Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Long-term growth responses in Sitka spruce populations to geoclimatic changes from IUFRO provenance trials… Xu, Ping 1998

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

Item Metadata

Download

Media
831-ubc_1998-0669.pdf [ 8.63MB ]
Metadata
JSON: 831-1.0075297.json
JSON-LD: 831-1.0075297-ld.json
RDF/XML (Pretty): 831-1.0075297-rdf.xml
RDF/JSON: 831-1.0075297-rdf.json
Turtle: 831-1.0075297-turtle.txt
N-Triples: 831-1.0075297-rdf-ntriples.txt
Original Record: 831-1.0075297-source.json
Full Text
831-1.0075297-fulltext.txt
Citation
831-1.0075297.ris

Full Text

LONG-TERM GROWTH RESPONSES LN SITKA SPRUCE POPULATIONS TO GEOCLLMATIC CHANGES FROM IUFRO PROVENANCE TRIALS IN BRITISH COLUMBIA by P F N G X U M.Agri., Northeast Forestry University, China, 1996 A THESIS SUBMITTED LN PARTIAL FULFILMENT OF THE REQUIREMEMTS FOR THE DEGREE OF MASTER OF SCIENCE in THE FACULTY OF GRADUATE STUDIES Faculty of Forestry Department of Forest Sciences We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA May 1998 ©PingXu, 1998 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department of (f~~~Z?T-e^T <3<St •e^XJL^s The University of British Columbia Vancouver, Canada Date MayjSr. DE-6 (2/88) TABLE OF CONTENTS PREFACE ; xii ACKNOWLEDGMENTS XV TABLE OF CONTENTS ii LIST OF TABLES V LIST OF FIGURES ix 1. Geoclimatic trends and the underlying major ecological factors in growth of 43 Sitka spruce provenances tested in British Columbia 1 Abstract : 1 1.1. Introduction 2 1.2. Data Profile and abbreviation 3 1.2.1. Experimental designs 8 1.2.2. Growth measurements 8 1.2.3. Geoclimatic data for test site and provenance origin 9 1.3. Methods of Data Analyses 10 1.3.1. Partitioning growth variations 10 1.3.2. Examining geographic trends in growth 11 1.3.3. Revealing phenotypic sensitivity to site climatic conditions 12 1.3.4. Unveiling the underlying major climatic factors for the geographic trends 13 1.3.5. Statistical criteria 16 1.4. Results and Discussions 16 1.4.1. Partitioning of variations in growth traits 16 1.4.2. Geographic trends inherited by the provenances 22 1.4.3. Climatic sensitivities to site conditions 28 1.4.4. Underlying major ecological factors that differentiate the provenances 33 1.5. Conclusions 38 1.6. References 39 2. Effect and probability of white pine weevil attacks on height growth of 14 Sitka spruce provenances 42 Abstract 42 2.1. Introduction 43 2.2. Materials 45 2.3. Methods of Data Analyses 47 2.3.1. Examining the effects of weevil attack on height growth over measured years 47 i i 2.3.2. Exploring the sources of variation accounting for weevil attack frequencies over measured years 50 2.4. Results and Discussions 52 2.4.1. Experimental effects, weevil effects, age trends and interactions 52 2.4.2. Impacts of weevil attacks on height growth at different ages 56 2.4.3. Explanatory sources accounting for weevil attack occurrence 61 2.5. Conclusions 74 2.6. References 75 3. Long-term growth responses in Sitka spruce populations to seed transfer from IUFRO provenance trials in British Columbia 78 Abstract 78 3.1. Introduction 79 3.2. Data Profile and Abbreviation 81 3.3. Methods of Analyses and Limitations of the predictions 83 3.3.1. Identifying effective geoclimatic predictors 83 3.3.2. Quantifying general volume response from geoclimatic changes 84 3.3.3. Predictions pertaining to site geoclimatic conditions 85 3.3.4. Graphic approach for seed transfer guidelines 86 3.3.5. Limitations of the predictions and modeling 87 3.3.6. Statistical assumptions and criteria 89 3.4. Results and Discussions 90 3.4.1. Effective geoclimate predictors 90 3.4.2. Trends from the scatter plots 92 3.4.3. General predictions on each effective geoclimatic predictor 94 3.4.4. Predictions pertaining to site conditions 104 3.4.5. Contours assisting the guide of seed transfer in BC 113 3.5. Conclusions 123 3.6. References 125 4. Sitka spruce IUFRO provenance trials in British Columbia: old experiment new approach 127 Abstract 127 4.1. Introduction 128 4.2. Materials and Methods 133 4.3. Results and Discussions 136 4.3.1. Growth response to thermal-climatic changes only 136 4.3.2. Growth response to thermal-climatic changes pertaining to site moisture condition 144 4.4. Conclusions 148 4.5. References 150 i i i RECOMMENDATIONS 152 APPENDICES 153 Appendix I. Site mortality rates 154 Appendix II. Examining climatic sensitivities by the 11 frequently tested provenances 155 Appendix III. Supporting information for the response surface analyses in Chapter Three 157 Appendix IV. Figures for differences in growth response between wet and less-wet sites 171 Appendix V. Technical report of the response surface analysis in Chapter Four 172 iv LIST OF TABLES Table 1-1. Geographic locations of the 12 test sites in the 3 series of Sitka spruce provenance trials 5 Table 1-2. Name, IUFRO number, and the geographic locations of the 43 IUFRO Sitka spruce provenances along with their planting series designation 6 Table 1-3. Coefficients of the original growth variables with the first two growth principal component (GPC1 and GPC2) and the percentage of original variations in growth accounted for by GPC1 and GPC2, respectively 17 Table 1-4. ANOVA for the first growth principal component (GPC1) of eight growth measures 18 Table 1-5. Partitioning of growth variations for each series of the Sitka spruce provenance trials in BC 21 Table 1-6. Ranks of the general growth performance of the provenances tested in Series II & III, where significant G x E interactions were detected 21 Table 1-7. Partial R2's for the significant predictive geographic variables ('factor') of provenance origins and the average model R's of all the regression models on the 5 growth measures at each test site specifically (from SAS REG procedure with forward selection at a = 0.01 level) 24 Table 1-8. Results of PCA on the 10 macro-climatic variables of the 11 test sites and 43 provenance origins altogether (based on standardized data, from SAS PRINCOMP procedure) 26 Table 1-9. Canonical correlations and explained variance percentage of each pair of canonical variables (Cans) between growth measures and site climate variables 31 Table 1-10. Canonical structure of the growth variables with the first two pairs of canonical variables 31 Table 1-11. Canonical structure of the site climate variables with the first two pairs of canonical variables, represented by the correlation coefficients 32 Table 1-12. Redundancies of the canonical variables for both growth and site climate variables 33 Table 1-13. Raw variations in growth and provenance origin place's climate conditions cross-explained by their opposite redundant variables from the redundancy analysis based on the within and between correlation matrices of growth measures and provenance origin climate variables 35 Table 1-14. Redundancy structure (correlation coefficients) of the original variables with the first provenance climate redundant variables 37 Table 2-1. Locations and the first principal component values (sitePCl) describes site mildness of the four weeviled test sites 47 Table 2-2. Names, IUFRO number, and place of origin for the 14 provenances tested at the four weeviled sites 47 Table 2-3. ANOVA of the repeatedly measured heights at the four test sites for the experimental effects, weevil attacks, and age trend and the interactions of weevil, age with experimental effects 55 Table 2-4. Probabilities of the computed F statistics greater than the critical F values in testing the null hypothesis that there were no weevil influences on height increment rates (HIR's) at each measurement intervals since weevil attack recorded, based on SAS ANOVA of effects on each single RHT and MANOVA of overall effects on all the HIR's 59 Table 2-5. Means for Height Increment Rates (HIRs, in percentage) and Heights (HT, in decimeter) and Student-Newman-Keuls (SNK) test results for these means corresponding to weevil attack codes at different measured years at the four test sites 60 Table 2-6. Maximum Likelihood estimates and logistic regression model information in predicting the probabilities of weevil attack occurrences at different measurement intervals (Prob(WV6), Prob(WVlO) and so on) by previous height (HT3, HT6 and so on)), site mildness (sitePCl), block and provenance origins' latitudinal (PLAT), based on individual trees observations and the use of backward selection at a = 0.05 level 64 Table 2-7. Mean Predicted probabilities of weevil attack(s) occurrence for previously weeviled and unweeviled trees at different measurement intervals 63 Table 2-8. Mean and standard deviation of the mean of previous height (in decimeter) for weeviled ('yes') and unweeviled ('no') trees immediately before weevil attack occurrence at each measurement intervals 67 Table 2-9. Height (HT, in decimeter) means and mean height increment rates (HIR, in percentage) along with the Student-Newman-Keuls (SNK) multiple range test results of these means for the 13 provenances tested at the four test sites 73 vi Table 3-1. The amount of original variations of the 'Devi-' variables and 'Diff-' variables explained by the redundant variables of their opposite group 91 Table 3-2. Correlation coefficients (i.e., redundancy loadings) between the original variables with the first redundant variable of the 'Diff-' group 91 Table 3-3. Estimated parameters for the prediction models along with the quality information of the models 97 Table 3-4. Mean predicted values for the percent deviations of volume growth from local performance (DeviVOL20 (%)) and the approximate standard errors of the means predicted from DiffLAT, DifiMAT, DiffMTCM, DiffDDO and DiffNFFD, respectively (arranged in the ascending order of DiffLAT) 98 Table 3-5. Volume-gains in northward seed transfer that are conditional upon site summer pricipitation 107 Table 3-6. Suitable geoclimatic ranges at BC for planting areas along with provenance origins if VOL20 = 150 dm3 is the level of individual tree volume growth to be achieved 123 Table 4-1. Estimated parameters for the prediction models relating NDeviVOL20 to the four thermal-climatic change variables, respectively, along with the quality information for these models 139 Table 4-2. Predicted volume responses (NDeviVOL20 (%)) from thermal-climatic changes between provenance origin and planting site in seed transfer of the provenance trials in the 20th year after planting 143 Table 4-3. Net expectable volume-gains to changes in mean annual temperature (DifiMAT) that are conditional upon site summer precipitation (SMSP) 147 Table 1-1. Mortality rate (%) of the trees over 20 years after planting at 12 sites of the three series of Sitka spruce provenance trials in BC 154 Table II-l. R 2 of the forward regression (a=0.01) of plot means of the 11 frequently tested provenances (from S to N) with site climate variables 155 Table III-l. Technical report of the response surface analysis for variable DeviVOL20 " with DiffLAT and SLAT (from SAS RSREG procedure) 157 Table III-2. Technical report of the response surface analysis for variable DeviVOL20 with DifiMAT and SMAT (from SAS RSREG procedure) 158 vii Table III-3. Technical report of the response surface analysis for variable DeviVOL20 with DifiMTCM and SMTCM (from SAS RSREG procedure) 159 Table III-4. Technical report of the response surface analysis for variable DeviVOL20 with DiffDDO and SDDO (from SAS RSREG procedure) 160 Table III-5. Technical report of the response surface analysis for variable DeviVOL20 with DiffNFFD and SNFFD (from SAS RSREG procedure) 161 Table III-6. Technical report of the response surface analysis for variable DeviVOL20 with DifiLAT and SMSP (from SAS RSREG procedure) 162 Table III-7. Technical report of the response surface analysis for variable VOL20 with SLAT and PLAT (from SAS RSREG procedure) 163 Table III-8. Technical report of the response surface analysis for variable VOL20 with SMAT and PMAT (from SAS RSREG procedure) 164 Table III-9. Technical report of the response surface analysis for variable VOL20 with SMTCM and PMTCM (from SAS RSREG procedure) 165 Table 111-10. Technical report of the response surface analysis for variable VOL20 with SDDO and PDDO (from SAS RSREG procedure) 166 Table III-ll. Technical report of the response surface analysis for variable VOL20 with SNFFD and PNFFD (from SAS RSREG procedure)...'. 167 Table IH-12. Technical report of the response surface analysis for variable VOL20 with SMSP and PMSP (from SAS RSREG procedure) 168 Table. 111-13. Pearson Correlation coefficients between pairs of geoclimatic distance variables (i.e., Diff- variables) 169 Table HI-14. Geoclimatic distance ranges of the seed transfer used in the three series of Stika spruce provenance trials in BC, i.e., the experimental span 170 Table V-l . Technical report of the response surface analysis for variable NDeviVOL20 (in logarithmic value) with DiffMAT and SMSP (from SAS RSREG procedure) 172 viii LIST OF FIGURES Fig. 1-1. Locations of the provenance origins and test sites for Sitka spruce provenance trials in British Columbia 7 Fig. 1-2. Growth variation explained by provenance location along with site mildness index and relative site mean for HT20 (%) 27 Fig. 1-3. Associations of site productivity (HT20%) and the amount of explained growth variation by provenance origin's locations (R2%) with test site mildness (sitePCl) 27 Fig. 2-1. Mean height trends of weeviled and unweeviled trees over measured years 54 Fig. 2-2. Density plots of weeviled and unweeviled trees at current measurement interval corresponding to the immediately previous height at the four measured years (illustrated on top of each plot), respectively 68 Fig. 2-3. Frequencies of weevil attack occurrences in measured years and at different test sites 69 Fig. 2-4. Frequencies of weevil attack occurrences at measurement intervals and on different provenances 71 Fig. 3-1. Scatter plot of DeviVOL20 versus DiffLAT and the quadratically smoothed trend 93 Fig. 3-2. Scatter plot of DeviVOL20 versus DiffMAT and the quadratically smoothed trend 93 Fig. 3-3. Scatter plot of DeviVOL20 versus DiffMTCM and the quadratically smoothed trend 93 Fig. 3-4. Scatter plot of DeviVOL20 versus DiffDDO and the quadratically smoothed trend 93 Fig. 3-5. Scatter plot of DeviVOL20 versus DiffNFFD and the quadratically smoothed trend 93 Fig. 3-6. Standard Error of the mean prediction on DiffLAT 100 Fig. 3-7. Standard Error of the mean prediction on DiffMAT 100 ix Fig. 3-8. Standard Error o f the prediction on D i f f M T C M 100 Fig. 3-9. Standard Error o f the prediction on D i f f D D O 100 Fig. 3-10. Standard E r r o r o f the prediction on D i f f N F F D 100 Fig. 3-11. Contour graph reflecting the predictions o f D e v i V O L 2 0 (%) on D i f f L A T (= P L A T - S L A T ) pertaining to site latitude conditions ( S L A T ) 108 Fig. 3-12. Contour graph reflecting the predictions o f D e v i V O L 2 0 (%) on D i f f M A T (= P M A T - S M A T ) pertaining to site mean annual temperature conditions ( S M A T ) 109 Fig. 3-13. Contour graph reflecting the predictions o f D e v i V O L 2 0 (%) on D i f f M T C M (= P M T C M - S M T C M ) pertaining to site mean coldest month temperature conditions ( S M T C M ) 110 Fig. 3-14. Contour graph reflecting the predictions o f D e v i V O L 2 0 (%) on D i f f D D O (= P D D 0 - S D D 0 ) pertaining to the amount o f winter coldness o f the planting sites ( S D D 0 ) I l l Fig. 3-15. Contour graph reflecting the predictions o f D e v i V O L 2 0 (%) on D i f f N F F D (= P N F F D - S N F F D ) pertaining to lengths o f annual frost free period o f the planting sites ( S N F F D ) 112 Fig. 3-16. Contour graph reflecting the predictions o f D e v i V O L 2 0 (%) on D i f f L A T (= P L A T - S L A T ) pertaining to lengths o f mean summer precipitation o f the planting sites ( S M S P ) 113 Fig. 3-17. Cntour graph o f V O L 2 0 (dm 3 ) on S L A T (site latitude) and P L A T (provenance origin's latitude) 116 Fig. 3-18. Contour graph o f V O L 2 0 (dm 3 ) on S M A T (Site M e a n A n n u a l Temperature) and P M A T (Provenance origin's M e a n A n n u a l Temperature) 117 Fig. 3-19. Contour graph o f V O L 2 0 (dm 3 ) on S M T C M (Site M e a n Temperature o f the Coldest M o n t h ) and P M T C M (Provenance origin's M e a n Temperature o f the Coldest M o n t h ) 118 Fig. 3-20. Contour graph o f V O L 2 0 (dm ) on S D D 0 (Site annual accumulated Degree Days be low 0 ° C ) and P D D 0 (Provenance origin's annual accumulated Degree Days below 0 ° C ) 119 Fig. 3-21. Contour graph o f V O L 2 0 (dm 3 ) on S N F F D (Site annual N u m b e r o f Frost Free Days) and P D D 0 (Provenance origin's annual N u m b e r o f Frost Free Days) 120 x Fig. 3-22. Contour graph of VOL20 (dm3) on SMSP (Site mean Summer Precipitation) and PMSP (Provenance origin's mean Summer Precipitation) 121 Fig. 4-1. Residual plot for the curvilinear regression model of DeviVOL20 = /(DiffDAY, DiffDAY2) 138 Fig. 4-2. Scatter plot of NDeviVOL20 (in logarithmic value) versus DifiMAT and the quadratically smoothed trend 138 Fig. 4-3. Scatter plot of NDeviVOL20 (in logarithmic value) versus DifiMTCM and the quadratically smoothed trend 138 Fig. 4-4. Scatter plot of NDeviVOL20 (in logarithmic value) versus DiffDDO and the quadratically smoothed trend 138 Fig. 4-5. Scatter plot of NDeviVOL20 (in logarithmic value) versus DiffNFFD and the quadratically smoothed trend 138 Fig. 4-6. Scatter plot of the provenance-by-site means for VOL20 (in logarithmic values) of 11 frequently tested provenances at the 11 test sites to the amount of site summer precipitation (SMSP) 145 Fig. 4-7. Contour of the quadratically smoothed response surface of NDeviVOL20 (%) to changes in mean annual temperature (DifiMAT) over site summer precipitation (SMSP) gradient 148 Fig. IV-1. Scatter plot of DeviVOL20 (in logarithmic value) versus DiffLAT (°N) and the quadratic smoothers for wet site and less-wet sites 171 Fig. IV-2. Scatter plot of DeviVOL20 (in logarithmic value) versus DiffMAT (°C) and the quadratic smoothers for wet site and less-wet sites 171 Fig. IV-3. Scatter plot of DeviVOL20 (in logarithmic value) versus DifiMTCM (°C) and the quadratic smoothers for wet site and less-wet sites 171 Fig. IV-4. Satter plot of DeviVOL20 (in logarithmic value) versus DiffDDO (degree days) and the quadratic smoothers for wet site and less-wet sites 171 Fig. IV-5. Scatter plot of DeviVOL20 (in logarithmic value) versus DiffNFFD (day) and the quadratic smoothers for wet site and less-wet sites 171 X I PREFACE In the history o f tree domestication and improvement, provenance trials served as the first step o f systematic and scientific research. They have been conducted for more than 200 years in many commercia l ly important tree species, in studies where seeds from different locations (populations) within the natural range o f a species were collected, and the seedlings were planted together in a test site, namely, a ' common garden', to observe the growth potentials or other interested traits o f these populations in the planted area. A s theories in forest genetics and experimental design developed over the latest forty years, the range and scale o f provenance trials also expanded, involv ing more seed sources and tested at more sites, some even being internationally cooperative. The original practical target o f s imply identifying suitable seed source for planting also developed into many theoretical purposes such as, the assessment o f the species' inter- and intra-population genetic variations and characterization, phenotypic flexibility and sensitivity, and genotype-by-environment ( G x E ) interaction. T h e most recently proposed approach o f using provenance trial data to simulate long-term growth response to rapid climate change (Langlet 1971; K o s k i 1989) has expanded the use o f provenance trials beyond forest genetics. Sitka spruce, Picea sitchensis (Bong.) Carr iere , a fast growing softwood species, occurs naturally along a narrow strip o f the western Pacific coast o f N o r t h A m e r i c a from A l a s k a to Cal i fornia over 22 degrees o f latitudes (Daubenmire 1968). Primari ly a coastal species, it extends wel l inland along river valleys in Brit ish C o l u m b i a ( B C ) in areas o f high humidity. It has become abundant as a plantation tree over large areas o f western Europe , a successful endeavor that is accounted for by the tree's qualities, inc luding its exceptionally great vigor, straight form, versatility to soil conditions and high timber quality (Holmes 1987). It is also expected to be the most prominent reforestation species in its native habitat, once the white pine weevil (Pissodes strobi) is under control ( Y i n g 1991). In order to guide seed transfer and screen for weevi l resistant populations, the British C o l u m b i a Minis try o f Forests ( M o F ) launched three series o f Sitka spruce provenance trials in 1973 and 1975, two series o f which are part o f the international provenance trials o f Sitka spruce, coordinated by the International U n i o n o f Forest Research Organizations ( I U F R O ) . These trials are located at 13 sites o f southern coastal B C and together test 43 Sitka spruce I U F R O provenances, the range o f which covers the species ma in natural distribution from southern A l a s k a to Oregon coast spanning over 17 degrees o f latitude. G r o w t h and health conditions o f all trees have been recorded periodically for 20 years since planting. Prel iminary reports on height growth from these trials were made by Illingworth (1978) and Y i n g (1997). However, the valuable 20-year data on every single tree had not be fully analyzed. T h e objective o f this study is to use the above mentioned data to address