CO 00 CM •sr LO 00 CO m T— CO T— «— +i 1 1 1 +i +i i i i +i LO in m CO CD •sr CO 00 CO o CO 00 CO CM •sr T o T— +i +i 1 +i i • +i t +i t CO CO cn CD CD •«r cn 00 LO •sr •sr CO CD O) CD CO CN T— CM CO T T— T— +i I +i i +i • i +i i +i CD CO •sr CO 00 CO CM •cr cn •sr CM •sr LO T_ CN CO T CN i — T— 1 +i i i i -1 +i i i i cn CO 00 CN CO •sr •sr 00 IO LO CN 00 CO T— CM CM •sr T— o T T +i +i +i +i • +i +i • • i CO •sr CD CD •q; cn •sr cn in CO •sr •sr •q- o oo •sr •sr ^_ i r T— CM CO T— r— CN T +i +i i i +i +i +i +i 00 h~ cn CO cn CM co co CO CO CO 00 CO •sr T— CO T— 1 i • +i +i 1 i i +i •a- CO cn CO •sr CO CM o 00 CM CO T_ T— CN t r CO T— T— T T +i +i • +i +i +i +i i -'. !-LO CO CO o r - CO CO in LO cn LO •sr •sr in in CO o CO ^_ m T— CM •sr T— T— +i +i i +i 1 +i +i • +i CO CD CO CD cn o CO LO •sr •sr •sr LO CM T— 1 +i i i 1 i +i • • • LO r~ co CO LO 00 CO •sr CO CN CO T— T— +i +i • • +i +i +i i • • o LO CO o m CO •=f CO •sr •sr •sr CO CO 00 in <J> CO T— CN co O +i 1 +i i +i +i i +i • +i co m •sr r~ m CM CO CO CO CO •sr LO CO CM in CN CO CN CN T- CN T— +i +i +i i i +i +i +i , 00 O) in CD CO CM in •<r CO in •sr r~- CD m T— CM x— o +i +i i i i - i +i i i i CO o o 00 CO •sr CD CO co o CM 00 CO T— CM •sr T— O +i +i i +i i -1 +i • +i i CO CD o •sr in co LO o> cn LO CO •sr CO ^ _ CM f— 1 +i i • i • +i i • i cn LO in o CN T— 1 +i i i i i +i i • r~ in _1 J DC O O _j i DC o O < DC LU —> CL < DC LU —7 CL o d £ cu •a Q-o .5 S to c | S .2 oo ° c l l . S •S •£ o o ca 4= Z. o c 'S s= .g> C 0) g> M_ to o CO E rem vel; ro JH Q. • in >. o TO o ke Us 13 • a ca c "co CD o o t QJ an ropo evi ropo <U TJ Q. V) 3 •o "D O CD O ctu ikel CO CO c: CS (/) o CQ 1 o o < l_ LL CD CD ien o •4— Fs sq £ c CD ra 03 o o " c 13 CD to "to a * CD s o Q. ~ CD CO S 3 > c/5 o .2 B CD 4> X T 3 C cn X •3 * cj O o .2 I - X 3 _ s-u S3 o £ vo ^ ii M !3 c w 'C c u .2 .2 > CL> O O CL> &. a o O O O H U H2 s c o CL, s o o CD o c ca "5 > <! CD 3 a CD X i cd -a 1-o fc <s is < u. c O O O lb C D C N C N ? S i _l L L O O d c X T O O ci 'xr 0 0 d Is d S or 1 C D 0 0 C O r-co O Lo" 0 0 d 0 C O <2 S i 3 _j ! C O o> C N xr 0 0 C N JJ d S DWIw c O O O O a C O C O d C N £ 0 0 d " 0 ^ X T CM 0 d 0 " > a: *co C D X T c XT C O C N CO °>" 'co C O C N _ i 'co CM C O C O C N d X T d d S . DWtw 0 0 0 d L O d *co CO 0 d Is d S -or C O C O C N C N O C O C N C O " S i a> _j t C O C D co c L O X f d 'cn r~-d C N j f d S -DWew 0 0 0 0 0 d C O oo" C N CO 0 d 0 ~ ? C N O <? d S L O r -c X T L O c CO* CD O x- 2. «? 5 ^ 2. - J 1 L O c C O L O d CO ° ^ d & l l a * -( O csi C M C O O Lo~ O 0 d 0 l l 0s-L U 1 C M X T C O d 11 d S -or ! O XT oi c h -co C N r: ' 0 C N i c O O O ci c co C D d C N P C O CD O d O ~"" C D co 0 0 0 d 0 < w 0 c 0 L O co C N * C N r>-to 0 0 w 0 C O C N ci C O C O d : co L O d d S w 0 ! L O C O C O c r» C O d : C N C D d C N t 5 s. 2 O t C N C O c C O C O L O ^ -oi S. O Lo" O 0 d 0 I ! C O C N O d d S O Lo~ O 0 d 0 L L Q 1 0 O ) C O CJ) C O C O C N C N cn C O O ) L O C O C O o 1 0 C O C N CD C N C O a: a . o C O