the multiple purposes o f provenance trial, either practical or theoretical as stated before, within the limitations set by the experimental design and data availability. Thus the thesis is written in four separate chapters each addresses one major aspect o f the study. Chapter One mainly addresses genetic variations and phenotypic sensitivity o f the species in growth traits. Emphases were given to the underlying ecological factors that drive geographical trends and cl imatic sensitivities o f trees. Since extensive attacks o f white pine weevil occurred at four o f the test sites under study, Chapter T w o is devoted to weevil resistance, assessing the damage o f weevi l attack to height growth over the 20-year period, and exploring sources accounting for the variability o f weevil xiii attack frequencies. Chapter Three is targeted at the primary goal o f provenance trials, i.e., evaluating the growth responses from latitudinal seed transfer, and defining suitable seed source range for planting under given site conditions in B C . Final ly , Chapter Four follows the new approach o f using provenance trial data to predict the impact o f global warming on this species in term o f vo lume growth. Thi s approach is feasible because o f the wide span o f latitudinal distribution in this species, but relatively small variations in elevational distribution and minimal soil diversity o f the seed sources and test site locations used in this study. A l though these four chapters are dealing with different aspects o f biological and ecological characters as well as forestry practice guidelines in the species, they are intrinsically related to each other. Therefore, 1 compi led them as one vo lume for the thesis. xiv A C K N O W L E D G M E N T S I am deeply grateful to my supervisor, D r . Yousry E l - K a s s a b y , who gave me insights into this research project and funded my degree program, and carefully reviewed the manuscript o f this thesis. I also w o u l d like to give my thanks to Drs . Gene N a m k o o n g and Sally Ai tken , from the Faculty o f Forestry o f this university, who have given me a great deal o f advices on my study during m y two-year stay at this university, and who have reviewed the draft o f this thesis and provided me a lot o f critical comments on the interpretation and writing o f the analytical results. Special thanks are given to D r . C h e n g C . Y i n g , from the Minis try o f Forests o f Brit ish Co lumbia , who not only arranged the data supply for this study, patiently responded m y every inquiry about the details o f data collection, but also advised m y study and carefully reviewed these chapters. Dr . V a l L e m a y , from the Faculty o f Forestry o f this university, has given me many useful suggestions on the analytical techniques for this research project. T h e Research Branch o f British C o l u m b i a Min i s try o f Forests are also highly appreciated for their data collection and entering. Thi s project was also funded by the E d w a r d W . Bassett M e m o r i a l Scholarship in Reforestation granted by this university. P ing X u M a y 27, 1998 XV 1. Geoclimatic trends and the underlying major ecological factors in growth of 43 Sitka spruce provenances tested in British Columbia Abstract: Sitka spruce, (Picea sitchensis (Bong.) Carr. ) , a highly moisture sensitive conifer, has populations (provenances) differentiated in growth by temperature (especially winter harshness) and photoperiod regimes o f source environments. T h e 20-year growth data o f 43 Sitka spruce I U F R O provenances tested at 11 sites in Brit ish C o l u m b i a were analyzed along with the geoclimatic conditions o f the test sites and o f provenance origins. Mult ivariate analyses, including multiple regression, canonical correlation analysis and redundancy analysis, were applied to reveal the inherent geographic trends among provenances, and the climatic sensitivities o f the tree in growth traits, as wel l as the underlying major climatic factors accounting for growth variation among provenances. Results indicated that about 63% o f the genetic variability in growth o f the species was explained by variation o f cl imatic conditions o f the provenance origins. Keywords: cl imatic sensitivity; geographic trend; growth trait; provenance trial; Sitka spruce (Picea sitchensis (Bong.) Carr) . 1 1.1. Introduction Sitka spruce, Picea sitchensis (Bong.) Carr . , a fast growing softwood species native to the Pacific west coast o f North A m e r i c a , occupies a long, narrow strip from A l a s k a to Cal i fornia spanning over 22 degrees o f latitude (Daubenmire 1968). Primari ly a coastal species, it also extends wel l inland along river valleys in Brit ish C o l u m b i a ( B C ) where high humidity is available. It has also become abundant as a plantation tree over large areas o f western Europe (British Isles in particular (Hermann 1987)), a successful endeavor that is attributable to the species' qualities, i.e., its exceptionally great vigor, straight form, versatility to soil conditions and high timber quality (Holmes 1987). It is also expected to be the most prominent reforestation species in its native habitat, once the threat from the white pine weevil (Pissodes strobi) is under control ( Y i n g 1991). W i t h a wide latitudinal distribution, Sitka spruce, in c o m m o n with most other North A m e r i c a n conifers that are widely distributed, is remarkably variable (Roche and H a d d o c k 1987). This variability is habitat-correlated and genetically based (Burley 1965; Roche 1969; Falkenhagen 1977). A l though studies have been conducted on the biological and ecological aspects o f the species (see Henderson and Faulkner 1987), the genetic potential o f the tree has not yet been fully exploited in forestry practice (Roche and H a d d o c k 1987). W i t h high genetic variability, S i tka spruce is not only expected to be able to overcome the weevi l problem, eventually, but also to be able to cope adaptively with the potential o f negative impacts from rapid climate changes (e.g., global warming). However, we were not very clear before this study how much genetic variability the species possesses in growth performance, what the underlying 2 ecological force(s) are that have resulted in differentiation o f these provenances, and how the species wi l l respond in face o f rapid climate changes. In order to guide seed transfer and screen for weevil resistant populations, the British C o l u m b i a Minis try o f Forests ( M o F ) established three series o f Sitka spruce provenance trials in early 1970s in southern coastal area o f Brit ish C o l u m b i a ( B C ) , two series o f which are part o f the international provenance trials o f Sitka spruce, coordinated by the International U n i o n o f Forest Research Organizations ( I U F R O ) . Preliminary reports on height growth from these trials were made by Ill ingworth (1978) and Y i n g (1997). However, the data collected on individual trees over the 20 years since planting have not be fully analyzed. T h e objectives o f this chapter are: 1) to assess the magnitude o f genetic variability in growth traits among Sitka spruce populations; 2) to examine geographic trends among the provenances in growth performance; and 3) to unveil the underlying ecological forces that driving the genetic variability as wel l as the phenotypic sensitivity in growth o f the species. 1.2. Data Profile and Abbreviation Three series o f Sitka spruce provenance trials in B C (Illingworth 1978; Y i n g 1991 and 1997) were used in this study. Growth and geoclimatic data were supplied by the Research Branch o f M o F . T h e plantations were established in 1973 and 1975, using 43 Sitka spruce I U F R O provenances, at 13 sites along coastal B C area including a few peripheral inner coastal sites. Locations o f the test sites and provenance origins are illustrated in F i g . 1-1, with details in 3 Tables 1-1 and -2. Orig inal ly , Series I was designed for guiding seed transfer on the Queen Charlotte Islands, north o f the natural range o f the white pine weevi l to al low for successful planting o f this species. There are five test sites in this series, four o f which are on the Graham Island at approximate 5 3 ° N and 1 3 2 ° W . T h e remaining site is on M o r e s b y Island, which is excluded from this study due to poor survival ( C C . Y i n g , B C M o F , personal communication). Geographic distances among the four test sites on the G r a h a m Island are very short, with elevation varying from 33 m to 460 m (Table 1-1). Series II and III, each has four test sites, are part o f the I U F R O international cooperative Sitka spruce provenance trials. These eight test sites are located from 4 9 ° 4 8 " to 5 5 ° 1 9 " N and from 1 2 6 ° 2 8 " to 1 3 2 ° 3 0 " W , with elevation variations from sea level to 600 m. U n l i k e those in Series I, the test sites in Series II and III are deliberately located in some contrasting environments ( Y i n g 1997) such that some peripheral inland habitats o f Sitka spruce (e.g., D r a g o n Lake , M a r o o n Creek, and Nass R iver ) are included. Consequently, the climatic conditions at these sites vary substantially, for instance, mean temperature o f the coldest month varies between 2.7 to -13 C , whereas mean annual precipitation from 1100 to 3850 m m . A l l the test sites are located in the Coastal Western H e m l o c k ( C W H ) biogeoclimatic zone ( B C ' s ecological classification o f the land, see Pojar et al 1987), except D r a g o n Lake which is in the Interior C e d a r - H e m l o c k ( ICH) biogeoclimatic zone (Table 1-1) and was excluded from this study due to h igh mortality (see below). W h e n pool ing the three series together, 43 provenances were tested in this study, the range o f which covers the species' main range from southern A l a s k a to Oregon coast, extending inland into the Sitka x white spruce hybridization zone, with elevation varying from sea level to 660 m. The range o f the 12 test sites covers coastal B C , extending from 4 9 ° 4 8 " to 5 5 ° 1 9 " N and from 1 2 6 ° 2 8 " to 1 3 2 ° 3 0 " W with elevations from sea level to 600m (Fig . 1-1, Tables 1-1 and -2). Table 1-1. Geographic locations of the 12 test sites in the 3 series of Sitka spruce provenance trials. Site Name Site Code BGC zone" Latitude Longitude Elevation (m) Series Graham Is. A l QC 1 CWHwhl 53°33" 132°20" 460 I Graham Is. A2 QC II CWHwhl 53°31" 132°11" 33 I Graham Is. A3 QC III CWHwhl 53°24" 132°16" 85 I Graham Is. A4 QC IV CWHwhl 53°22" 132°16" 100 1 Dragon Lakeb DL ICHmc2 55°19" 128°58" 210 11 Holberg Site HG CWHvhl 50°44" 128°07" 60 11 Maroon Creek MN CWHws2 54°46" 128°39" 600 11 Nass River NS CWHwsl 55°04" 129°26" 15 11 Head Bay HB CWHvml 49°48" 126°28" 15 111 Juskatla JU CWHwhl 53°34" 132°30" 20 111 Kitimat Valley KT CWHwsl 54°12" 128°33" 100 111 Rennell Sound RS CWHvh2 53°23" 132°28" 50 III 'BGC zone = Biogeoclimatic zone (ecological classification of the land in BC, see Pojar et al 1987). This test site was excluded from the study due to high mortality. 5 Table 1-2. Name, IUFRO number, and the geographic locations of the 4 3 IUFRO Sitka spruce provenances along with their planting series designation. U 1 U V V 1 1 U 1 1 V V J C» 1 \-' 1 I £ . Prov. Name IUFRO No. BGC zone" Latitude Longitude Elevation(m) Tested in Series Forks WA 3003 USA 48°04" 124°18" 137 11, 111 Hoquiam WA 3008 CWHvml 47°05" 124°03" 5 11,111 Necanicum WA 3012 USA 45049" 123°46" 45 Brookings OR 3018 USA 42°15" 124°23" 90 1 1 Yakutat AK 3021 USA 59°31" I39042" 12 Duck Creek 3024 USA 58°22" 134°35" 30 I, 111 Ohmer Creek 3025 USA 56°35" 132°44" 15 1 Derrick Lake 3026 ICHmcl 55°41" 128°41" 240 1 Craig 3027 USA 55°30" 133°08" 0 1 Old Hollis 3028 USA 55°28" 132°40" 0 1 Cranberry R. 3029 ICHmc2 55°28" 128°14" 510 1, III Ward Lake AK 3030 USA 55°25" 131°42" 15 Dragon L.Prov 3031 ICHmc2 55°21" 128°57" 255 1 Kitwanga 3032 !CHmc2 55°10" 127°52" 660 1, II Zolap Creek 3033 ICHmc2 55°09" 129°13" 15 1 Fulmar Creek 3034 ICHmc2 55°09" 128°58" 390 1 Moss Point AK 3035 USA 55°02" 131°33" 0 1 Cedarvale 3036 ICHmc2 55°01" 128°19" 240 1 Kitsuinkalum Lk 3039 CWHwsl 54043" 128°46" 135 1 Usk Ferry 3040 CWHwsl 54°38" 128°24" 135 I, III Shames 3041 CWHwsl 54°24" 128°57" 30 1 Kasiks River 3042 CWHvml 54°17" 129°25" 30 1 Inverness 3044 CWHvh2 54°12" 130°15" 30 I, II, III Aberdeen Cr. 3045 CWHvh2 54°12" 129°55" 0 1 Wedeene R. 3046 CWHvml 54°08" 128°37" 165 1 Humpback Cr. 3047 MHwhl 54°02" 130°22" 300 1 Masset Sound 3048 CWHwhl 53°55" 132°05" 0 1 Link Road 3049 CWHwhl 53°30" 132°10" 90 I, 11, III Copper Creek 3050 CWHwhl 53°08" 131°48" 75 1 Moresby Camp 3051 CWHwhl 53°03" 132°04" 60 1 Tasu Creek 3052 CWHvh2 52°52" 132°05" 15 1 Jedway 3053 CWHwhl 52°17" 131°13" 15 1 Holberg Prov. 3056 CWHvml 50°37" 128°07" 30 I, II, III Salmon Bay 3058 CWHxm2 50°02" 125°57" 0 1 Fair Harbour 3059 CWHvml 50°03" 127°02" 30 1 Squamish R. 3060 CWHdm 49053" 123°15" 30 I, II Tahsis Inlet 3061 CWHvml 49°50" 126°40" 0 Big Qualicum R 3062 CDFmm 49°23" 124°37" 0 1, II, III Haney 3063 CWHdm 49°14" 122°36" 300 1 Vedder 3064 CWHxm 1 49097" 121°56" 30 1 Port Renfrew 3065 CWHvhl 48°35" 124°24" 15 1 Muir Creek 3066 CWHxm2 48°23" 123°53" 0 1 Blenheim Mt. 3073 CWHvml 48°54" 124°57" 240 1 -a BGC zone = Biogeoclimatic zone (ecological classification of the land in BC, see Pojar et al 1987). 6 Fig. 1-1. Locations of the provenance origins and test sites for Sitka spruce provenance trials in British Columbia. 7 1.2.1. Experimental designs Completely randomized block designs were used at all test sites. However, the number of blocks used varied over test sites (Ying 1991 and 1997). The four test sites on Graham Island (i.e., QC I to IV) of Series I has four to six blocks each, randomly accommodating 38 provenances. Series II and III have identical experimental design, that is, nine blocks at each site and each block accommodating ten provenances with six common to both series, which makes a total of 14 provenances tested in these two series. In all the series, provenances were represented by a 9-tree-row plot within each block. Trees were planted at a spacing of 3 x 3 m. The number of trees planted in Series I is 6840 (9 trees x 38 provenances x 20 blocks for 4 sites together), while the number of trees planted in Series II and III is 3240 each (9 trees x 10 provenances x 9 blocks x 4 sites). As of year 20, mortality rates were low (0.04 ~ 14.69%) at all the test sites except Nass River (NS) and Dragon Lake (DL) where 31.98% and 51.73% of the trees died, respectively (see Appendix I). Mortality at NS was caused mainly by road expansion, whereas winter killing was the major cause at DL which was therefore excluded from this study. Thus, the remaining 11 test sites under this study have a total of 1390 plot-means and 220 provenance-by-site means for each growth measurement (see below). 1.2.2. Growth measurements At the 11 test sites studied, growth and health condition of the trees were recorded on individual tree base. Height (HT) of each tree was measured to the nearest decimeter in the 3rd, 6th, 10th, 15th and 20th year after planting (referred to as HT3, HT6, and so on). Diameter at breast height (DBH) of each tree was measured to the nearest millimeter at the 6th (occasionally), 8 10th, 15th and 20th year (referred to as DBH6, DBH10, etc.). In cases where a tree was less than 4 meters in height, diameter was measured at 1/3 of the total height. When both height and diameter of an individual tree were available, the tree volume (referred to as VOL6, VOL 10, etc.) was calculated in cubic decimeters, using Kovats' (1977) volume function for juvenile conifer trees. Diameter and volume data were not complete for all test sites until the 20th year after planting. 1.2.3. Geoclimatic data for test site and provenance origin Geoclimatic data are available for the 11 test sites as well as for the 43 provenance origins. The three geographic variables, i.e., latitude (LAT), longitude (LONG), and elevation (ELEV), were used in this study and were identified by adding the prefix 'S-' for test site, 'P-' for provenance origin. That is, the site geographic variables were abbreviated as SLAT, SLONG, and SELEV for test site while PLAT, PLONG, and PELEV specify provenance origin. Details of the geographic locations of the test sites and provenance origins are listed in Tables 1-1 and -2, respectively. Two sets of climatic data with ten macro-climatic variables (see below) were used in this study, one set for test sites and one for provenance origins. Since long-term growth response over the macro-geographic range is the primary goal, macro-climatic data are suitable for use. These ten macro-climatic variables generally define temperature, moisture and photoperiod conditions of the test sites as well as of provenance origins. The acronyms and units of these climatic variables are as follows, adding prefix 'S-' where they are for test site while 'P-' for provenance origin: 9 NFFD FFP DD5 DDO DAY* MAP MSP MAT MTCM MTWM Mean Annual Precipitation (mm) Mean Summer Precipitation (mm) (May ~ September) Mean Annual Temperature (°C) Mean Temperature of the Coldest Month (i.e., January) (°C) Mean Temperature of the Warmest Month (i.e., July) (°C) annual Number of Frost Free Days (day) annual continuously Frost Free Period (day) annual accumulated Degree Days above 5°C (degree day) annual accumulated Degree Days below 0°C (degree day) accumulated available day-length (hour) of the growth season (April ~ October) In Series II and III the macro-climatic data for test sites were obtained from the weather station closest to each test site. However, since no close weather station is applicable to the four sites on Graham island, the macro-climatic data in Series I were derived from climate models developed by Rehfeldt et al (1998). These models are applicable to BC, the United States above 48°30" N, the Alaska panhandle, and the narrow strips of Alberta and the Yukon along the BC border (CC. Ying, BC MoF, personal communication). Macro-climatic data for the provenance origins were obtained from IUFRO information system ( C C Ying, BC MoF, personal communication). 1.3. Methods of Data Analyses 1.3.1. Partitioning growth variations Repeated growth measurements are highly correlated variables. To avoid redundancy of analysis of variance (ANOVA) when partitioning the variations of growth measurement variables, these variables were synthesized by principal component analysis (PCA) using the * Using computer program downloaded from the web site: http://www.netti.fi.//~jjlammi//sum/html 10 SAS PPJNCOMP procedure to generate uncorrected (orthogonal) principal components (Jolliffe 1986). The variance-covariance matrix of the growth variables based on logarithmically transformed plot-means was the input matrix for this PCA procedure. The first principal component of growth measurements (GPC1) was used as an index of growth performance in general as it accounts for most (95%) of the original variations in growth measurements (see below, Table 1-3). The partition of variation in GPC1 was performed by the SAS MIXED procedure to obtain the Random Effect Maximum Likelihood (REML) estimates of the variances of all variation sources assuming they are random effects. The percentage contributions of the variation sources to total variation in growth were calculated based on these REML estimates. The significance of the variation among sources was tested by the SAS GLM procedure. Although the assumption of randomness of all experimental effects is statistically arguable under real situation of provenance trial, it is a prerequisite to obtain the REML estimates and to interpret the variation sources, representing the species' span by these provenances and environmental gradient by the test sites. 1.3.2. Examining geographic trends in growth Based on plot-means of the growth variables, multiple regression was applied site specifically on each growth variable using the SAS REG procedure, by relating the growth performance to geographic location of provenance origin (including quadratics and cross-products of geographic variables). Diameter and volume growth before year 20 was not examined due to high rate of missing observation in several hash test sites. Forward selection (a = 0.01) was used to screen the significant geographic factor(s) and thus examine the inherent 11 geographic trend(s) in growth traits. The extent to which the geographic trends was expressed, evaluated by partial and model R 2 (coefficient of determination), was related to age, growth trait and site mildness. In this study, site mildness is represented by the first principal component for site climatic conditions derived by PC A on the ten macro-climatic variables (see below). 1.3.3. Revealing phenotypic sensitivity to test site climatic conditions To reveal the general climatic sensitivity of all the provenances to site conditions, canonical correlation analysis was performed by the SAS CANCORR procedure on two groups of variables. One group is of the eight growth variables (i.e., HT3, HT10, HT15, HT20, DBH15, DBH20 VOL 15 and VOL20) and the other of the ten macro-climatic variables for test sites (see above). Diameter and volume growth measurements in year 6 and 10 were excluded due to the high rates of missing observations at several harsh sites in these years. Canonical correlation analysis is a multivariate statistical approach suitable for the study of relationships between two groups of self-correlated variables (Gittins 1985). As stated before, repeated growth measurements on same individuals are highly correlated, and so are many climatic variables. Therefore, the relationships between these growth variables and climatic variables should be addressed based on multivariate correlations rather than pair-wise simple correlations, because simple correlations do not take into account the inter-relationship among these highly correlated variables. Canonical correlation analysis produces two sets of canonical variables that are linear combinations of the original variables of the two groups, respectively, while maximizing the correlations between each pair of the canonical variables such that, the first pair of canonical 12 variables has the maximum correlation with each other, and the second has the second highest correlation, and so on. The relationships of the two groups of variables are evaluated by the canonical correlations between pairs of canonical variables, and the significance of each canonical correlation is tested by likelihood ratio F-test under the assumption of multivariate normality. The loading and cross-loading of each original variables onto the two sets of canonical variables were defined by the correlation coefficients between original variables and the canonical variables within and between groups, namely, canonical structures. Thus, the relative importance of each original variable to the relationships between the two sets of canonical variables can be quantified. The extent to which the original variables were represented by the canonical variables can also be evaluated by the proportions of raw variations in original variables explained by both sets of canonical variables within and between groups. To avoid the scale problem of the original variables, canonical variables in this analysis were derived based on the correlation matrices within and between the two groups of original variables at provenance-by-site means level. Consequently, the ratio of raw variation explained is based on the analysis with standardized original variables. 1.3.4. Unveiling the underlying major climatic factor for the geographic trends In order to unveil the major ecological factors underlying the observed geographic trends, redundancy analysis was applied on the two groups of variables, i.e., growth variables and climatic variables for provenance origins, through matrix algebra manipulated by the SAS IML procedure (Dr. Val Lemay, Faculty of Forestry of UBC, personal communication). In multivariate analyses, redundancy analysis is 'an alternative to canonical correlation analysis' 13 (Wollenberg 1977) which, instead of maximizing the correlation between two groups of self-correlated variables, maximizes the variations in one group cross-explained by, or say, 'redundant' on variations of the opposite group variables. Consequently, the linear combination of original variables in one group maximally accounts for variations of the opposite group but not necessarily accounts for variations of its own group to the maximum. In other words, redundancy analysis generates two sets of eigenvalues for determining the redundant variables of the two groups, respectively, such that the first redundant variable of one group maximally accounts for the variations of the opposite group (but not necessarily accounts for variations of its own group to the maximum), and the second redundant variable accounts for second maximal variations of the opposite group (but not necessarily accounts for the second maximal variations of its own group), and so on (see Wollenberg 1977 for details). Therefore, redundancy analysis is suitable for determining the variations in some multiple correlated traits purely due to some other suspected environmental sources that are also closely correlated. The notion that genetic variability is habitat-correlated in adaptive mode (Burley 1965; Roche 1969; Falkenhagen 1977) implies that, to some extent, growth variations among different provenances are conferred by seed origin's climatic conditions. Using redundancy analysis, growth variations among the provenances that are accounted for by provenance origin's climatic conditions can be quantified to the maximum. Similarly, the loading (represented by the correlation coefficient) of an original variable to the first redundant variable of its own group indicates the relative importance of this original variable in explaining variations of the opposite group. Thus, the major climatic factor(s) driving the genetic differentiation among the provenances in growth can be unveiled by comparing the redundancy loadings of the original climatic variables to the first redundant 14 variable of its own group. Again, the redundancy analysis performed here is based on the correlation matrices within and between groups to avoid the scale problem. Although redundancy analysis is more suitable than canonical correlation analysis for the interpretation of the variation in one group of variables explained by another group of variables, it also has a drawback that, as indicated before, when a redundant variable of one group maximizes the redundancy of the other group it does not maximize the redundancy of its own group. Consequently, the first redundant variable of either group could be a good representative of the other group, but not necessary of its own group. Therefore, it is not as advantageous using redundancy analysis to interpret simultaneously the mutual relationships of two groups of variables, as using canonical correlation analysis. On the other hand, canonical correlation analysis only maximizes the canonical correlations between two groups of variables, not the redundancy of one group upon another group. Therefore, it is more advantageous using redundancy analysis to determine the variation in one group (which is the interested group) explained by another group of variables (which are causal factors). In the previous section's analysis on the relationships between growth variables and site climatic variable, attentions were given to the multiple correlations between these two groups, not the redundancy of growth variation upon site climatic variability which could change substantially over different experimental settings. Therefore, canonical correlation analysis was used in that section's analysis. However, in this section, interests are on the determination of the magnitude. of explained growth variation by provenance climatic variations and in conjunction with this redundancy, the major climatic factors underlying the differentiation among provenances in growth traits. Therefore, redundancy analysis is more suitable for the purposes of this study. 15 1.3.5. Statistical criteria All the growth measurements were transformed into natural logarithmic values before analyses to achieve approximate normal distributions of the response variables. To avoid scale problem of growth and geoclimatic variables, data standardization (i.e., subtracting the mean and then dividing by the standard deviation of that mean) was performed whenever needed (e.g., multiple regression). All the significance related tests were performed under the assumptions of single variable and multivariate normalities and homogeneous variances of the response variables across different levels of experimental effects and geoclimatic regimes. The significance criterion was set at a = 0.01 level unless otherwise specified. All data analyses were performed with SAS procedures (SAS Inc. 1990). 1.4. Results and Discussions 1. 4.1. Partitioning of Variations in growth traits Based on plot-means, the eight growth variables (i.e., HT3, HT10, HT15, HT20, DBH15, DBH20, VOL15, and VOL20) were used to generate principal components of all the growth measurements. The PCA results indicated that the first principal component (GPC1) accounted for the majority (94.7%) of the original growth variations (Table 1-3). The contributions of the original growth variables to GPC1, represented by the correlation coefficients, varied greatly among the eight growth measurements (Table 1-3). However, the volume growth (VOL 15 and VOL20), also the most important growth trait, contributed the most to GPC1. Therefore, GPC1 is considered a good representative of all the growth measurements to be used as the response variable in the 16 following analysis of variance (ANOVA) to partition the growth variations into different sources of variation. Table 1-3.* Coefficients of the original growth variables with the first two growth principal component (GPC1 and GPC2) and the percentage of original variations in growth accounted for by GPC1 and GPC2, respectively. Coefficient of the original variable with Original growth variables GPC1 GPC2 HT3 0.1076 0.6545 HT10 0.1782 0.4447 HT15 0.2152 0.1697 HT20 0.2408 -0.081 DBH15 0.2174 0.0221 DBH20 0.2151 -0.3000 VOL15 0.6147 0.2555 VOL20 0.6173 -0.428 Variance explained 94.7% 2.6% *From the SAS PRINCOMP procedure on the eight growth variables based on covariance matrix, using logarithmically transformed plot means. Pooling the 11 test sites, assuming the response variable (GPC1) is normally distributed, and homogeneous variance exists across different sites, blocks and provenances, and all levels of these experimental effects were randomly chosen (though it is difficult to achieve these assumptions in a real situation of provenance trial), the SAS GLM model has high coefficient of determination (R = 0.87). The variance component of each variation source was estimated by the SAS MLXED procedure which computed the REML estimates for the variances of all the experimental effects under the random assumption, and from which the relative contributions of these effects to the total variations were calculated and listed in Table 1-4. 17 Table 1-4. ANOVA for the first growth principal component (GPC1) of eight growth measures. Variation Type III REML Relative Variance Source DF Mean Square F Value Pr > F Estimate Contribution Site 10 190.30 26.76* O.0001 2.1165 65.42% Provenance 42 7.82 8.14* O.0001 0.2559 7.91% Site x Prov. 159 0.96 2.22 O.0001 0.0871 2.69% Block(Site) 64 6.57 15.23 O.0001 0.3504 10.83% Exp. Error 1014 0A3 0.4256 13.15% * pseudo F -test Clearly, site effects were dominant, accounting for 65.4% of the total variation in growth (Table 1-4). This is not surprising as test sites are located in very contrasting environments (Ying 1997). In addition to the prevailing site influence, growth variation due to provenance variability was also highly significant. The ratio of inter- over intra-provenance variation can be approximated from the REML estimate of variance for Provenance and for Experimental Error which is the variation due to sources within block, provenance and site sources. Thus, the ratio was 0.2559 : 0.4256 » 1 : 1.66. Note this approximation likely overestimated the ratio because the inter-provenance variance contained both genetic and non-genetic components, and by using plot-means the within-plot variation which is also part of intra-population variation was excluded. The ratio estimate is thus much higher than that from Yeh and El-Kassaby's (1980) allozyme study with ten IUFRO provenances of the species, in which they estimated 92% of the genetic (allozyme) diversity reside within the populations. The difference, however, is a very common phenomenon in forest tree species in genetic partition based on morphological and growth traits versus biochemical markers (Morgenstern 1996). Block effects and provenance-by-site (G x E) interaction were also significant. Comparing to the sources of Site and Provenance, the G x E interaction had a low rate of 18 contribution (2.69%) to growth variation on average, although its magnitude was about 1/3 of that of Provenance. The large amount of degree of freedom for the Error term, on which the G x E interaction was tested, seems also to suggest that the significance of the G x E interaction was statistically marginal on average. However, the real situation of G x E interaction in growth of this species could be far from the results hitherto because of the following reasons: First, full expression of G x E interaction needs adequately wide geoclimatic span of test sites, as well as sampling range of provenances. In the above ANOVA, three series of the provenance trials were pooled together in order to include maximum numbers of provenance and test site for the scale of this study, while ignoring the differences of experimental setting among these series (see 1.2.1. of this chapter). Secondly, low expression of the G x E interaction on average could also be resulted from the pooled ANOVA which gives an equal weight to the provenances tested in different environments. So that in the above unweighted ANOVA, the 38 provenances tested in Series I had greater influences than the 14 provenances in Series II and III due to their large amount of observation, while the environments in Series I were too similar to allow for expression of G x E interactions (see below). More detailed ANOVA on GPC1 were also performed for the three series specifically, with results presented in Table 1-5. These smaller-scaled ANOVA clearly indicate the differences among the series. As mentioned above, the four sites in Series I are closely located with few environmental differences except for elevational variation. Consequently, the G x E interaction was not significant in Series I. In contrast, the three sites in Series II are located in very contrasting environments, therefore, the G x E interaction constituted a great portion of growth variation (35.5%) for Series II, while the effects of provenance variation were not 19 significant on average at all. The case in Series III was somehow intermediate between Series I and II since it has moderate environmental variations for the four sites. The contradictory of the above ANOVA results suggests that G x E interaction is subject to different experimental settings (the amplitude of environmental gradient and provenance sampling), and to different levels of analyses as well as different methods of analyses (e.g., weighted or unweighted ANOVA). It is imprudent and could be misleading if drawing conclusions based on a single approach of ANOVA analyses. In forestry practice, the G x E interaction is a major concern of provenance trial, because a provenance can retain high growth vigor only within certain geoclimatic ranges, and these ranges are critical in selecting seed sources for a planting area. As the expression of this interaction is conditional on different experimental settings, it can be of practical importance involving harsh inland environment in provenance trials (Ying 1997). The causes of the G x E interactions also deserve attention and will be discussed in later analyses (e.g., Chapter 3). Comparisons of the growth ranks of the 14 provenances tested in Series II and III (Table 1-6), where the G x E interactions were statistically significant, imply the causes of the interactions were mainly due to some reversal responses of a few provenances to weevil attack and harsh winter conditions, which certainly are very important to seed transfer. For instance, provenance Necanicum (No. 3012) was a high-yield provenance at most test sites, but dropped to the lowest rank at Maroon Creek (MN) which is the harshest site among these sites; while provenance Kitwanga (No. 3032) had low yield at mild sites, but relatively thrived at MN due to its noticeable resistance to white pine weevil attack (see Chapt.2). 20 Table 1-5. Partitioning of the general growth variation in GPC1 for each series of the Sitka spruce provenance trials in B C . Series I Series II Series III Var. Source DF MS F Var% DF MS F Var% DF MS F Var% Site 3 130.5 15.7 54.2 2 38.2 16.6 50.1 3 129.1 43.1 61.8 Provenance 37 2.2 11.0 8.6 9 3.0 1.4NS* 0.1 9 12.2 14.6 14.0 Site x Prov. 111 0.2 0.9NS — 17 2.4 16.7 35.5 27 0.8 3.1 2.8 Block (Site) 16 8.3 36.4 18.2 21 0.5 3.7 3.5 32 2.4 9.0 9.5 Exp. Error 590 0.2 19.0 161 0.1 10.8 288 0.3 11.9 model R 2 0.8231 0.8764 0.8850 # of plots 758 (760) 211 (270) 360 (360) % of data 57% 16% 27% * NS = not significant (a = 0.05). Table 1-6. Ranks of the general growth performance of the provenances tested in Series II & III, where significant G x E interactions were detected. Growth Rank H G a Series II NS MN HB Series III RS JU K T 1 2 3 4 5 6 7 8 9 10 a. Test site code (see Table l-l);Test sites were listed in descending order (left to right) of site climate mildness within each series. b. Provenance represented by Sitka spruce IUFRO number starting with "30-". 21 1.4.2. Geographic trends inherited by the provenances Based on plot-means, geographic trends among the 43 Sitka spruce provenances were examined site specifically by relating growth performances (HT, DBH and VOL) at different ages of each provenance to its origin's geographical variables, including quadratics and cross-products, using multiple regression analysis. Forward selection (a = 0.01) was applied in these regressions. The regression models were highly significant (p < 0.0001) for all the sites and growth traits with a few exceptions where geographic trends in growth were suppressed by winter harshness and/or weevil attack at sites J U and M N (see below). The significant 2 2 predictor(s) and the partial R's of them as well as the model R of each regression model were presented in Table 1-7. The results proved that there were strong geographic trends, mainly latitudinal, underlying the 43 Sitka spruce provenances in growth traits at different ages. However, there was also great variability of the geographic trends over sites, growth traits and ages, even when considering the latitudinal trend only. This variability implies that the expression of genetic control in growth traits was highly conditional upon environmental condition (which is one of the expression of the G x E interaction) and can change with age. The following implications can be drawn by comparing the significant predictors (i.e., 'Factor') and the R values over sites, ages and growth traits in Table 1-7: First, the geographic trends were highly site-dependent. At extremely mild sites (e.g., HG and HB) the latitudinal trend tended to be linear and explained a high level of variation in growth. This means that the more southern a provenance is the better the growth performance it has at these sites. At less favorable sites (e.g., QC III and KT) the latitudinal trend was quadratic (concave down), which means that provenances from either extreme north or south were less 22 favored than those from the central part of the latitudes (i.e., the species range). This was because at these sites southern provenances were more susceptible to winter injuries than the central ones which could still grow better than further northern ones. At the very harsh sites (with low winter temperatures, e.g., MN and NS) the latitudinal trend was suppressed, or substituted by longitudinal or elevational trends with very low levels of variation in growth explained by provenance origin's locations. Secondly, the latitudinal trend (LAT and/or LAT 2) varied with age and growth traits. Considering height growth only, the latitudinal trend was almost linear at early ages but later switched to be quadratic, most apparent at northern and inland sites (e.g., KT and MN). This could be explained by the fact that the southern provenances suffered winter injuries at northern harsh environments from year 3 to 10 and slowed down their height growth. The 20-year data did not show any evidence suggesting that the quadratic latitudinal trend will switch back to be linear when trees grown older. Instead, there were perceivable declines in the expression of the latitudinal trend in height growth on average with increasing age, except for a few test sites that are extremely mild and wet (e.g., HG, RS and QC II). This agrees with previous knowledge that heritability of some traits in trees, especially growth traits, declines with age (Namkoong and Kang 1990). 23 43 Cd o 13 •o ll 6 B Cj ?d o T3 CD L . Q . S cd o 43 43 O a u cd J 2 <o r—1 O 6 _o 'co to CJ o cs —1 O > 4= 1 / 5 -O § * co o £= « ; M 43 O £ c .5 cd "O S3 <u S CJ > 2 2 a o o CN a Q s w O cn 43 u 3 >i ed = I •C cd & .a > g 43 g j O m <D •!-> on cn U ' > o H a H S3 3 co cd CD 6 43 O <-bC >/-> u 43 e s-.O vo CN vo oo oo CN o CN oo o vo i> r- r- oo CN oo m co CN d d d d d d d d d © o o CN vo oo 00 oo oo 00 CN o CO </-) CN in CO <—1 o p p o <—1 p CO CN ! d O O d © d d d © © d d d © © 1 Pi r-p >/0 CO '—i o o © © © d o © OH p4 CM I* > cj W J —1 W # M ca cd > cj W * —1 > W * > cj W cd cd ca ca ca —I ca —1 ca p-1 CO a la —1 > 3 —1 J > w * ca oo o Ov o oo CN oo CN o o o o o o o o o 00 a o —i # ca —1 o C5 ca —1 la > CD w ta « —1 -J > w * —i > ca ca •J —1 T 3 a J3 O 00 a o * o 00 a o * cd > ID s * O O > CO W * td —1 00 c « O cd 00 c o -1 * td CO O CN CN CN '—1 00 £3 O o^o Jr1 o ca ca J J J ca -J ca oo S3 o >-) * * Id cd 00 a o Id cd —1 00 a o > W ta 00 o ca —1 J ca oo *H c cd O ca —1 oo S3 o # ta —1 oo cd O cd cd > W * —1 Os CO m CO CN o oo r~; p p CN o p I T ) vq o o o o d d d d d d d d d d d 00 a « O cd 00 c o * la ca oo vo 00 ca o o > ta £ oo oo N « S3 cd O vo CO cd oo O cd —1 oo Ov 00 S3 O I N * 4—* 4-* cd cd —1 J >/~> CO cd CO oo CO u oo CO u o oo CO u oo CO u C/l z •a S3 o z -a S3 CJ o Z CO oo CO CO CN OS '—1 o p CN p CN o p d d d d d d d d > OJ U * 00 S3 o —1 o 00 CN oo VO uo VO _ CO Ov CO CO oo o vq oo CN p o >—1 p CN CN p o CN p o d d d d d d d d d d d d d d d d d d d 00 S3 ^ O cd -4 J O VO o CN — —< 00 a- « c cd cd O —1 —1 J CN C3S z CN Further analyses with the site-dependency of the geographic trends confirmed that the extent to which the geographic trends expressed (i.e., the genetic characters expressed in growth traits) was closely associated with site mildness. PCA was used to reduce the ten macro-climatic variables to a few, informative and orthogonal variables for both test sites and provenance origins. The first principal component (climPCl) accounted for 62.8% of the total variation in the original ten macro-climatic variables after data standardization (Table 1-8). Except for MTWM, the macro-climatic variables contributed almost evenly to climPCl, indicating that climPCl is an effective index representing most macro-climatic variables. The negative coefficients of DDO and DAY with climPCl and the positive coefficients of the remaining climatic variables with climPCl imply that a higher value for climPCl means milder climate or more southerly located. Accordingly, the higher the climPCl value for a test site (referred to as sitePCl), the higher the average growth performance at that test site. The associations of site productivity (represented by the site means for HT20 in percentages relative to the highest site mean for HT20), site mildness index (sitePCl) and the 2 2 average model R (transformed into percentage, i.e., R %) are illustrated in Fig. 1-2 and Fig. 1-3. With a few exceptions (see below), the levels for these three different statistics varied concomitantly over the 11 test sites. This provided evidence that the milder the site, the higher performance achieved in height growth and the more pronounced the geographic trends exhibited at that site (since a higher model R means a greater proportion of growth variations are accounted for by the provenance's geographic origin). 