r--d o C D C D C N T-1 +i 1 +i C O L O C O C O C D L O C N '— C N 1— ' +i ' +i O cn C N C O C O L O zi a: +1 co •<1-C D C O C N C N • 1 +i +i • +i C N C O CO C N L O CO xr C N C D 1 T— -: 1 1 +i C O C O C O CO L O cn C O —^ C N t— ' +i +i ' +i C D C O C O C N L O xr L O C D C O CNI T • ! • +i • +i L O L O C D C O L O 1 0 xr L O C N C N 1 +i +i O co L O xr C N 1 0 CO C N C N 1 +i +i +i +i C N 0 xr C N xr C O L O C N C N 1 1 +i • 1 +i O CO L O C O xr C O C O x- C N T +i • +i 1 1 +i 0 r~- co C N L O xr xr C O L O xr C N C N T— • ! 1 +i +i 1 +i C O N . C D O xr 1 0 in L O r~- L O C N T T— - 1 1 +i 1 +i C O C D C O xr CO C D 1 0 1 0 c 8 u= c O) CO 0 c c •fa" a) to CO 0 = ai > ,— 0 d ca CD 0 c ca CO '> CO CO CO TJ CO sz T 3 O c O CD ra ai a. a> ^= 0 c SO c CO CO > ra cn to CO L L § C TO ific an< C 0 Ul L P CO c * O) CO cu _ <D V) • s s l iS co d "te •o ffi ra ai § g •o O C I D > o oi CO ~ _ w aj cn o M I - L L I 2 a> 2 2 g eS •- <° I"2"- c^ - g> ra CO ! t -a CD ca g CD 5 I GO s — ' = s X , c a < 2 X -t-» c 5 -2 o .S 6? J—i 'w c2 -tt CD £ O c c o D. VC CD °5 i - ca O > O o CD oo. o CD 4 - 3 I n O C • c " CD — g S o o CD O -t-» e CO cd > ed a: J c O C O c C D C N o o to* o 'co C O C O C J ) d d d d. d C O g e C O c C O c O Lo" c O m I T ) 5 C O o o C O C M T C N o d o d D o d d d or C D X T C O C D cn C D o o d I T ? o d. ' C N m C N xr d C O 5 co or co LU or oo o or L L o o q x ^ x -£ — CM CO o CN d d CO CN xr o xr o o d S i o O o xr d o cn o m o to" o CN o o o o o CN O d o d. d d d ~ d C O co o m o o d o . : CD i n co in CN 'f-m x J d d d d c o O o c CO r -c CJ) o o CM o_ o d d d d c CN CO c i n CN CO c O o d if> o d_ d CN I CO cn c CN ! xr CD c O O x - d CN d CD CM CN * o o> xr o xr in' d 0.00 c r -co d 'm c O o d i n CO CT> oo o o Lo" q 'co CN CO CD d o CN CO o o c r -CN CN o o o o d_ o o o d d d d ! CO xr "a> co c CN d d d 115.4" c O CO c CJ) d CN cT CM c xr i n xr c CN O c CN O c O o q c xr o d d d d x - i n CJ) CO CJ) CO co or u. U. c55 CO cn xr CJ) o r-. co CD m xr iri co cn i n C O d co d oo CN xr xr in o d xr m d LL. o CO CO r -CN CN T — T — +i +i +i +i CD xr m CD CO , cq xr CO xr oo CO xr CO . — CM CN T — T — T — +i +i +i +i +i CO r~ CJ) r~ m m CN xr CD xr xr i n xr CN +i +l o CD CN +i in +i xr CD +l CO +l CO +l CD i n +i m co 00 CN cn i n .30+. .51±. +i i n xr .47±. CD CN r~-CN .10±. ,52±. +i cn .47±. m CN CN CO CD .03±. .38±. .05±. .39±. ALL or ° ALL RR N *~ +l i n +l co xr +i o xr +i co +l m co co > 0) o d ra 8 c ^ ro to > co 55 co -o ^ 3 c ~ CO CO O ( CO -"co - S o £ CO CO ro ._ ti S ro ro ~u co c ^ O J , Q_ (0 "O . a) E "TO 5 ^ £ tj « CO ro <2 i= CO §2 d S ra a3 I °-> c CO CO •D O T 3 S o c O CD = CO "D J£ C= — 3 £ o c= ™ o —' <4— 5) ° CJ o I ro g to 5 c CO CO •_ iH ° 3 ro ro co to 'io i s ct: 1 cr: 5 5 CC to o o d o o o r-CD o cn •sr i r i d c O O CO O •-d d o o d CO c O CO o d d c o c o d d o r--o co d d o cn o oo d •sr c O ~~ o o *— d o o c o LO d *~ d CD (0 LO t o c CO o d d LO CO ^ o d ! I s- eg o CM r--! d • o o o i o o o O d LO LO d o LO