25 Table 1-8. Results of PCA on the 10 macro-climatic variables for the 11 test sites and 43 provenance origins altogether, showing correlation coefficients and percentages of explained variance (based on standardized data, from the SAS PRINCOMP procedure). Climatic variables climPCl climPC2 climPC3 climPC4 DAY -0.330 0.261 -0.066 0.553 MAP 0.295 0.297 0.555 -0.128 MSP 0.216 0.490 0.487 0.243 MAT 0.380 -0.133 -0.026 0.007 MTCM 0.379 0.107 -0.204 -0.025 MTWM 0.097 -0.593 0.354 0.563 NFFD 0.366 0.094 -0.309 0.128 FFP 0.348 0.026 -0.309 0.477 DD5 0.259 -0.452 0.225 -0.215 DDO -0.371 -0.094 0.202 0.099 Variance explained 62.8% 20.0% 8.1% 3.9% (Note: climPCl = the 1st principal component of the 10 climatic variables, climPC2 = the 2nd principal component of the 10 climatic variables, and so on.) 26 Test site Fig. 1-2. Growth variation explained by provenance geographic location (average model R 2's in previous regressions) along with site mildness index and relative site mean for HT20 (%). i r -6 -5 - 4 - 3 - 2 - 1 0 1 2 3 Site mildness index (sitePCl) Fig. 1-3. Associations of site productivity (HT20(%)) and the amount of explained growth variation by provenance origin's locations (R**2(%)) = R2*100) with test site mildness (sitePCl). 27 Note that the number of provenances used in these regression analyses varied from 10 to 38 among different sites; that is, the sites in Series I test 38 provenances while those in Series II and III test only 10 provenances (Table 1-7, also see section 1.2.1.). Model R 2 declines with the number of levels for the predictive variables, especially quickly within the range of n = 30 (Draper and Smith 1966). Therefore, the average model R's for the four sites in Series I (i.e., QC I ~ IV) were relatively low for their site mildness compared to the remaining sites. The extra low level of model R for site HB was due to the occurrences of extensive white pine weevil attack during the measured years. That is, weevil attack might also suppress expression of geographic variation in tree growth. 1.4.3. Climatic sensitivities to site conditions The previous results indicate that site influences were dominant on the overall growth variation (Table 1-4). With ten macro-climatic variables defining site climatic conditions, the climatic sensitivities of Sitka spruce in different growth traits to different climatic variables can be revealed by canonical correlation analysis. As mentioned before, canonical correlation analysis is effective for presenting an overall view of the relationships between two groups of self-correlated variables in the multivariate sense. Two groups of variables were used in this analysis, one consisting of the eight growth variables (i.e., HT3, HT10, HT15, HT20, DBH15, DBH20, VOL15 and VOL20), and the other of the ten macro-climatic variables for test sites (i.e., SDAY, SMAP, etc.). Eight canonical variables were derived based on the correlation matrices within and between the two groups of variables, namely, Growl, Grow2, and so on for the growth canonical 28 variables, and Scliml, Sclim2 and so on for the site climatic canonical variables. The results presented in Table 1 -9 show that there were strong positive correlations between growth and site climatic variables, indicating that high growth performance is associated with milder site climatic conditions. The first pair of canonical variables, accounting for 59.2% of the total canonical variation, has the maximum canonical correlation as R = 0.954, and the second pair has the second highest canonical correlation as R = 0.872. Under the assumption of multivariate normality of the two groups of variables, canonical correlations between each pair of canonical variables were tested by F-tests on the likelihood ratios for the hypotheses that the canonical correlations between the current pair and the following pairs do not differ from zero. The probabilities that the hypotheses are true were presented in Table 1-9 under the column, "Pr > F (likelihood)", which indicate that all pairs of canonical variables were significantly correlated with each other, except the last one. However, for the ease of interpretation, only the first two pairs of canonical variables (i.e., Growl with Scliml and Grow2 with Sclim2) will be used in further analyses. These two canonical variables were selected due to their high eigenvalues (> 1) and the fact that together they accounted for 7 8 . 5 % of the total canonical variation (Jackson 1993) (Table 1-9). Canonical structures (Tables 1-10 and -11) of the first two pairs of canonical variables reveal the relationships between the original variables and the canonical variables. For the growth measurements, all the growth variables had moderately strong correlation with Growl, but relatively low correlation with Grow2 (Tables 1-10). This was also true with the correlations between growth variables and Scliml and Sclim2, since the first pair of canonical variables had the maximum canonical correlation. Focusing on the canonical structure of the first pair, Growl 29 was most strongly correlated with HT20 and HT15, followed by diameter and volume growth, and then HT10, but least strongly correlated with HT3. The same pattern of correlation ranks was found between growth variables and Scliml. This indicates that the environmental component of height growth due to test site increased while genetic variation among provenances decreased with age, that is, accumulative effects of site climatic conditions tended to become a determinant in height growth as age increased. Diameter and volume growth were also strongly influenced by site effects though not as strongly as height growth. On the other side of the canonical structures with site climatic variables (Table 1-11), SMSP had the strongest correlation with Scliml, while SMAP, SMAT, SMTCM and SNFFD ranked second in their correlations with Scliml. Therefore, the first canonical correlation implies that site climatic conditions, mostly defined by moisture regimes (SMSP and SMAP) and next determined by site winter harshness (SMTCM and SNFFD), had the strongest influences on later height growth, less strong but considerable influences on volume and diameter growth, and least strong influences on early height growth. The case with the second pair of canonical variables was quite different from that of the first one. On the growth variable side (Table 1-10), all the heights had negative correlations with Grow2 and Sclim2, while diameter and volume variables had positive but weak correlation with grow2 and Sclim2. Negative correlations between heights and Sclim2 declined sharply with increasing age, indicating height growth became more and more unaffected by those site climatic conditions defined by Sclim2. On the other hand, Sclim2 was strongly and negatively correlated with temperature conditions (SMAT, SMTWM, SDD5 and SMTCM), and had the strongest and positive correlation with SDAY. This implies that a higher value for Sclim2 means colder and/or 30 further north site conditions. Therefore, the canonical correlation between Grow2 and Sclim2 could be interpreted as that seedling height growth (before Year 10) was very sensitive to site temperature conditions, and diameter and volume growth were also influenced by site conditions (mainly in photoperiod, summer temperature SMTWM and heat sum SDD5) to certain degree. However, later height growth was almost unaffected by site temperatures and latitudinal location, and became to be more strongly influenced by site moisture conditions (i.e., Scliml), known from the first pair of canonical variables. Table 1-9. Canonical correlations and explained variance percentage of each pair of canonical variables (Cans) between growth measures and site climate variables. Cans CanR2 Eigenvalue Variance (%) Pr > F (likelihood) 1st 0.954 20.89 59.2 O.0001 2nd 0.872 6.81 19.3 O.0001 3rd 0.806 4.16 11.8 O.0001 4th 0.693 2.26 6.4 O.0001 5th 0.398 0.66 1.9 O.0001 6th 0.266 0.36 1.0 O.0001 7th 0.088 0.10 0.3 0.0002 8th 0.040 0.04 0.1 0.0167 Table 1-10. Canonical structure of the growth variables with the first two pairs of canonical variables. Growth Growth canonical variables site climatic canonical variables variables Growl Grow2 Scliml Sclim2 HT3 0.483 -0.570 0.471 -0.532 HT10 0.794 -0.260 0.776 -0.243 HT15 0.916 -0.174 0.895 -0.163 HT20 0.960 -0.064 0.937 -0.060 DBH15 0.801 0.120 0.782 0.112 DBH20 0.813 0.229 0.794 0.214 VOL20 0.855 0.013 0.835 0.012 VOL20 0.877 0.045 0.856 0.042 31 Table 1-11. Canonical structure of the site climate variables with the first two pairs of canonical variables, represented by the correlation coefficients (listed in the descending order). site climatic canonical variables Growth canonical variables Climate variable Scliml Sclim2 Growl Grow2 SMSP 0.864 -0.318 0.844 -0.297 SMAP 0.609 -0.463 0.595 -0.432 SNFFD 0.596 -0.453 0.583 -0.423 SMTCM 0.555 -0.471 0.542 -0.440 SMAT 0.441 -0.673 0.430 -0.628 SFFP 0.289 -0.250 0.282 -0.233 SDAY 0.167 0.787 0.163 0.735 S D D 5 -0.350 -0.584 -0.348 -0.546 SMTWM -0.337 -0.653 -0.329 -0.609 SDDO -0.300 0.202 -0.293 0.188 It is worthy of notice that the climatic sensitivities of different growth traits in Sitka spruce delineated in the above canonical correlation analysis are general to all the provenances tested. Different provenances, however, could have different climatic sensitivities even when tested at same environments. Preliminary regression analyses on the plot-means of the growth measurements of 11 frequently tested provenances (tested at least at eight sites) to site climatic variables (screening by forward selection at a = 0.01) had indicated that when these provenances were tested in British Columbia, the growth of southern provenances was more sensitive to winter temperatures than northern provenances did in general, and that the latitudinal line between southern and northern provenances for this purpose could be drawn at 46 - 48° N , approximately (see Appendix II). The redundancy, i.e., explained raw variance, appended to the above canonical correlation analysis (Table 1-12) shows that the first canonical variable for site climatic variables, which carries 34.0% original variation of its own group, explained 70.5% original variation in growth 32 variables at provenance-by-site mean level. This means that growth variation was well accounted for by site climatic conditions, which agrees with the previous ANOVA result that site influences were dominant on overall growth variation. On the other hand, the first growth canonical variable, which carries about 73.9% of original variation of its own group, accounted for 32.5% of the raw variation in site climatic conditions. Grow variations were well explained by the opposite group, suggesting high predictability of growth performance in Sitka spruce from planting site climatic conditions. Table 1-12. Redundancies of the canonical variables for both growth and site climate variables. Growth canonical variables Site climatic canonical variables Original Variation Growl Grow2 Scliml Sclim2 Growth variation 73.9% 1.7% 70.5% 1.5% Site climate variation 32.5% 17.3% 34.0% 19.9% 1.4.4. Underlying major ecological factors that differentiate the provenances Significant geographic trends, mainly latitudinal, were observed which underlie the growth performance of the 43 Sitka spruce provenances. Since genetic variability is habitat-correlated and genetically based (Burley 1965; Roche 1969; Falkenhagen 1977), it is desirable to examine the major ecological force(s) driving the geographic trends of growth among the provenances. In order to do this, growth measurements were related to climatic conditions of the provenance origins by redundancy analysis. As mentioned in section 1.3.4., redundancy analysis 33 is especially useful in genetics data analysis for determining possible causal relationship with environmental factors. Ten growth variables (i.e., HT3, HT10, HT15, HT20, DBH10, DBH15, DBH20, VOL10, VOL15 and VOL20) and the ten macro-climatic variables for provenance origin (i.e., PMAP, PMSP, and etc.) were used in this redundancy analysis. The involving of DBH10 and VOL 10 sacrificed some observations at a few harsh test sites, but would be helpful to examine the age trend of genetic control in these growth traits, which was not analyzed in the previous multiple regression analyses (see 1.4.2.). The analysis was performed on the within and between group correlation matrices that were obtained by the SAS CANCORR procedure based on original data at plot-mean level to avoid scale problem. The redundancy analysis results indicate that the amount of growth variation was poorly explained by the provenance origin's climatic conditions when all the test sites were pooled together (Table 1-13). Only 5.23% of the growth variation at plot-mean level was explained by the first redundant variable of climatic variables for provenance origin (Pcliml). Low expression of genetic effects in growth traits is not surprising, because heritability of trees in growth is usually low compared to other traits such as morphological traits (Falkenhagen 1977). This rate (5.23%), however, should not be considered as a surrogate for heritability. It could be greater if the input matrix for growth variables is at provenance-by-site level (in this case, the rate was 8.7%, details not shown here). The plot-mean level for growth variation was chosen to comply with the level of former ANOVA to make the following approximation, under the knowledge that different levels of growth variation could only affect the magnitude of explained variance, not the ranks the original climatic variables in contributing to the redundant variables. Recalling 34 the previous ANOVA result that 7.91% of the overall general growth variation (represented in GPC1, which carries 95% of the totally growth variations) was due to provenance variability (Table 1-4), a conclusion could be drawn that nearly 63% (= 5.23% * (7.91% * 95%)) of the genetic variability of the provenances in growth was accounted for by the climatic variation of provenance origin environments, acting as an agent of natural selection. This conclusion supports the assumption that genetic variability is habitat-correlated. On the other side, variations in provenance origin's climatic conditions were better explained (maximally 18.98%, Table 1-13) by the growth variables than vise versa. This is because there were less variations in the climatic variables than in growth variables as the former are actually on provenance level while the latter on plot-mean level. Table 1-13. Raw variations in growth and provenance origin place's climate conditions cross-explained by their opposite redundant variables from the redundancy analysis based on the within and between correlation matrices of growth measures and provenance origin climate variables. Growth redundant variables Prov. climate redundant variables Raw variation (%) Growl Grow2 Grow3 Grow4 Pcliml _PcJlirn2^  Pclim3 Pclim4 Growth Var. 5.23 0.26 0.16 0M~ Prov. Climate Var. 18.98 0.68 0.20 0.08 In this analysis, the redundancy structure (Table 1-14), assessed by the correlation coefficients between redundant variables and original variables, are probably more interesting than the redundancy itself as it presents an overall view of the contributions of the original variables onto the redundant variables and thus, reveals relationships between the original 35 variables of the redundant variables with the opposite group. Since none of the redundant variables, but the first one, accounted for the variations of the opposite group noticeably, only the structure for the first pair of redundant variables deserves concern. Again, as we are interested in the climatic factors that underlie growth variation among provenances, attentions were only given to the first redundant variable of the ten provenance climatic variables (i.e., Pcliml). Low rate of redundancy of growth on provenance climatic conditions resulted in low correlations between growth variables and Pcliml, none of which was greater than |Rj = 0.4 (Table 1-14). Relatively, HT3 was most closely correlated with Pcliml, followed by the diameters and volumes. There were rapid declines in the correlations of heights with Pcliml as age increased, which agrees with the previous multiple regression analyses of geographic trends in growth (Table 1-7). However, the correlation between diameter and volume with Pcliml remained as tree grew older, indicating more persistent genetic control in diameter and volume growth than in height growth, though not so strong as in early height growth. On the other side, the correlations of the provenance climatic variables with Pcliml are more meaningful as they indicate the relative importance of different climatic variables for provenance origins when their linear combination (Pcliml) maximally accounted for variation of the growth variables. Results in Table 1-14 indicate that all the climatic variables were closely associated with Pcliml, except PMAP, PMSP and PMTWM which define provenance origin's moisture conditions as well as warmth in summer. Among the closely correlated variables, PDDO and PMTCM (which define winter harshness of the provenance origin places) had the strongest correlations with Pcliml. The length of frost-free period (PNFFD and PFFP) as well as the mean annual temperature (PMAT) had second-highest correlations with Pcliml. PDAY, a 36 function of provenance latitude, also had strong negative correlation with Pcliml, indicating that provenances from northern areas were less favored than southern ones in general. PDD5 had less strong correlation with Pcliml than PDDO did, indicating that the amount of warmth of provenance origin environments was less important to growth performance than winter coldness. Therefore conclusions were made that temperature related climatic conditions, especially those defining winter harshness of the provenance origins, along with photoperiod_condition, contributed the most in differentiating growth performance of the provenances; while moisture related conditions and the amount of warmth of provenance origins did not affect much in characterizing growth performance of these provenances. The results can be well explained by the natural distribution of Sitka spruce. That is, with different temperature and light-climatic regimes while relatively even moisture conditions along the long, narrow strip of the Pacific west coast where Sitka spruce occurs naturally, the species could differentiate its populations in growth only by temperature and photoperiod regimes. Table 1-14. Redundancy structure (correlation coefficients) of the original variables with the first provenance climate redundant variables (Pcliml). Correlation coefficients of the Pcliml with Original growth variable Original provenance climatic variable HT3 0.359 PDAY -0.835 HT10 0.229 PMAP 0.368 HT15 0.153 PMSP 0.080 HT20 0.120 PMAT 0.845 DIA10 0.241 PMTCM 0.863 DIA15 0.231 PMTWM 0.199 DIA20 0.232 PNFFD 0.858 VOL10 0.232 PFFP 0.825 VOL15 0.204 PDD5 0.668 VOL20 0.209 PDDO -0.873 37 1.5. Conclusions 1. Great genetic variability in growth performance among the 43 Sitka spruce provenances were observed despite the dominant influences of site conditions. Nearly 63% of the growth associated genetic variability among the provenances was directly explained by the climatic conditions of the provenance origins. 2. Strong geographic trends, mainly latitudinal, underlie the growth performance of the provenances: southern provenances outgrew northern provenances, but the southernmost provenances did not fare well. The expression of the geographic trends was highly site, age and trait dependent. The milder the planting site, the stronger the geographic variation expressed, and the greater the likelihood that the latitudinal trend being linear. The strength of geographic patterns of variation was greatest in early height growth and declined rapidly with increasing age, but remained significant for diameter and volume growth at year 20. 3. The major ecological forces driving the geographical trend in growth performance of the provenances were predominantly temperature related, particularly for climatic elements related to winter harshness. That is, Sitka spruce populations were differentiated by temperature and photoperiod regimes of their origins. 4. The phenotypic expression of a provenance at different test sites was mostly correlated with the moisture conditions of the site, especially to summer precipitation. Moisture related site influences enhanced on later height growth while temperature related site climatic influences were influential on early height growth and diameter growth. 38 5. Growth variation was well accounted for by site climatic conditions, which implies high possibility of predicting growth from planting sites using climatic models. 1.6. References Box, G.E.P., Draper, N.R. 1987. Empirical Model-Building and Response Surfaces. John Wiley and Sons, Inc. New York. Burley, J. 1965. Karyotype analysis of Sitka spruce. Silvae Genetica. 14: 127-132. 1966. Genetic variation in seedling development of Sitka spruce, Picea sitchensis (Bong.) Carr. Planta. 99: 283-289. Campbell, R.K., Pawuk, W.A., and Harris, A. S. 1989. Microgeographic genetic variation of Sitka spruce in southeastern Alaska. Can. J. For. Res. 19: 1004-1013. Cooley, W.W., and Lohnes, P.R., 1971. Multivariate Data Analysis. Wiley, New York. Daubenmire, R. 1968. Some geographic variations in Picea sitchensis and their ecological interpretation. Can. J. Bot. 46: 787-798. Draper, N.R. and Smith, H. 1966. Applied regression analysis. John Wiley and Sons Inc., New York. Falkenhagen, E.R. 1977. Genetic variation in 38 provenances of Sitka spruce. Silvae Genet. 26: 67-75. . 1978. Parent tree variation in Sitka spruce provenances, an example of geographic variation. Silvae Genet. 27: 24-29 . and Nash, S.W. 1978 Multivariate classification in provenance research. Silvae Genet. 27: 14-23. Farr, W.A. and Harris, A.S. 1979. Site index of Sitka spruce along the pacific coast related to latitude and temperatures. For. Sci. 25:145-153. Gittins, R. 1985. Canonical Analysis: A review with applications in ecology. In: Biomathematics Vol. 12, Springer-Verlag, Berlin. Gregorius, H.R. and G. Namkoong 1986. Joint analysis of genotypic and environmental effects. Theor. Appl. Genet. 72: 413-422. Hall, P.M. 1994. Ministry of Forests' perspectives on spruce reforestation in British Columbia. In: Alfaro, R.I., Kiss G. and Fraser R.G. (eds) The white pine weevil: biology, damage and management. Symp. Proc. ppl~6. Jan. 19-21, 1994, Richmond, B.C. Canada. Henderson, D.M. and R. Faulkner (eds) 1987. Sitka spruce. In: Proceedings of the Royal Society of Edinburgh. (Section B, Biological Sciences). Edinburgh: 22 George St. 234pp. Holmes, G.D. 1987. Foreword. Proceedings of the Royal Society of Edinburgh, 93B, v. Hormann, R.K. 1987. North American tree species in Europe. J. Forestry. 85: 27-32. 39 Ilingworth, K. 1978. Sitka spruce provenance trials three years after planting in British Columbia. In: Proc. IUFRO Joint meeting of Working Parties, Douglas-fir, lodgepole pine, Sitka spruce and true firs. Vol.2 Pp 311-326. Vancouver, B.C. Canada. Jackson, D.A. 1993. Stopping rules in principal components analysis: a comparison of heuristical and statistical approaches. Ecology. 74(8): 2204 - 2214. Jolliffe, I.T. 1986. Principal Component Analysis. Springer-Verlag, New York, USA. Kendall, M.G. and Stuart, A. 1967. The Advanced Theory of Statistics: Vol2. Inference and Relationship. (2nd ed). Hafner Publishing Company, New York. Khalil, M.A.K. 1993. Performance of genotypic stability of Sitka spruce (Picea sitchensis (Bong.) Carr.) in Newfoundland, Canada. In: Ying, C.C. and Mcknight, L.A. (eds) Proc. IUFRO International Sitka spruce provenance Experiment. Pp 59-80. Edinburgh, Scotland. Ministry of Forests of B.C., Victoria, BC Canada and The Irish Forestry Board, Co. Wicklow, Ireland. Kovats, M. 1977. Estimating juvenile tree volumes for provenance and progeny testing. Can. J. For. Res. 7: 335-342. Lines, R. 1973. Sitka spruce IUFRO collection. Rep. For. Res. 42-45. MacSiurtain, M.P. 1981. Distribution, management, variability and economics of Sitka spruce in coastal British Columbia. M.Sc. thesis. Fac. For., UBC, Vancouver, B.C., Canada. 256p. Malcol, D.C. 1987. Some ecological aspects of Sitka spruce. In: Henderson, D.M. and Faulkner, R. (eds) Proc. Royal Soc. Edinburgh, 93B. Pp 85-92. Edinburgh, Scotland. Morgenstern, E.K. 1996. Geographic variation in forest trees: genetic basis and application of knowledge in silviculture. UBC Press, Vancouver, British Columbia, Canada. Namkoong, G. and Kang H. 1990. Quantitative Genetics of Forest Trees. Plant Breeding Review. 8 (Chpt.5): 139-188. Pojar, J., K. Klinka and D.V. Meidinger. 1987. Biogeoclimatic ecosystem classification in British Columbia. For. Eco. Manag. 22: 119-154. Rehfeldt, G.E., C.C. Ying, D.L. Spittlehouse and D.A. Hamilton. 1998. Genetic responses to climate for Pinus contora in Brititsh Columbia: niche breadth, climate change, and reforestation, (in press) Roche, L. 1969. A genecological study of the genus Picea in British Columbia. New Phytologist. 68: 505-554. and P.G. Haddock, 1987. Sitka spruce (Picea sitchensis) in North America with special reference to its role in British forestry. Proceedings of the Royal Society of Edinburgh, 93B: 1-12. SAS Institute Inc. 1990. SAS/STAT User's Guide, Version 6,4th edition. Vol .1 (pp 943) and Vol. 2 (pp 846). Cary, NC, USA. Wetherill, G.B. 1986. Regression Analysis with Applications. Chapman and Hall Ltd., New York. Wollenberg, A.L. 1977. Redundancy analysis — An alternative for canonical correlation analysis. Psychometrika 42(2): 207 -219. Worrell, R. and Malcolm, D.C. 1990. Productivity of Sitka spruce in northern Britain; 1. The effects of elevation and climate. 2. Prediction from site factors. Forestry 63: 105-118. Yeh, F.C. and El-kassaby, Y.A. 1980. Enzyme variation in natural populations of Sitka spruce (Picea sitchensis); 1. Genetic variation patterns among trees from 10 IUFRO provenances. Can. J. For. Res. 10: 415-422. 40 Yeh, F.C. and Rasmussen, S. 1985. Heritability of growth in 10-year old Sitka spruce. Can. J. Genet. Cyt. 27 (6): 729-734. Ying, C.C. 1991. Genetic resistance to the white pine weevil in Sitka spruce. Research Notes No. 106, Research Branch, Ministry of Forests B.C., Victoria, B.C., Canada. Ying, C.C. 1997. Effects of site, provenance, and provenance and site interation in Sitka spruce in coastal British Columbia. Forest Genetics. 4(2): 99-112. 41 2. Effect and probability of white pine weevil attacks on height growth of 14 Sitka spruce provenances Abstract: White pine weevil (Pissodes strobi (Peck)) attacks were recorded along with height growth of Sitka spruce (Picea sitchensis (Bong.) Carr.) periodically for 20 years since planting at four test sites of IUFRO provenances trials in British Columbia. Data were analyzed in two directions: examining the effects and persistence of weevil attacks on height and height increment rate and exploring the possible sources of variation accounting for the probability of weevil attack occurrence at different measurement intervals. Results indicated that both early and later attacks (before and after year 10) significantly slowed down height growth, though later attacks were about three times more frequent than early attacks. Weevils caused loss in height increment rate of around 30% over the measurement intervals. Height loss in absolute value was barely perceivable before year 10 as early weevil attacks tended to occur on taller than shorter trees within plantations. Height loss from weevil attack was 12 and 23% at year 15 and 20, respectively. Tree could resume normal height growth in about three years from early weevil attacks, but would not recover in terms of height growth rate until five to 10 years after attacks. The occurrences of later weevil attacks highly depended on previous attack history. Probabilities of weevil attacks on previously attacked trees were about three times greater than on unweeviled trees. Previous tree heights influenced the probabilities of weevil attacks at early years (before year 10) such that taller trees were more at risk to weevil attacks. The frequency of weevil attacks also varied among blocks within plantation, which implies non-random spatial pattern of 42 weevil activity. At young ages, weevil attack rates increased with site mildness and provenance latitudinal gradient (north to south), but these trends were not evident after year 15. This suggests that possibly height at the time of attack, not site mildness and provenance latitudinal location, caused variations of weevil attack rates among sites and provenances in early years. Different provenances did not exhibit different levels of weevil resistance until after year 10, suggesting weevil resistance observed in several provenances of Sitka spruce could be a stimulated biological response by weevil attack which has genetic background that varies among the provenances. Of the 14 provenances studied, three are recommended for further study as weevil resistant and/or tolerant provenances. Keywords: height growth; probability; provenance trial; Sitka spruce (Picea sitchensis (Bong.) Carr.), white pine weevil (Pissodes strobi (Peck)). 2.1. Introduction Sitka spruce (Picea sitchensis (Bong.) Carr.), a fast growing conifer native to the Pacific west coast of North America and a thriving plantation species in Great Britain (Herman 1987), could have been the most productive reforestation species in coastal British Columbia (BC) if not for the impact of the white pine weevil (Pissodes strobi (Peck)) on height growth at juvenile ages (Ying 1991; Hall 1994). Weevil damage results from larva and adults (occasionally) feeding on the tree leader and consequently, disabling the main stem growth and/or causing defects of the 43 stem or even deformity of the whole tree (Alfaro 1989a; Alfaro and Omule 1990; Alfaro 1994). Control of the weevil damage has been long concerns to forestry practices and the Ministry of Forests (MoF) of BC (Ying 1991; Hall 1994). However, none of the control techniques tested thus far, including shading, clipping, insecticides, or biological control, have proved to be sufficiently effective and practical (Cozens 1983; Hall 1994). There is increased interest in genetic control, alone and in combination with other control methods, since the discovery of apparent provenance differences in tolerance of weevil attack (Wood 1987; Alfaro and Ying 1990; Ying 1991). However, long-term benefits from genetic resistance is questionable until the mechanism of genetic resistance, the mode of inheritance, and the integration of resistant trees with silvicultural systems have been well understood (Ying 1991). Although weevil damage on Sitka spruce is frequently reported, the details of weevil attack pattern over years (i.e., temporal pattern) have seldom been studied, which is particularly valuable and interesting in understanding weevil behavior in relation to host selection. In early 1970s, the BC MoF launched a Sitka spruce provenance testing program, in cooperation with the international Sitka spruce provenance trials coordinated by the International Union of Forest Research Organizations (IUFRO). Extensive weevil attacks occurred at four out of eight IUFRO test sites located at coastal BC (Ying 1991). Weevil attacks were recorded at these four sites along with the height growth on individual trees periodically over 20 years since planting. The data provided a good opportunity to systematically investigate the temporal pattern of weevil attacks at different locations and ages and on different Sitka spruce provenances. In this chapter, the 20-year height growth along with weevil attack are analyzed to quantify weevil damage and height growth temporal patterns, and to examine the explanatory sources of variation 44 accounting for the probabilities of weevil attack occurrences across different levels of experimental factors and at different ages. 2.2. M a t e r i a l s Data from four test sites of Sitka spruce provenance trials in BC were applied in this study (for more detailed information regarding the trials, see Illingworth (1978) and Ying (1991 and 1997), or Chapter One). The four sites are Head Bay (HB), Kitimat Valley (KT), Maroon Creek (MN), and Nass River (NS). Except HB, the remaining three sites are located at peripheral inner coastal areas that are less favorable (harsh in winter) for Sitka spruce to grow (Table 2-1). The unfavorable growing conditions of these sites were assessed by summarizing the climatic data for 10 macro-climatic variables that were collected from the British Columbia Sitka spruce provenance trials, which totally involve the test of 43 Sitka spruce IUFRO provenances at 12 sites (see Chapt. 1 for details). Principal component analysis (PCA) was used for reducing these climatic variables into a few, informative variables for defining climatic conditions of test site and of provenance origin (see Chapt. 1 for explanation). The first principal component (climPCl) accounted for 62% of the original climate variations and was used as an index of test site mildness (sitePCl): a higher value for sitePCl is an indication of milder site. The experiment followed a completely randomized block design with nine blocks per site. Ten provenances were tested at each site with six common to all the four sites (Table 2-2). A total of 14 provenances were involved, which cover the species' main coastal range from southern Alaska to Oregon coast, extending inland into the hybridization zone of Sitka x white 45 spruce (Ying 1991; Table 2-2). Within block each provenance tested is represented by a 9-tree-row plot. Trees were planted at even space of 3 x 3 m. The number of trees planted per site is 810, that is, totally 3240 trees are planted. Extensive weevil attacks occurred during the first 20-year period after planting. Upon year 20, only 8.3, 34.8, 36.3 and 4.4% of the trees escaped from weevil attack at site HB, MN, KT and NS, respectively. Accordingly, the mortality rate at year 20 was 11.9, 14.7, 2.7 and 32.0% for these sites, respectively. The high mortality at site NS is partly (about 4%) due to road construction in those years. Tree height was measured to the nearest decimeter at the 3rd, 6th, 10th, 15th and 20th year after planting (referred to as HT3, HT6, and so on) on individual trees. At year 3, the length of tree leader was also recorded (referred to as Lead3). Since year 6, weevil attacks were observed and recorded during each measurement interval that is between two successive measurement years of height. For instance, weevil attacks during year 3 to 6 were recorded at year 6 and referred to as WV6, and same rule with WV10, WV15 and WV20. Ordinal codes were used to classify the intensity of weevil attacks. A value of 0, 1, 2, and up to 5 were given for no, one, two attacks and so on, respectively. A value of 6 was given to dead trees (i.e., the tree died either from 1, 2, or more attacks) which no longer had height measures. It should be mentioned that any weevil code value equal to or greater than 2 implies the attacks could have occurred at either successive years or non-successive years during the measurement interval. In this chapter, the five height measurements along with the four weevil attack codes at individual tree level were employed in the analyses. 46 Table 2-1. Locations and the first principal component values (sitePCl) describes site mildness of the four weeviled test sites. Site Code BGC^one Latitude Longitude Elevation (m) sitePCl2 Head Bay HB CWHvml 49°48" 126°28" 15 1.71 Kitimat Valley KT CWHwsl 54°12" 128°33" 100 -0.98 Maroon Creek MN CWHws2 54°46" 128°39" 600 -5.92 Nass River NS CWHwsl 55°04" 129°26" 15 -2.34 'BGC Zone = Biogeoclimatic Zone (BC's ecological classification of the land, see Pojar et al 1987). 2See Chapter One for explanation. Table 2-2. Names, IUFRO number, and place of origin for the 14 provenances tested at the four weeviled sites. Provenance IUFRO BGC a Latitude Longitude Elevation Tested atb Name No. Zone (m) HB KT MN NS Forks WA 3 (USA) 48° 04" 124° 18" 137 X X X X Hoquiam WA 8 CWHvml 47° 05" 124° 03" 5 X X Necanicum WA 12 (USA) 45° 49" 123° 46" 45 X X X X Brookings OR 18 (USA) 42° 15" 124° 23" 90 X X Yakutat AK 21 (USA) 59°31" 139° 42" 12 X X Duck Creek 24 (USA) 58° 22" 134°35" 30 X X Ward Lake AK 30 (USA) 55°25" 131°42" 15 X X Kitwanga 32 ICHmc2 55° 10" 127° 52" 660 X X Usk Ferry 40 CWHwsl 54° 38" 128° 24" 135 X X Inverness 44 CWHvh2 54° 12" 130° 15" 30 X X X X Link Road 49 CWHwhl 53° 30" 132° 10" 90 X X X X Holberg 56 CWHvml 50° 37" 128° 07" 30 X X X X Tahsis Inlet 61 CWHvml 49° 50" 126° 40" 0 X X Big Qualicum 62 CDFmm 49° 23" 124° 37" 0 X X X X a BGC Zone = Biogeoclimatic zone (Ecological zone in British Columbia): CWH= Coastal Western Hemlock Zone; ICH= Interior Cedar-Hemlock Zone (ICHmc2 is a hybridization zone of Sitka x white spruce). b see Table 2-1 for test site code. 2.3. Methods of Data Analyses 2.3.1. Examining the effects of weevil attack on height growth over measured years In order to examine the effects of weevil attacks on height growth over the assessed years along with experimental effects, height variation sources were tested by analysis of variance 47 (ANOVA), using the SAS GLM procedure based on repeated measures of height. As heights and weevil attacks were recorded on individual trees at five different ages, the use of repeated measurement analysis (Kuehl 1994) not only examines the sources of weevil attacks and experimental effects (i.e., Site, Provenance, Site-by-Provenance interaction, and Block within Site,), but also reveals age (time) trend and age interactions (i.e., Age x Site, Age x Provenance, and Age x Weevil attacks). Age trend of height growth reflects the tree's responses to experimental effects and to weevil attacks over years, which is very important for understanding the temporal host-pest interaction. In this analysis, heights were treated as different observations (not different variables) for the single dependent variable, HT, which is contingent with the measurement years (i.e., ages) under the variable, Age. Weevil attacks were tested as a discrete covariate along with other experimental effects. Leader length at year 3 (i.e., Lead3) was also included in the GLM model as a continuous covariate due to its significant effect on later height growth (see below). The interactions between weevil attack and the experimental effects were also included in the GLM model, which can be illustrated as follow: HT = / {(Site, Provenance, Site x Provenance, Block within Site), (Weevil, Weevil x Age, Weevil x Site, Weevil x Provenance, Weevil x Site x Provenance, Weevil x Block within Site), (Age, Age x Site, Age x Provenance, Age x Block within Site) and Lead3)}. Height growth-loss from weevil attack is more likely to be reflected by height increment rates (HIR) at different measurement intervals than by current height (HT) itself. Therefore, the significance of weevil effect on current height and it's lasting effect on later height growth were tested by Multivariate Analysis of Variance (MANOVA) using HIR's at different measurement intervals as multiple response variables, while weevil attacks (when applicable) as well as other experimental effects as explanatory variables, using MANOVA within the SAS GLM procedure. 4 8 MANOVA is an extension of ANOVA which is to include more than one dependent variable in the analysis (Bernstein 1988). When several responses to experimental effects are measured, one possibility is to perform an analysis of each dependent variable separately using ANOVA. However, this will result in very high Type I error rate (i.e., a) for all analyses combined and the analysis ignores the dependence among the response variables. Specifically, the error rate for the k independent tests results in a Type I error rate of (l-(l-a)*). However, by using MANOVA this kind of problem can be avoided as MANOVA treats all dependent variables simultaneously, while controlling for a specific a level. The methodology of MANOVA is analogous to that of ANOVA as it tests the equality of vectors of means, instead of one single mean as in ANOVA, over different treatments. In ANOVA, the total sum of squares is divided into that for treatment and for error. For MANOVA, the total sums of squares and cross products matrix (SSCP) is divided into the SSCP for treatment and for error. Similar to the test of Mean Square treatment to Mean Square error for ANOVA, MANOVA uses the ratio of SSCP for error to SSCP for treatment to test for significantly different vectors of means. In this analysis, however, we only have five repeated measurements (i.e., k = 5), by using a = 0.01 for separate ANOVA we still can achieve the overall level of a = 0.05 without severe inflation of the Type II error rate (i.e., P = 1-a) for the ANOVA. Therefore, both ANOVA and MANOVA were performed in analyzing the height growth rates for better understanding of the weevil attack impacts. The analyses were performed site specifically, as the four test site differed noticeably in site mildness, early weevil attack frequency and height growth rate. The extent of weevil caused height growth loss was evaluated by multiple comparisons of the height means and 49 means of height increment rate that correspond to different levels of weevil attacks by Student-Newman-Keuls (SNK) range test within the GLM procedure. There are a few limitations for analyses of this section. First, the four test sites are virtually from two series of the Sitka spruce IUFRO provenance trials (i.e., Series II and III, see Chapt. 