co CM o cn d ! o ' •sr CM c o 5? c •ST CO LO o o CM CM o CM o o O d d d d c s o O) CO •sr i n o CO •sr •sT c^ d d d d o_ o c CD c O LO~ o to 1 o O O o o o o CN o o d. d O d d c O O •ST o cr o 00 00 o " d d d d d o c CD : cn o CO* C\j o •sT LO CM d d d d o O d d ~ •sr CD O IT) o o O d d " 3 CD LO o •sr m oo d co CO •sr d CO •sr CM •sr m CM CO •sr •sr d cn co i n co cn co LL cn cn m i n CD •sr *— CM CM T T— • +i +i • i +i +i cn CD r~ •sr •sr CM •sr •sr CN •sr •sr LO CD •sr T— CN T— +i +i +i < +i o cn CO CD CO •sr CO •sr +1 00 •sr +l CO CN +1 o CN +1 00 •sr LO i*-CM T— 1 i +i i 1 +i 00 CO CO o 00 00 CO CO •sr CM CN CN +i +i +i +i +i +i co co r-- •sr •sr o CN •sr i n CM •sr m CM +i CM CO CM CD CM +i O i n i n CN +i 00 CO i n CN +i r-CN CN +i CO LO LO CN +i DC DC +1 i n •sr CO m cn co t— CN CM o T— +i +i +i +i +i +i CO cn CN cn m CM LO CD •sr •sr LO •sr r - r - •sr CO T— CN •c— T— +i +i i +i +i i •sr CO CO o CO i n CO i n +1 r -•sr CD •sr CO •sr +i CO CO +1 CO CO +1 •sr CD +i 5; CC DC A P P E N D I X 6: R A N K S O F F A M I L I E S A F T E R S E L E C T I O N A C C O R D I N G T O D I F F E R E N T RISK R E D U C T I O N S T R A T E G I E S 179 Table A6.1 Ranks of families after selection based on different objectives, at selection intensity /=/. EKRR PGQ Selection Objectives Selection Objectives VOL DWGHT VOL_RD VOL DWGHT VOL_RD Fam Rank non-lin lin non-lin lin non-lin non-lin lin non-lin lin non-lin 1 29 29 29 29 29 161 125 14 125 33 2 78 78 78 78 78 5 139 161 139 165 3 85 76 85 76 85 138 18 167 18 37 4 70 70 50 70 70 167 3 138 3 3 5 75 75 75 75 50 85 5 85 5 79 6 50 85 70 85 75 16 16 30 16 125 7 76 50 76 50 76 116 141 115 141 139 8 104 71 104 71 104 14 167 16 167 128 9 40 5 49 5 49 18 85 147 85 141 10 49 40 40 40 40 87 161 5 161 122 11 71 68 20 68 71 142 116 87 116 111 12 20 49 71 49 20 109 87 29 87 72 13 68 20 34 20 68 29 56 109 56 124 14 34 90 68 90 34 1 33 26 33 64 15 90 81 90 81 90 30 165 142 165 163 16 5 82 28 82 23 56 1 116 1 174 17 28 57 23 57 5 115 66 84 66 160 18 23 6 83 6 83 125 14 18 14 127 19 83 23 9 18 28 71 6 9 6 94 20 9 18 5 23 82 26 128 126 128 156 21 82 32 82 32 9 121 117 56 117 18 22 57 34 30 34 30 17 98 17 98 117 23 30 9 57 9 57 137 121 71 121 135 24 12 12 12 12 12 139 79 8 17 66 25 81 83 123 83 123 66 17 6 79 42 26 123 30 109 30 81 96 122 125 122 1 27 42 112 81 112 109 147 109 11 109 133 28 109 77 42 77 42 9 26 98 26 76 29 6 25 6 25 112 8 53 149 53 6 30 112 28 103 28 6 53 143 137 143 129 31 32 104 66 104 32 143 126 121 126 56 32 18 124 112 124 31 126 137 139 137 98 33 31 31 111 31 108 98 138 1 138 151 34 111 42 32 42 18 141 142 66 142 121 35 108 108 31 108 111 112 30 143 30 112 36 66 41 108 41 66 6 37 53 37 53 37 103 19 18 19 103 120 124 96 124 21 38 25 111 41 111 25 117 94 117 94 143 39 77 88 25 88 77 94 156 120 156 136 40 41 47 80 47 41 129 129 62 129 120 41 80 123 77 123 124 3 71 141 71 5 42 124 80 19 80 88 55 120 55 120 116 43 19 52 132 52 19 93 9 119 9 93 44 88 66 88 66 80 149 64 112 64 173 45 132 132 58 132 132 84 127 129 127 34 46 58 74 124 74 58 42 112 3 112 17 180 Table A6.1 continued EKRR PGQ Selection Objectives Selection Objectives VOL DWGHT VOL_RD VOL DWGHT VOL_RD Fam Rank non-lin lin non-lin lin non-lin non-lin lin non-lin lin nO:!