1). When pooling the four sites with different experimental settings in the repeated measurement ANOVA, it is virtually intangible to isolate the effects of weevil attack from the effects of test site and provenance as they are already confounded with each another. Secondly, repeated measurement ANOVA has the advantage of being able to reveal the temporal patterns, but has the disadvantage of inflating the degree of freedom for the error term when sampling size is already large (in this case, five times the original measurement size), and thus cause inflation of the F test values for the variation sources tested directly against the error term. Third, significance related tests in the separate ANOVA for height increment rates were made under the assumptions of univariate response variable across experimental levels and weevil attack intensities, while those in MANOVA were made under the assumption of homogeneous variance-covariance matrix (SSCP) across different experimental levels and weevil attack intensities. These assumptions are not always valid in real world. Therefore, the AVOVA and MANOVA of this section should be considered referential rather than inferential. 2.3.2. Exploring the sources of variation accounting for weevil attack frequencies over measured years In order to explore the possible sources accounting for the temporal pattern of weevil attacks, logistic regressions were performed on weevil cases (i.e., weeviled tree and unweeviled 50 tree), using site mildness index (sitePCl), provenance latitude, block, previous weevil attack(s) and previous tree height before current weevil attacks as explanatory variables. The analysis was performed for each measurement interval specifically, using the SAS LOGISTIC procedure with backward selection (a = 0.05) for the predictive models. The general predictive model for the probability (p) of a tree being attacked at current measurement interval is Ln (p) = f (sitePCl, provenance origin's latitude, block, previous height, previous weevil attack). The use of backward regression is for the purpose of retaining as many as possible the explanatory variables in the model within the limitation of significance level (i.e., a = 0.05). There could be some minor differences in the results of multiple regression between backward and forward selection, as backward selection starts with the elimination of the independent variable which has the smallest contribution to model R among all the independent variables, while forward selection starts with the entering of the independent variable that has the biggest contribution to model R . Both backward and forward selection continue the same process as per their first steps for the remaining independent variables till that the remaining ones are all significant in backward selection model, while those are all not significant for entering into forward selection model. Therefore, backward regression is considered suitable for retaining as many as possible (set by a level) independent variables, while forward selection more suitable for retaining as few as possible independent variables in the regression model. In this analysis, the suspected sources accounting for variation of weevil attack probability deserves full consideration, therefore, backward selection was used in the logistic regression modeling. In all the analyses of this chapter, values for growth variables (HT's and HIR's) were transformed into natural logarithmic values in order to approach normal distributions. All the 51 significance related tests were under the assumptions of multivariate normality and homogenous variance and/or covariance of the response variable(s) across different levels of experimental factors and weevil attacks. Significance criterion for all the tests was set at a = 0.05 level if not specified otherwise. All data analyses were performed with SAS procedures (SAS Inc. 1990). 2.4. Results and Discussions 2.4.1. Experimental effects, Weevil effects, age trends and interactions With a high coefficient of determination (R2 = 0.899), linear model was constructed by using five repeated height measures (HT3, HT6, HT10, HT15 AND HT20) over years as the response variable, while the experimental factors (site, provenance, site x provenance, and block within site) as independent variables, in addition with three types of covariates that are 1) weevil attack effects and its interactions with age and experimental effects, 2) age and its interactions and 3) Lead3. Assuming levels of all the experimental factors were randomly chosen while those for the other effects are fixed, using the SAS GLM procedure, F tests and pseudo-F tests proved that the effects of all these sources were significant with respect to height variation (Table 2-3). However, it should be noted that those variation sources tested directly against the experimental error term could be inflated due to large amount of degree of freedom for this term (see section 2.3.1). These variation sources are the three-factor-interactions (e.g., WV*Site*Provenance) and Lead3 (referred to the EMS column in Table 2-3). The variance component for weevil effects ranked the highest level (45.02%), indicating that weevil attack became the predominant effects (instead of site effects as indicated by the 52 ANOVA in Chapt.l) accounting for height variations. That is, height variations at the four weeviled test sites were largely due to this biological effect, aside of experimental effects. The experimental effects still contributed considerably to height variations, except that Site effects were suppressed by weevil effects to certain degree. The provenance-by-site interaction endured the weevil effects, suggesting that when attacked by the weevil, different provenances could change height growth rates differently, sometime inversely, over environmental gradients. The exceptionally great F values for the age-by-site interactions suggest that mild site could have higher early heights but lower later heights due to more severe weevil attacks compared to harsh sites. The age-by-weevil interaction also had a large F value, though not a large variance contribution (0.27%), indicating there could be a temporal switch of the sign in host-pest correlation pertaining to tree height. The plot of the height means for this interaction (Fig. 2-1) shows that height at early ages (before year 8) was greater for weeviled trees than unweeviled trees on average. Height means for unweeviled trees began to surpass weeviled trees at year 8, approximately. This age-by-weevil interaction suggests that height was not merely the influenced term by weevil, but also could be a causal factor to weevil attack occurrences at early years. The significant interactions of weevil-by-site and weevil-by-provenance means that weevil effects were also conditional on site and provenance effects and that the latter are genetically controlled. The age-by-site and age-by-provenance being significant implies that the ranks of average heights for site as well as for provenance could change substantially over measured years in the presence of weevil attacks, which means that early height growth at weeviled sites is not reliable for determining rank of later growth. 53 However, one must notice the limitations in the present study. As stated before (section 2.3.1.), the experimental effects are confounded to each other to certain extent due to unbalanced setting, which caused virtual intangibility of the isolation of weevil effects from experimental effects as well as the partitioning of all the variation sources, under rigorous statistical criteria. In addition, the repeated measurement analysis introduced large amount of degrees of freedom for the experimental error term (five times greater than the size of using individual measurements independently), which resulted in very small (hence, sensitive) Mean Squares for the error term (MSE) and thus caused inflation of the F tests in which the MSE was used as the denominator for testing. Therefore, the interpretation for this ANOVA of repeated height measures should be considered as descriptive rather than inferential, which are based on the information from the GLM that has valid assumptions as stated before. Weevil Case weeviled — unweeviled 3 6 9 12 15 18 21 Age (years after planting) Fig. 2-1 Mean height trends of weeviled and unweeviled trees over measured years. 54 CD 43 -*-» -o 3 T3 C CD t --*-» CD ca T3 c ca o ca 1 > CD CD , 4 > CD 13 c CD s c CD a. xi CD CD 43 CD CD 4= cd C/l <u u 03 -a ii u u , ST It*! 4= 60 « 'CD O jg Cj-, » 1 0 >- "3 3 «S 8 1 CD « E | 4^ 1> I •8 ^ •B CD ^ _ CD 43 a.+3 CD CD I + J c d 1 O 72 < s O c^  CZ) CO ,' .2 CD cd s ^ H .5 o o o o t t t t w_ w u c3 c3 p3 c3 > > > > - ~ ON oo © CNI T f CN I - ; ir i oo vi © o o o o o o 00 o o o o o o o o o o * T l - ° - o -0 CJ s a o a CN & 5 5S CO I CD .1 VO CN VO ON oo VO CN d d w u o pa * w o < o 0. * w f-w o < > o Pi w o < w o < > b °-* LU W H Q So > O a. * w a a < > * w o < uS w o o < < w So w o o „ J cn 03 * * # W W W a a o < < < c7 a a + + + t t t t t t W W W M W, W, o3 c3 c3 c3 t3 <3 > > > > > > CN p >/-! C^ — TJ- CN CN —< oo CN i r i d d d CN d It * * J _ O -H _H O O o o o o o o o o o o o o o o o <=! o d d o o o o o o o t t t t t W W W W UJ c3 c9 c3 c3 c3 > > > > > » m N "J- O CO o o o p CN d d d d * # * ^ — — — © o o o o © © o o o o o © o o - 2 * vo -co ~ ?° 2 « CO CO CO CN o o o ON CO „ -H o o CO CN 3 (zi Cu H O O CAI pa K cj 2 VO VO CN O d TI-' — CN co r~ CO CO d d ON CN <£> CM > > >n oo CO _ o d ^ CN 00 ^-c o CJ ON I T ) ON r- co ON tN d <=> <=> s oo °^ o .g IP § 1 C3 < < w 2 ^ ^ Ed H I/) U 0 „ — l t/J 0 3 * * 5 w - J a + o fc w > oo o d o o o r--i n CN ON vo CO CO ON CN o o o VO oo CN >/-) VO CN CN w w w u a a a a • < < < oo CN s = <" cn I ^ (X CD <u £ u U >JO uo 2.4.2. Impacts of weevil attacks on height growth at different ages Using height increment rates (HIR's) at different measurement intervals as multiple response variables, the significance of weevil attacks were tested by ANOVA (on each HIR) and by MANOVA (testing all the HIR's simultaneously), aside of the experimental effects (omitted from Table 2-4). The tests were performed site specifically because of substantial differences in site mildness and early weevil attack frequencies as well as the provenances tested among these four test sites. The MANOVA results in Table 2-4 indicate that, at each measurement interval, weevil attacks significantly reduced height growth rates of the current measurement interval and subsequently (with only one exception at NS site where WV15 did not influence the growth increment rate of the same period and subsequently). However, the ANOVA results indicate the persistence of the weevil effects varied from age to age and site to site. Generally, early weevil attacks (WV6) did not influence height growth rate of later measurement intervals (i.e., HIR10, HIR15, and HIR20), but later attacks (i.e., WV10, WV15, and WV20) did (see Table 2-4). This possibly is due to stronger recovery ability of younger trees, and could also be due to the fact that early weevil attack is much less extensive than later attacks. It could be proposed that the impacts of weevil attacks lasted for no more than three years before year 6, but for about four years during year 7 to 10, and five to ten years during year 11 to 15. This agrees with Cozens (1983) that 'a major weevil attack may destroy three years of growth, as the destruction of the previous year's stem will kill the current year's growth as well'. Though not adequately precise, it appears that the earlier the attack, the less the number of years required for the resumption of normal height growth rate due to higher vigor of younger trees. However, recovery of stem form 56 is not clear in this study. Stem defects after weevil attack were once reported by Alfaro et al (1989a). From Table 2-4 one could also see that the lack of early attack (WV6) at sites KT and MN seems making it more difficult for the tree to recover from later attacks as compared to those at site HB and NS. This suggests that weevil tolerance could be possibly a stimulated response to weevil attacks of young trees. However, as for the limitations stated before (see section 2.3.1.), these test results should not be considered robust (i.e., indifferent of validations of the assumptions). Since weevil attacks were scored as ordinal variables, one might also wonder if height growth-loss corresponded to different levels of weevil attack intensity. This was resolved by using Student-Newman-Keuls (SNK) range tests on the means for height increment rate (HIR) and height (HT) that correspond to different levels of weevil attacks at different measurement intervals (Table 2-5). Results indicate that the rank of weevil attack reflected the rank of height increment rate but did not well reflect the rank of height itself. That is, once weeviled, height means for the trees did not vary much among different levels of weevil attacks. The reason for this phenomenon could be that weevils attacked more intensively on taller and robust trees at early ages than on shorter and less robust trees to ensure better food resources (Silver 1968, Gary et al 1971; Alfaro 1989b and 1994), so that the growth increment rates were affected but the absolute height values did not vary accordingly to different levels of weevil attacks. Hence, pooling different levels of weevil attacks in the analyses of the remainder of this chapter is justified. By pooling the levels of weevil attacks, weevil caused height-loss was manifested at a rate of 12 to 23% between year 10 and 20, and the loss in growth increment rate compared to 57 unweeviled trees was 20 to 32% at all measurement intervals after pooling the four sites studied (Table 2-5). The following implications, drawn from Table 2-5, deserve further consideration. First, weevil seems to be inclined to attack taller trees at early ages (before year 6) but later switched to shorter trees after year 10 (also see Fig. 2-1) because of repeated attacks (see below). Second, although weevil attack before year 6 caused loss in height increment rate during year 4 and 6, the ranking of total height at year 6 remains the same as year 3 between weeviled and unweeviled trees, indicating weevil attack occurred more often on taller trees but not severe enough to upset their rankings. Third, the influence of early attack lasted for no more than three years or so and tree could resume normal height growth rate from early weevil attack; but the effects of later attacks (year 10 and after) will last for about five to 10 years, in terms of height growth rate. 58 cn <u O 3 CD 3 3 3 4 3 43 o CD c o CD 43 O CD <+* O 43 > o .22 2 3 < 43 c — O 2 "° 3 CD .5 CD cn i_ CD O ~ O 3 P o <-l—I CD CD 43 60 s o T3 CD CD ex o cn cci CD CD CD o 3 cd > * « ^ ^  t: •3 U U g 8 6 • C CD U O c o CD 43 CD .5 CD . M CD «a c« . * * « I 3 ° cd T"! CD -4-> c tn O cn _ CD « H S w CD 3 S S cn r- l _ H J3 -3 • C o * " CD cd =3 •g a> cd §•? § cn tn O o « -« cd CD > o .2, s 4 3 cd 4 0 O 2> < 6 3 £ C .5 O - f 4= ^ .2 's ^ 1—1 O cd CM 5} Z I/-} Z O N M C O 2 SB [5 CO •7 *d-Z 00 M cn 2 r» C_( 00 CO f-SC c— Z io Z •* S C O 2 » aa C N 3°. Z H C N 2 f -CO <o X N O Vi f -o 00 — O N C N O O d 0 o o d d o o o o o o TC vo T- V O O r-p r r d d 1-H O V O O O C O O V O O N O O O N o d d ON - * V© ro O CO O O O O O O o d d <N vo © -<r 0 0 0 0 0 0 o d d rt N » O CN 00 O CN V I © © TT © © © i - H 0O V O f -© u~i vo © O N V O © © C N C O © o d d 0 0 0 0 3 s c c »-H © 00 r~-o » » © — CN — © —; © ON o d d © o in o ^ *H »N oi oi oi oi * N * N X X I I o 3 3 . u 43 3 cd cu 43 CJ ' 1 G O >/-> © © II a , § CJ o CJ CJ ,cd O 43 cj _3 "cd > o 3 (D z oo CD i c a B CD z c CD 3 T 3 C u +H o CN H I H EC a SO H o PC (u J3 « PC <L> 60 ed £ 6 crt OH W S en -—- "O g * (50 &0| '5 .S PC "° ^ cn 0 0 4 ) e t O SO OS ca <D *5 cn C S cn U od 06 CN CN O c CO 43 O 00 SO CN C O OS t— CN 00 ~ T> OS Xi ™ i n OS CN in ^ o M m a r - © so os r~ so 00 ca CB CO x> x> x> X> ea CI cn cn OS "*! in O l 00 CO so CO SO in in t~ in in in in i n m m in in so CN CN X) ca X) ca CN OS i n SO t— OS CO *—1 cn CO •<=t-(loss XI X) X) X) X) ea — OO C N 00 C N C N © © C N t— r^-X) C3 * * -s -s C O 00 •*r •—1 00 p 00 — ' co SO i n O 0 0 00 CO OS CN CN CN C O C O C O CN CN CN CN X) ca p —< p in so -"a-X) o cd cd cfl d W SO CN CN OS p " 5 f p CN in >n O C O so i n 00 C O C O C O C O C O C O C O C O C O cd cd cd Si i n 00 00 >n OS C O i n C O m i n m m T3 O X) i n C8 CN CO CO SO O CN w in —< os m so 00 os X i ca X> X) 00 o Os in — CN o O O X) X) X) C3 —< f ; CN >n in >n co r~ 00 o co co TJ- -^ r \O cd ca ca x> X> X> so sq vq 00 r~ i n o r~ o so o — — 00 00 00 ca ca ca ca ca ca C N — ; p — ; os r~; >n r~ o o\ 00 co O O -H O S O S o ^ 00 i n so C N ' C N 00 00 so ^ — C O SO C N >n 0 C O X) X) C O CN "3- CN CN 00 00 c/l c/l O C O C O C O S o i n i n i n in so — o o 0 X) X) O C N C N t— 00 S O O S O S C N t— C O C O C O - r t O CN xi X) xi xi ca o •^ f C N in C N C\ i n ^ - iri C O C O C O co - ^ f o XI C O C O t— O — O S O X) ca sq o\ r-; co coi r~ 00 so t~ O O S o 00 00 ca ca ca ca ca —< p 00 C O 1—1 so —I o C N i n os o o o o 2 * § I 2 so g CO CN - H CS If) CO CN — O o CN >-CS T T ro CN — O C Q > co •a ~ ca o II > a <u « CJ ca 11 0 II ca CN u > u c ~ II o „ w 'S 60 w cd > C o SO 2.4.3. Explanatory sources accounting for weevil attack occurrence By pooling the levels of weevil attacks, trees at each measured year were categorized into weeviled or unweeviled trees. The frequency of weeviled trees varied among different sites, provenances, and ages (Figs. 2-2 and -3). When pooling the four sites together, the frequency of weevil attacks at each measurement intervals was 4.4, 27.5, 69.2 and 75.9% during year 3 to 6, year 7 to 10, year 11 to 15, and year 16 to 20, respectively. This shows that later attacks (after year 10) were about three times more extensive than early attacks. The probability of weevil attack occurrence is therefore examined by logistic regression for each measurement interval separately. Weevil case is used as the binary response variable which is encoded by transforming the original weevil codes into a binary variable, such that, 1 = unweeviled tree, 2 = weeviled tree (pooling all levels of weevil attacks). Previous weevil attacks might induce more attacks in later years. The age x weevil interaction illustrated in Fig. 2-1 suggests that, to certain degree, previous height itself accounted for the occurrence of weevil attack at early ages (i.e., weevil seems to attack taller trees at early ages). Therefore, to examine possible sources of variation accounting for the probability of weevil attack, logistic regressions were performed on experimental effects as well as on previous weevil attacks, and previous height immediately before the current measurement interval. The experimental effects used are Site, Provenance and Block within Site (excluding Provenance-by-Site interaction for ease of interpretation), where, the sites were represented by the values for site mildness index (i.e., sitePCl) while the provenances by provenance origins' latitudes (PLAT). This was chosen because Sitka spruce has provenances differentiated primarily by temperature and photoperiod regimes along the north-61 south Pacific coastal line (see Chapt.l for explanation). Backward selection (a = 0.05) was applied to the logistic regression models to eliminate as few as possible insignificant variation source(s). The results are summarized in Table 2-6. As indicated in Table 2-6, the prediction models were all highly significant (p < 0.0001), which means the probabilities of weevil attack occurrence were well accounted for by the models. However, the goodness-of-fit tests (Hosmer and Lemeshow 1989) indicate that the associations of predicted probabilities and observed responses were significantly departed, except for the earliest model (i.e., the one predicts Prob(WV6)). Nevertheless, the concordant rates (i.e., the percentages of correctly predicted weevil cases) were around 80%, suggesting the predictions are still reliable, especially when taking into account the undetermined targeting of weevil attacks. The maximum likelihood estimates and their associated chi-square significance tests showed that the suspected sources significantly accounted for the variations of weevil attack probability (Table 2-6). Previous weevil attacks was the number one source accounting for later attacks, especially after year 10. The estimated coefficient for previous weevil attack(s) accounting for the probability of attacks during year 7 to 10 (i.e., Prob(WVlO)) was 2.84, and those for year 11 to 15 and year 16 to 20 were 2.88 and 3.81, respectively (Table 2-6). The mean predicted probabilities of weevil attack occurrence for previously weeviled and unweeviled trees at different measurement intervals were computed and presented in Table 2-7, from which one could see that the probabilities of weevil attack(s) occurrences on previously weeviled trees (immediately before the current measurement interval) were about two to three times greater than on previously unweeviled trees. This is in accordance with Alfaro and Ying (1990) and Alfaro et 62 al (1993) who studied the association of weevil attack probability on a tree and the distance of that tree from the nearest previously attacked tree, and reported 're-attack' tendency of the weevil. Table 2-7. Mean Predicted probabilities of weevil attack(s) occurrence for previously weeviled and unweeviled trees at different measurement intervals. Previous weevil attack case Probability of weevil attack(s) Prob(WVlO) Prob(WV15) Prob(WV20) Year6: Unweeviled 0.2713 0.6588 0.7146 Weeviled 0.8353 0.9643 0.9446 YearlO: Unweeviled 0.5592 0.6499 Weeviled 0.9433 0.9257 Yearl5: Unweeviled 0.2889 Weeviled 0.9550 63 o -O c t - - T i - ^ o o o OS O O p p o © © © OS OO ^ oo T|- m o re o — in CN © d o so SS oo <^  CN 2 t— OS O T | -CN — < © © 5 + i + i 2 — — oo o —' r- o T f o o o © o m m so d cn o cn CN t -s© —• © d +1 +1 T|- oo CN 00 —' CN in < 2 o o o d o so t— SO >n cn d +1 TI-oo CN '1 CD 3 o T l -in so oo d < 2 o o o o o +1 •<t so O fi SO 1-1 - H > > > £ £ £ o o o CN 00 >n cn d +1 OS cn sS —' I ca -St 3 o o o © d m r- o — n ^ K o — cn < < +1 2 2 8 in so — oo oo o —i cn o cn p o o d d O CN P in so o o d d < +1 +1 S £J 2 P © o o — SO o o o ° ~ © o p o o d d so in cn «n- ^ 00 so CN r-» T|-so cn -H .