-lin 47 52 103 52 103 47 122 111 81 111 78 48 47 58 74 58 52 127 147 34 147 137 49 74 48 47 48 74 128 42 31 42 55 50 101 55 101 55 101 24 133 133 133 87 51 53 67 53 67 53 11 96 21 96 16 52 55 39 33 39 39 156 29 156 29 24 53 39 109 55 109 33 124 115 93 115 62 54 33 51 39 51 55 79 8 122 8 126 55 48 53 87 53 48 119 55 128 55 4 56 87 100 48 100 87 34 174 94 174 96 57 100 101 100 101 100 31 62 42 62 31 58 67 87 65 15 67 133 34 24 34 85 59 65 15 67 87 65 62 21 35 21 172 60 51 107 51 107 51 64 93 46 93 35 61 107 89 107 89 107 111 151 64 151 26 62 59 64 59 64 59 21 119 127 119 119 63 102 33 102 102 102 35 84 151 84 71 64 89 102 89 33 89 33 72 124 72 167 65 105 59 105 59 105 174 163 111 163 109 66 95 110 95 110 95 46 24 165 24 46 67 106 35 106 35 106 165 31 79 31 9 68 13 13 94 13 94 136 11 174 11 161 69 94 95 13 95 13 37 160 136 160 8 70 110 106 69 106 110 72 136 4 136 142 71 69 94 110 94 69 151 149 172 149 81 72 15 65 114 65 15 81 35 78 35 14 73 35 105 119 105 114 163 81 37 81 147 74 114 36 15 36 35 4 135 163 135 30 75 119 114 35 114 119 160 46 135 46 11 76 72 119 72 119 72 78 76 33 76 84 77 64 69 36 69 64 76 78 173 78 29 78 36 72 64 72 36 173 4 160 4 138 79 38 38 38 38 38 135 173 72 173 149 80 63 63 63 63 63 172 172 76 172 115 181 Table A6.2 Ranks of families after selection based on different objectives and risk reduction strategies. Selection Objectives and Risk Reduction Criteria Vol Tww VOL_Tp VOL_TR VOL RD VOLTwwTRI MaxiMax Fam. Rank All All MaxiMax MaxiMin MaxiMax MaxiMin MaxiMin 0 Change 1 115 115 115 117 4 3 115 117 2 94 94 94 4 115 5 94 4 3 66 66 66 9 121 66 66 9 4 137 137 137 173 151 24 137 167 5 3 3 3 167 46 16 3 173 6 24 24 24 33 129 94 24 33 7 35 35 35 81 165 14 35 81 8 85 85 85 111 117 112 85 111 9 125 125 125 121 81 29 125 121 10 5 5 5 165 173 137 5 165 11 16 16 16 84 133 17 16 84 12 122 122 122 174 122 125 122 17 13 8 8 8 17 163 143 8 174 14 163 163 163 116 8 85 163 116 15 29 29 42 46 35 87 29 46 16 42 42 29 151 11 93 42 151 17 87 87 87 14 84 116 87 14 18 112 112 112 143 42 33 112 143 19 93 93 93 133 167 167 93 133 20 129 129 129 11 9 42 129 11 21 11 11 11 129 85 174 11 129 22 133 133 133 93 111 35 133 S3 23 143 143 143 112 174 84 143 112 24 14 14 14 87 33 111 14 87 25 151 151 151 29 87 11 151 29 26 46 46 46 42 125 8 46 42 27 116 116 116 163 116 9 116 163 28 17 17 17 8 93 122 17 8 29 174 174 174 122 137 115 174 122 30 84 84 84 16 143 163 84 16 31 165 165 165 5 17 133 165 5 32 121 121 121 125 29 165 121 125 33 111 111 111 85 94 81 111 85 34 81 81 81 35 112 117 81 35 35 33 33 33 24 14 129 33 24 36 167 167 167 3 66 46 167 3 37 173 173 173 137 16 173 173 137 38 9 9 9 66 24 121 9 66 39 4 4 4 94 5 151 4 94 40 117 117 117 115 3 4 117 115 182 

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.831.1-0090560/manifest

Comment

Related Items