—. p o p o d d +1 +1 +1 so cn CN in o CN o d d o g b © © © p § o oo os so TT >n «i h CN m cn o CN O O ITi SO i-H FH H H H I I I o +1 +1 r~ o +1 id m o SO so d d i d K ca 5 C U H •* xpe QH cu LA blocl '3 blocl CN (N tin * Q ©" —r ? ° © © — © 60 O CN CN SO p S in © r-~ © £ <=> 11 ^ I—I 3 60 O HJ CN © /—, — in © Q CN " °° © 00 o —I o II v 60 O J CN cn r- " <^  o m © r~ © °° © " V 60 ^ O —1 CN 6* .5 SO b 00 Q 00 o o © © V 3 | T f t-; CN 00 a 00 ^ 11 © b cn Q ts v -SI T f t— SO >n a SO CN OO II cu td z: v e eB T 3 1— O O c o U a oo O S II U-r~ Q r-II >n T t " co -4-» CC CN cn ai © II C CO "H 3 ou CN a OS o O II u a s 03 C CB O a OS 00 OS „ o J3 O en os J> os E « to ^ •a c oo <« < <U o 15 £ . > « o •*-* o .23 2 tO c n J ci 60 O 0 1 H-J cn CN eo 1 a II -g J o 6 0 O —1 II T f so Previous tree height also contributed significantly to the encounters of weevil attacks at early ages, especially before year 10. Positive estimates of the coefficients for the previous heights in the models (Table 2-6) indicated that taller trees at early ages were more susceptive to weevil attacks than shorter trees. This supports previous report that the weevil intends to feed on the highest trees with the longest, thickest leader (Alfaro 1989a). However, there were no significant associations between previous height and weevil attack occurrence after year 10. As weevils inclined to re-attack (i.e., attack those previously attacked), there was a switch in the sign of the correlation between previous height and probability of weevil attack before and after year 10. Taller trees had higher susceptivity to early weevil attacks which caused growth-loss to the trees and enhanced probability for those tree to encounter later attacks due to re-attack tendency of the weevil. The density plots of weeviled and unweeviled (i.e., healthy) trees corresponding to the immediately previous height at different measurement intervals (Fig. 2-2) also revealed the height 'preferences' of weevil attacks at different ages. Another hypothetical reason explaining this phenomenon can be proposed here as that, weevils can attack successfully only within certain range of tree height, namely 'height-window'. Before year 6, trees seldom suffered from attack as tree height had not reached the height-window for weevil attack. Younger taller trees who reached the lower height limit early in their life became the ones that encountered weevil attacks first. After year 15, trees that had grown above the upper height limit for weevil attack were more likely to escape the attack. According to the data under this study, the tree height range within which extensive weevil attack occurred was between 0.89 and 4.4 m (Table 2-8). However, the observed large standard deviations of these height means suggest that the actual 65 height range prone to weevil attacks was much wider and undetermined. The age of the tree reaching the lower height limit for weevil attack was around year 6 to 8, and this was also the critical period for weevil controlling (e.g., clipping or eliminating weeviled trees). It could be very difficult to control the pest after year 10 since weevil populations exploded at this stage (deduced from the difference between the numbers of trees weeviled and unweeviled, Table 2-8) and repeated attacks became very common. However, the number of years for Sitka spruce seedlings to reach the proposed 'height-window' largely depends on planting site conditions and seed source used, therefore, the tree age should not be considered as the only criteria for timing of weevil control. Block effects on the probability of weevil attack occurrence were significant over all the measurement intervals (Table 2-6), indicating group activity tendency of the insect, or micro-environmental variation affecting weevil activities. This agrees with the observed pattern of weevil population aggregation in the dispersal studies using mark and release techniques (Harman 1975). 66 Table 2-8. Mean and standard deviation of the mean of previous height (in decimeter) for weeviled ('yes') and unweeviled ('no') trees immediately before weevil attack occurrence at each measurement intervals. Previous WV6 WV10 WV15 WV20 height no yes _ no yes no _ n o Y e s HT 3 : ~ mean 7.6±3.1 8.9±4.1 SNK b a n 3098 142 HT6: mean 15.3±6.7 15.2±6.1 SNK a a n 2349 891 HT10: mean 28.8±9.8 29.4±12.1 SNK a a n 997 2243 HT15: mean 49.3±15.2 43.9±15.2 SNK a b n 780 2460 67 Year06 YeaMO 100 noweevil weeviled Case noweevil weeviled Case Year15 Year20 100 100 noweevil weeviled Case E -o g ) cu I CD noweevil weeviled Case Fig. 2-2. Density plots of weeviled and unweeviled trees at current measurement interval corresponding to the immediately previous height at the four measured years (illustrated on top of each plot), respectively. 68 Site mildness also accounted for the occurrence o f weevi l attacks at early ages before year 10. Positive estimates o f the regression coefficient with s i t e P C l (i.e., site mildness index) suggests that weevils were more active at milder sites in early years (Table 2-6). This could possibly due to the fact that trees at milder site grew faster so that reached the lower height limit o f weevi l attack earlier than at harsher site. However, when the tree grew above the lower height l imit o f weevi l attack, site differences diminished so that after year 10 there were no more significant differentials in weevil attack frequency among different sites (Table 2-6 and F ig . 2-3). The outbreak o f weevi l attack reached 'equi l ibrium' (Alfaro 1994) during year 15 to 20 after its 'exponential increase' o f attack frequency since year 4 (Fig . 2-3). There has not yet a perceivable decline o f attack frequency at these sites that is supposed to occur around 30 to 40 years after planting (Alfaro 1994). 100 CD CU •D > cu cu i • wv6 • wv10 • wv15 • w v20 KT MN Test site code NS Fig. 2-3. Frequencies of weevil attack occurrences in measured years and at different test sites. Provenance latitude ( P L A T ) seems also related to early weevi l attacks. T h e negative estimated regression coefficients with P L A T up to year 15 suggest that southern provenances 69 were more subject to weevil attacks than northern ones before year 15 (Table 2-6). However, after year 15 this latitudinal trend also diminished, that is, the effect of PLAT no longer significantly affected the probability of weevil attack occurrence and hence are removed from the predictive model (Table 2-6). Previous analyses (Chapt. 1) have indicated that the latitudinal trend of height growth among different provenances declined as trees grew older. Therefore, it might be the inherited height growth rate of the provenances, not inherited weevil resistance, which induced the latitudinal trend of weevil attack frequency at early ages. Aside from the effects of provenance latitude, different provenances did show different levels of weevil resistance after year 10 (Fig. 2-4). The observed mean frequencies of weevil attacks on different provenances indicated that provenances Nos. 3032, 3021, 3062 and 3061 were relatively resistant to later weevil attacks which, based on previous analyses (see above), were more extensive and influential on height growth than early attacks. Age trend of the weevil attack frequencies also indicated that these provenances did not exhibit perceivable weevil resistance until after year 10. Within each provenance, the resistance level during year 11 and 15 was very consistent with that during year 16 and 20 (Fig. 2-4). The above results suggest that weevil resistance and/or tolerance, if any, in some of the Sitka spruce provenances, are stimulated responses by weevil attack (during year 6 and 10) and are genetically controlled, so that the abilities of developing the stimulation varies among different populations. The mechanisms of weevil resistance and/or tolerance, however, are not clearly understood yet. Similar induced resistance to the weevil attack was also found in white spruce and discussed by Alfaro (1995), which gave us the hope that Sitka x white spruce hybrids might be more resistant to weevil attack than non-hybrids. 70 100 90 80 70 60 50 40 30 20 10 0 LdUJJ j j i l l 32 21 62 61 56 44 12 40 3 49 Provenance (IUFRO No. minus 3000) 30 24 Fig.2-4. Frequencies of weevil attack occurrences at measurement intervals and on different provenances. By examining the height means (HTs) and mean height increment rates (HIRs) of the 13 provenances (excluding No. 3018 which almost jeopardized at these sites) at different ages, the ranks of height and height increment rate of the provenances were presented in Table 2-9. It is noticeable by this table that provenance Nos. 3008, 3032 and 3062 suffered from weevil attacks but still remained as the tallest or taller provenances, which means that they were more tolerant or resistant to weevil attacks. Provenance Kitwanga (No. 3032), a poor grower at other test sites where the weevil did not attacked (data not shown), exhibited exceptionally strong resistance and tolerance to weevil attack among the studied provenances. It kept the lowest level of weevil attack frequency during outbreak of the weevil (Fig. 2-4), while its height growth rate topped the other provenances over the measured years and consequently, it joined the 'tallest group of provenance' at year 15 and almost surpassed the tallest provenance (No. 3062) at year 20 (Table 2-9). Provenance Hoquiam (No. 3008) showed possible weevil tolerance in that it kept both highest frequency of weevil attack and top class of its height. In contrast, provenance Big 71 Qualicum (No. 3062) showed its possible weevil resistance in that it maintained lower level of attack frequency during the outbreak of weevil attack while its height kept at top class among the provenances. The mechanisms of possible weevil tolerance and/or resistance in provenance Kitwanga and Big Qualicum are most interested to forest geneticists, since Kitwanga is a suspected hybrid of Sitka x white spruce (El-Kassaby et al 1988), while Big Qualicum is a high yield 'generalist' among the provenances tested (Ying 1997). Further studies are highly recommended on the three provenances, i.e., Kitwanga, Hoquiam and Big Qualicum, to exploit the genetic potential of weevil tolerance and/or resistance in Sitka spruce populations. 72 0 0 CD • c ca £ CD c cu •o 3 -4-* oo cu % e o CD oo ca o CN H se H 8 H u o CD I D •»= I C - 'co .S « ., CD s id | 1 E to CO 4) o CU CD CO Q . « ~7 a cn H o CN a s s s "O 1) C X3 ca D 0 0 ^ £ u •n x: £ « a « u 3 DS £ 2 E •as CD 1) . ca ON *-l CU H £ O N N O i n oo 00 i n O Tf CN CN i n CN >n CN O N cn Tf N O O N d cn cn cn >n N O r~ cn Tf m Tf i n v-> i n v-i N O N O N O •8 •« ON NO CN <n CN CN 00 p C-; in Tf ON NO TT NO CN CN c n © Tf v i cn CN Tf NO c n Tf NO c n NO Tf >n oi Tf cn >n c n O N CN O N O CN O p i n O CN p r~; Tf CN v i CN CN od CN Tf CN t— CN v i CN ON CN CN CN CN cn cn v i cn CN c n * <~ <~ -B t> £ 4^—» M-I -a CU CU o T D T 3 X ) U O S o u X> Xi ca ca o u o 0 o 0 0 " i= £ -§ £ -§ £ "§ -§ ca co ca O u CJ 0 O " -g £ -§ £ -§ « -S « « •§ 13 1) — oo CO O O ^ X) o CO "2* -S la ^ ca o X) ca T r o o v - j O N v - > r " - ; p - - N q c N v - > — " T r c N C N c n v i c n c — o d o N N d 00 T3 >t-i O CO xi -a X) T3 CB O * X) CB X ) CB p p v - i c n — T f t - - - c x ) v ^ T f T r O N , > n v i T f v i t ^ N O N O N d r - ~ N o o d o \ O N o d ca X) cu co T J CU <+-< o X I c^ -i -rt c^ cu o cu t~- T f T|- c n oo —< CN O N — O T f O N v i d T f v i — CN cn cn cn T f c n c n cn T f ca x> -a O T3 X) T 3 CU * -s O N c n CN cn O N — cn 00 Tf ON Tf N O N O d O N N O Tf" Tf cn Tf Tf Tf v-i v i Tf d Tf N O Tf Tf cn O N cn cn N O _, cu o "p, o _ a T3 X) g X> - ° O O O CN v-i Tf 00 p —; O N O N cn 00 vn ON v i 00 v i Tf cn v i d d O N 00 O N 00 00 00 oo O N C B X J O C J C J O O O O O O O O 00 p O ON oo NO CN CN cn cn CN CO o cn O cn O CN O CN O d o O N ^ cn CN r -K g N O Tf Tf O N ^ O N O N O N N O O «n m ^ , T f i n c N v - i c n i n . - O N - o t - ^ c n T f O c n c n o O N ^ — T f O c N CN — — H C N — - C N — CNCN m c N c N i r i f n T f T f T f N o " 1 0 0 — NO 2.5. Conclusions 1. Weevil attacks significantly slowed down height growth of young Sitka spruce at each current measurement interval. The average loss in height growth rate from weevil damage was around 30%. Weevil caused height loss was barely perceivable before year 10 because early weevil attacks tended to occur on taller than shorter trees within a plantation. Height loss was 12 to 23% between year 10 and 20. However, once a tree was attacked, the loss in height did not vary greatly among different levels of weevil attack intensity. It took about three years for the tree to resume its normal height growth rate from early attacks (before year 10), but five to ten years to recover from later attacks. 2. The probabilities of weevil attack highly depended on age, previous weevil attack history, and previous tree height at early ages. The average frequency of weevil attacks at each measurement intervals was 4.4, 27.5, 69.2 and 75.9% during year 3 to 6, year 7 to 10, year 11 to 15, and year 16 to 20, respectively. Weevil attacks in later years (after year 10) were approximately three times as extensive as in early years (before year 10). Early weevil attack occurrences were highly correlated with tree height such that, the taller the tree the higher the risk it had to encounter weevil attack. The occurrences of later weevil attacks highly depended on previous attack history. The risk of repeated attack on previously weeviled tree was about two to three times greater than on previously unweeviled tree. The frequency of weevil attacks also varied among blocks within plantation, which implies group activity tendency of the weevils, or spatial sensitivity to micro-environmental differences. 3. Dependencies of weevil attacks on both a multivariate measure of site mildness and provenance latitudinal origins were perceivable only at early ages, but not after year 15. Weevil attacks were more severe at milder than harsher site, and on southern provenances than on northern ones during early years. This suggests that tree height at attack, not site mildness and provenance latitudinal origin, might be the cause accounting for different attack rates among sites and provenances at early ages. Different provenances did not exhibit different levels of weevil resistance until after year 10, suggesting that weevil resistance, observed in several provenances, could be a biological response that is stimulated by weevil attack and has genetic background varying among the provenances. Three of the 14 provenances tested, i.e., Kitwanga, Hoquiam and Big Qualicum, were recommended for further study as weevil resistant and/or tolerant provenances. 2.6. References Agresti, A. 1984. Analysis of Ordinal Categorical Data. New York: John Wiley & Sons, Inc. Alfaro, R.I. 1982. Fifty-year-old Sitka spruce plantations with a history of intensive weevil attack. J. Entomol. Soc. British Columbia 79: 62-65. . 1989a. Stem defects in Sitka spruce induced by Sitka spruce weevil, Pissodes strobi (Peck). In: R.I. Alfaro and S.G. Glover (eds). Proc. IUFRO Working Group on insects affecting Reforestation. Vancouver, B.C., pp. 177-185. . 1989b. Probability of damage to Sitka spruce by the Sitka spruce weevil, Pissodes strobi. J. Entomol. Soc. B.C. 86: 48-54. . and C.C. Ying. 1990. Levels of Sitka spruce weevil, Pissodes strobi (Peck), damage among Sitka spruce provenances and families near Sayward, British Columbia. Can. Ent. 122: 607-615. . and S.A.Y. Omule. 1990. The effects of spacing on Sitka spruce weevil damage to Sitka spruce. Can. J. For. Res. 20: 179-184. . 1992. Forecasting spruce weevil damage. In: Ebata, T. (ed.) Proceedings of a Spruce weevil Symposium. Terrace, BC, March 12, 1992. BC MoF, Price Rupert Region. Pp 10-16. 75 . Hulme, M. and Ying, C.C. 1993. Variation in attack by Sitka spruce weevil, Pissodes strobi (Peck), within a resistant provenance of Sitka spruce. J. Entomol. Soc. B.C. 90: 24-30. . 1994. The white pine weevil in British Columbia: biology and damage. In: Alfaro, R.I., Kiss G. and Fraser R.G. (eds) The white pine weevil: biology, damage and management. Symp. Proc. Pp7~22. Jan. 19-21, 1994, Richmond, B.C. Canada. . 1995. An induced defense reaction in white spruce to attack by the white pine weevil, Pissodes strobi. Can. J. For. Res. 25: 1725-1730. Bernstein, I.H. 1988. Applied multivariate analysis. Springer-Verlag. Pp 315-344. Cozens, R.D. 1983. The spruce weevil Pissodes strobi Peck (Coleoptera: Curculionidae): a review of its biology, damage and control techniques with reference to the Prince George Timber Supply Area. B.C.Min. For., Victoria, B.C. Internal Rep. PM-PG-3. . 1987. Second broods of Pissodes strobi (Coleoptera: Curculionidae) in previously attacked leaders of interior spruce. J. Entomol. BC. 84: 46-49. El-Kassaby, Y.A., A. Sigurgeirsson, and A.E. Szmidt. 1988. The use of restriction analysis of chloroplast DNA in classifying hybrid spruce seedlots. In: Proc. Frans. Kempa Symp., Molecular Genetics of Forest Trees. June, 1988. Umea, Sweden (J.E. Hallgeren, ed.). Institute of Forest Genetics and Plant Physiology, Swedish University of Agricultural Sciences. Umea, Sweden. Report No. 8:67-88. Gary, R.I., R.L. Carlson and B.F. Hrutfiord. 1971. Influence of some physical and host factors on the behaviour of the Sitka spruce weevil, Pissodes sitchensis, in southern Washington. Ann. Ent. Soc. Am. 64:467-471. Kuehl, R.O. 1994. Statistical principals of research design and analysis. (Chapt. 15: Repeated Measures Designs). Duxbury Press, Behust, Canada. Pp 499-528. Hall, P.M. 1994. Ministry of Forests' perspectives on spruce reforestation in British Columbia. In: Alfaro, R.L, Kiss G. and Fraser R.G. (eds) The white pine weevil: biology, damage and management. Symp. Proc. ppl~6. Jan. 19-21, 1994, Richmond, B.C. Canada. Harman, D.M. 1975. Movement of individually marked white pine weevil, Pissodes strobi. Environ. Entomol. 4: 120-124. Hormann, R.K. 1987. North American tree species in Europe. J. Forestry. 85: 27-32. Hosmer, D.W. and Lemeshow, S. 1989. Applied Logistical Regression. New York: John Wiley & Sons, Inc. Ilingworth, K. 1978. Sitka spruce provenance trials three years after planting in British Columbia. In: Proc. IUFRO Joint meeting of Working Parties, Douglas-fir, lodgepole pine, Sitka spruce and true firs. Vol.2 Pp 311-326. Vancouver, B.C. Canada. Kiss, G.K. and A.D. Yanchuk. 1991. Preliminary evaluation of genetic variation of weevil-resistance in interior spruce in British Columbia. Can. J. For. Res. 21: 230-234. Mitchell, R.G., K.H. Wright, and N.E. Johnson. 1990. Damage by Sitka spruce weevil (Pissodes strobi) and growth patterns for 10 spruce and hybrids over 26 years in the Pacific Northwest. U.S. Dep. Agric. For. Serv. Res. Pap. PNW-RP-434. Pojar, J., K. Klinka, and D.V. Meidinger. 1987. Biogeoclimatic ecosystem classification in Brititsh Columbia. Forest Ecology and Management. 22: 119-154. SAS Institute Inc. 1990. SAS/STAT User's Guide, Version 6, 4th edition. Vol .1 (pp 943) & Vol. 2 (pp 846). Cary, NC, USA. Silver, G.T. 1968. Studies on the Sitka spruce weevil, Pissodes sitchensis, in British Columbia. Can. Ent. 100: 93-110. 76 Wood, P . M . 1987. Development of Sitka spruce phenotypes resistant to the spruce weevil: a summary of recent and planned research projects in B.C.. BC MoF, Vancouver Forest Region. Internal Rep. PM-V-10. Ying, C.C. 1991. Genetic resistance to the white pine weevil in Sitka spruce. Research Notes No. 106, BC MoF, Victoria, British Columbia. . 1997. Effects of site, provenance, and provenance and site interaction in Sitka spruce in coastal British Columbia. Forest Genetics. 4(2): 99-112. 77 3. Long-term growth responses in Sitka spruce populations to seed transfer from IUFRO provenance trials in British Columbia Abstract: The 20-year growth of Sitka spruce provenances tested in Brititsh Columbia were analyzed towards the goal of defining latitudinal seed transfer limits for higher-than-local growth performance of the provenances planted at coastal BC areas. Growth responses were quantified by modeling volume deviation of ecdemic (i.e., non-local) provenances from local sources at year 20 to geoclimatic changes resulted from seed transfer of 41 Sitka spruce IUFRO provenances at 11 test sites. The predictions were again related to test site's geoclimatic conditions by response surface analyses to predict the volume response from seed transfer pertaining to site geoclimatic conditions. Volume contours were also developed responding to each effective predictor for planting site as well as for provenance origin to attempt to define suitable ranges of seed source and planting area. Results indicated that northward seed transfer along the Pacific coast is favored in this species, an average ultimate volume-gain was predicted from using seed source 5.5 0 of latitude south of a planting site. The volume response is contingent with site geoclimatic conditions such that, the milder and/or the more southern the planting site is, the wider the limits of northward seed transfer that allows for pursuing higher-than-local growth, and also the greater the amount of volume-gain can be achieved through the transfer. High dependency of growth response on site summer precipitation was also found such that, a minimum of 500 mm summer rainfall was required to achieve higher-than-local growth performance. With high site precipitation (e.g., SMSP > 700 mm), about 40% of volume-gain 78 could be achieved by northward seed transfer up to 12 0 of latitude. The selection of planting site is more important than selection of provenance origin. A minimum of 670 mm summer precipitation at planting site is required for high volume production with this species. Keywords: growth response; seed transfer; provenance trial; Sitka spruce (Picea sitchensis (Bong.) Carr.). 3.1. Introduction Sitka spruce, Picea sitchensis (Bong.) Carr., a fast growing conifer native to North America, occurs naturally in the Pacific coast 'fogbelt', a long, narrow strip adjacent to the Pacific Ocean spanning over 22 degrees of latitude (Daubenmire 1968; Pojar, et al 1987). Its growth vigor, high wood quality and versatility to soil conditions made it a recommended species for reforestation in coastal areas where the white pine weevil (Pissodes strobi (Peck)) threat is low (Ying 1997). In order to exploit the potential of genetic superiority of ecdemic (i.e., non-local) provenance over local provenance in growth, the Research Branch of Ministry of Forestry of British Columbia (BC MoF) established three series of Sitka spruce provenance trials in early 1970s (Illingworth 1978a; Ying 1991 and 1997; Chapt. 1). These trials under this study involve a total of 43 provenances that are tested at 11 sites in coastal BC (see Fig. 1-1 in Chapt. 1). The provenances were collected along the Pacific coast from Oregon coast to southern Alaska and formed the core sample of the Sitka spruce RJFRO (International Union of Forest Research 79 Organizations) provenance trials. Height and diameter of the trees were measured on individual trees over 20 years since planting. The primary objective for provenance trial is to identify suitable seed source(s) by comparing the performance of local to ecdemic (non-local) populations. Many provenance trials have shown that local seed sources are often not the optimum in growth, and that provenances from mild and/or southern areas usually outgrow those from harsh and/or northern areas, but are more vulnerable to winter injuries (e.g., Mergen et al. 1974; Campbell and Sorenson 1978; Illingworth 1978a and b; Rehfeldt 1983 and 1995). Previous analyses with the 20-year growth data of these Sitka spruce provenance trials found considerable genetic variability among the 43 provenances, despite the dominance of site effects over contrasting environments (Ying 1997; Chapt. 1). Linear and quadratic trends that are inverse to latitude were found in growth measures where the test site is favorable for Sitka spruce and free of weevil attack. With increasing age, the latitudinal trends tended to be stable in both height and diameter growth, with different levels of expression, by the 20th year after planting (see Section 1.4.4. in Chapt. 1 for details). The geographic trends stabilizing by year 20 justifies the. notion that this is an appropriate age to address the species' seed transfer limits, a question that silviculturists are most interested in, though it is still too early to make a final assessment if considering the species' rotation length (Ying 1997). This chapter is a summary of the predictions with the 20-year growth data towards the primary goal of provenance trials. The objectives are 1) to quantify the volume-gain and -loss, i.e., growth response, resulting from seed transfer in these trials; 2) to define the limits of seed transfer that allow for higher-than-local growth performance under a giveri plantation's 80 geoclimatic conditions, assuming volume production is the primary goal of reforestation with this species; 3) to provide graphical views of suitable geoclimatic ranges for planting area and seed source selection within the experimental span. 3.2. Data profile and Abbreviation Data from three series of Sitka spruce provenance trials in British Columbia (supplied by Research Branch of BC MoF) were applied in this study. These trials together have 43 provenances and 11 test sites (see Tables 1-1 and -2 in Chapt.l for the names and geographic locations of the provenances and test sites). The provenance trials were established using a completely randomized block design, with 4, 5, 6 or 9 blocks at different test sites. Within blocks, each provenance is represented by a 9-tree-row plot. Not all provenances were tested at all sites so that a total of 220 provenance-by-site means were available for each growth measurement. Growth measurements available are height at year 3, 6, 10, 15 and 20 after planting (termed as HT3, HT6 and so on), and diameter at breast height at year 10, 15 and 20 (termed as DBH10, DBH15 and so on). Individual tree volume was calculated after Kovats (1977) when both height and diameter are available for a tree. For long-term simulation and silvicultural concerns, volume at year 20 (termed as VOL20) was selected as the growth trait of interest. Macro-climatic data were obtained from the nearest weather stations to each test site (i.e., site climatic variables) as well as to each provenance origin place (i.e., provenance climatic 81 variables). Adding "S-" as a prefix for site climatic variables while "P-" for provenance climatic variables, the acronyms of the 10 macro-climatic variables that define temperature, moisture and photoperiod conditions are as follows: Following the same convention, the acronyms for latitude, longitude and elevation are LAT, LONG and ELEV, respectively, with the prefix "S-" for test site and "P-" for provenance origin. Geoclimatic differences between provenance origin and test site were obtained by subtracting the values for provenance origins from those for test sites where the provenances were tested. These geoclimatic distance variables are named as "Diff-" variables, i.e., DiffLAT, DiffMAT, etc. For instance, DiffLAT = PLAT - SLAT. These Diff- variables are mostly correlated with each other (see Table JJI-13 in Appendix III). The variation ranges of these Diff-variables are presented in Table III-14 in Appendix HI, which should be used as the range limits for interpreting the prediction results, namely, within the experimental span. Growth responses are expressed in the ratio of an ecdemic provenance's growth performance over the local growth performance where the ecdemic provenance is tested (after Schmidtling 1993). These growth responses are termed as "Devi-" variables, i.e., DeviHT3, DeviVOL20, etc. For instance, DeviVOL20 = (Ecdemic provenance's VOL20) -H (Local MAP MSP MAT MTCM Mean Annual Precipitation (mm) Mean Summer Precipitation (mm) (May ~ September) Mean Annual Temperature (°G) Mean Temperature of the Coldest Month (i.e., January) (°C) Mean Temperature of the Warmest Month (i.e., July) (°C) annual Number of Frost Free Days (day) annual Frost Free Period (day) annual accumulated Degree Days above 5°C (°C) annual accumulated Degree Days below 0°C (°C) accumulated day length (hour) of the growth season (April ~ October) (calculated as a function of latitude) MTWM NFFD FFP DD5 DDO DAY 82 provenance's VOL20). Again, for the long-term simulation and silvicultural concerns, DeviVOL20 was selected as the growth response of interest among all the Devi- variables.. The ecdemic provenance's growths were evaluated as the Least-Squares means for the growth measurements at provenance-by-site level, computed from the SAS GLM procedure (adjusting for missing observations that are plot-mean based). Local performances were obtained either from the local or "close-to-local" provenances. Close-to-local provenance performances were derived from regression models that are site specific in relation to provenance geographic locations. 3.3. Methods of analyses and limitations of the predictions 3.3.1. Identifying effective geoclimatic predictors Predicting growth response (i.e., DeviVOL20) using all the geoclimatic distance variables is difficult to interpret and unnecessarily complicated. In order to screen for effective geoclimatic predictors among the Diff- variables for predicting the growth response, redundancy analysis was applied on the multivariate relationships of the two groups of variables, i.e., Diff-variables and Devi- variables. The analysis was performed through matrix algebra with the SAS IML procedure, based on the correlation matrices within and between these two groups to avoid scale problem of the original Diff- and Devi- variables. Redundancy analysis is effective for determining the variation in one group of self-correlated variables accounted for by the variability of another group of self-correlated variables. It derives two sets of redundant variables that are linear combination of the original variables of 83 the two groups, respectively, such that the variations in one group were maximally accounted for by the opposite group's redundant variables (Wollenberg 1977). From the loadings of the original variables onto the redundant variables, the relative importance of the original variables of one group in explaining the variations of the other group is determined. In this study, growth response variables (i.e., Devi- variables) are closely correlated, and so are the Diff- variables (see Table 111-13). Therefore, the redundant loadings of the Diff- variables to their own redundant variables can reveal the relative importance of the Diff- variables in explaining variations of the Devi- variables. 3.3.2. Quantifying general volume response from geoclimatic changes The growth-gain or -loss resulted from each predictive Diff- variables, as determined by the redundancy analysis before, were quantified by modeling the volume response (i.e., DeviVOL20) with each effective Diff- predictor, using the SAS GLM procedure. Predictions based on the five geoclimatic variables simultaneously would be largely redundant because of the close inter-correlations among the independent variables (see Table HI-13 in Appendix DT). Predictions with each variables separately, on the other hand, could also be deemed as over simplification, but in fact it is the way to quantify the volume response without turning into linear transformations of the original predictive variables (e.g., through Principal Component Analysis (PCA)) and thus making the results difficult to interpret and apply. If predicting the volume response using the principal components of these five geoclimatic distance variables, which Matyas (1994) called 'ecodistance' between seed source and planting site, it is not only unable to interpret the prediction in the original geoclimatic sense, but also difficult to 84 apply the predictive results to seed transfer practices. Every time when one needs to foresee a growth performance for a new provenance or a new planting site, he would have to go back to the PCA to derive a new ecodistance of the seed transfer in order to determine the growth response from that transfer. In other words, predictions with ecodistance are data specific. Provisionally, all the geoclimatic data needed are available for provenance origin as well as for planting site which is uneasy, if not impossible, to obtain at a real situation. Hence, the predictions and subsequent response surface analyses in this chapter would be using one geoclimatic predictor at one time to present simple predictions and to avoid the above mentioned problems. The models were set up to the second power of the independent variables to comply with the quadratic trends found in scatter plots, and under the constraint that the intercept of the model equal zero. By making null intercept, the predicted value for DeviVOL20 at the point of testing a provenance without geoclimatic distance from the origin is zero, which means the ratio of the tested provenance's growth over the local source's is 1:1 (i.e., the performance of the tested provenance is identical to local one's). The geoclimatic distance ranges that allow for pursuing higher-than-local growth performance in seed transfer were then delineated from the mean predicted values for DeviVOL20. 3.3.3. Predictions pertaining to site geoclimatic conditions As volume response is highly conditional on site geoclimatic conditions (see below), the general prediction could only present the average level of the response. In order to provide more operable transfer limits, the volume response to each effective Diff- predictor was related to the 85 corresponding geoclimatic variable for test site by response surface analysis, using the SAS RSREG procedure. Site summer precipitation (SMSP) was also used in the response surface analysis with DiffLAT to DEviVOL20, because of high sensitivity of this species to SMSP (see Section 1.4.3. in Chapt. 1). Illustrated by a series of contour graphs, the results presented a more sophisticated view of the previous general predictions pertaining to site geoclimatic conditions. However, there are also a few limitations for the application of these contour graphs, from the bias sources as stated in the following section (3.3.5.). 3.3.4. Graphical approach for seed transfer guidelines In conjunction with the above predictions and response surface analyses, the ranges of suitable planting area and seed source under given geoclimatic conditions within the experimental span were also discussed by another series of contour graphs. The graphs were constructed after the method proposed by Kung & Clausen (1983), using pairs of the predictive geoclimatic variables for test site as well as for provenance origin as the two-dimensional independent variables and the growth vigor (represented by VOL20, based on provenance-by-site means) as the response variable. The six predictive geoclimatic factors (i.e., LAT, MAT, MTCM, DDO, NFFD and MSP) were determined by previous analyses as stated before (Section 3.3.3.). This kind of contour graph can present a direct view of volume productivity of different provenances (represented by the geoclimatic conditions of their origins) at different areas. The graphic approach can reveal the relative importance of site and provenance selection in seed transfer practice. The graphs are also capable of showing the presence or absence of provenance-86 by-site interactions for a particular geoclimate factor. However, the use of VOL20 as the response variable brings back the experimental effects (see Section 1.4.1. in Chapt.l) into the modeling. Consequently, there could be significant 'lack-of-fit' error due to these experimental effects, and to lack of fine geoclimatic gradients in site location selection and lack of randomness in provenance sampling, coupled with the drastic weevil damages observed at four test sites (Chapt.2). However, by substituting the relative growth vigor (i.e., DeviVOL20) with the observed VOL20, the models can also avoid the impact of possible inaccuracy from some of the local source performance that are derived from regressions in this study (see Section 3.3.5.). Therefore, the graphic approach of this section and the last section both have merits and demerits compared to each other, and the results of both approaches should be considered descriptive rather than inferential. 3.3.5. Limitations of the predictions and modeling Several bias sources of the predictions should be mentioned here, some of which are applicable to all the predictions, some are to a specific analytical approach, as per the following: First, the latitudinal seed transfer in this species is virtually northwest-southeast oriented, because the Pacific coast of British Columbia is oblique, not paralleled with longitudes. The prediction using latitudinal distance as predictor can not distinguish a northwest-southeast trend from a north-south trend. Therefore, the predictions with DiffLAT could be.biased somehow due to this reason. IN THIS STUDY WHEN "NORTHWARD SEED TRANSFER" IS REFERRED, IT ACTUALLY MEANS "SOUTHEAST TOWARD NORTHWEST SEED TRANSFER". Again, response surface analysis only works well when the data structure is symmetric. The test 87 sites are not symmetrically distributed, so are sampling of the provenances as well as is the Pacific coast which is northwest-southeast oriented. Therefore, the response surfaces that are associated with latitudinal factors can not avoid biases from this structural problem, and hence are less reliable than those associated with climatic factors. Second, climatic gradients along the coast line of BC are generally gradual, but very steep from coast (maritime) to inland (submaritime). Therefore, the response curves for coastal region can be very different from those of inland. However, there are only three submaritime test sites (see below in Section 3.4.2), and the sampling of provenances at these test sites were not ample enough to allow for separate prediction from those maritime test sites. The pooling of maritime and submaritime sites caused the general predictions being general to the whole region of coastal BC, but not specifically good for either outer coastal area or peripheral inland area. Third, the three test sites, i.e., Head Bay, Nass River and Rennel Sound, had no local source tested. The local growth performance at these sites were derived from site-specific regression models (details not presented). These regression models are set up by relating the performances of those provenances at the test site to their origin locations, at provenance-by-site mean level. These "close-to-local" performances could bring in certain degree of bias to the predictions relative to local performance. Fourth, the four test sites that weevil attacked (i.e., HB, KT, MN and NS) are included in the predictions under the assumptions that, there were no overall substantial differences in weevil resistance among the provenances tested at the attacked sites, and that provenances (local and ecdemic) suffered similar weevil damages at a specific site. These 'equal-weevil-effects' assumptions were made for not sacrificing the four weeviled sites which have contrasting 88 environments. The including of these test sites can help better expression of the G x E interactions in growth performance (see Chapt.l), which is important to assess the seed transfer limits, although it could bring in certain degree of inaccuracy to the predictions in the fact that, previous analyses show that a few (i.e., three or four) provenances exhibited considerable weevil resistance and/or tolerance at these attacked sites (Chapt.2). The last but not least limitation of the predictions is the tree age. Although the 20-year data has show the stabilizing of the geographic trends in growth traits by year 20 (Chapt.l), these trees have not been exposed to extreme climate events that might happen once every five to ten decades. Comparing to the species' rotation length (100 years long, see Ying 1997), the 20-year period is still short for prudent assessment. 3.3.6. Statistical criteria All data analyses were performed by SAS procedures (SAS Inc. 1990). The predictions were made on the response variables that are transformed into natural logarithmic values to approach normal distribution, but results are interpreted in the original units of the response variables. Statistical tests and inferential analyses were based on the assumptions that, multivariate normality and simple normality exist in the growth response variables, variations of the response variables are homogeneous across different levels of experimental effects, and the residuals from the predictive models are normally and independently distributed, with a zero mean and a common variance. The significance criterion was set at a = 0.05 level if not stated otherwise. 89 3.4. Results and Discussions 3.4.1. Effective geoclimatic predictors Redundancy analysis was used to determine the effective geoclimatic predictors by the 'loadings' of the original variables onto the first pair of redundant variables which maximally cross-explains the variations of the opposite group. The two groups of variables are the five Devi- variables (i.e., DeviHT3, DeviHTIO, DeviHT20, DeviDJ3H20 and DeviVOL20) and the 12 Diff- variables (i.e., all the geoclimatic distance variables but DiffDAY). The results show that 46.4% of total variations in 'Devi-' variables were explained by the first redundant variable of the 'Diff-' group (Table 3-1). That is, to the maximum, 46.4% of the total variations of the growth responses were accounted for by the geoclimatic distances between seed source and test site. This implies the growth responses are predictable by the geoclimatic distances in seed transfers of this experiment. Since the first redundant variable of the 'Diff-' group accounted for the maximum variations of the growth response variables (i.e., opposite group), attentions were directed to the correlations (loadings) between the original 12 geoclimatic distance variables onto its first redundant variable, and between the original five growth response variables onto this redundant variable (Table 3-2). These redundancy loadings indicate that only the 'Diff-' variables describing thermal difference (especially winter temperature) and latitudinal distance between seed source and test site were essential to reflect the growth response variations. The top five Diff- variables with highest correlation coefficients to the first redundant variables of their own group were thus selected as the effective geoclimatic predictors to be used in the subsequent modeling. They are DiffDDO, DifiMTCM, DifiMAT, DiffNFFD (i.e., thermal differences) and 90 DiffLAT (i.e., latitudinal distance). It should be pointed out that, DiffDAY is also an effective predictor for DeviVOL20, just as DiffLAT is, but was.excluded from the analysis because it is a function of latitude. Redundancy loadings in the other group (i.e., 'Devi-' variables) indicate that upon year 20, DeviDBH20 was most predictable among the five growth response variables (Table 3-2). However, for silvicultural concerns, the remainder of this chapter is focused on the predictions of the volume response in year 20 (i.e., DeviVOL20), which is also highly redundant on the variation of the Diff-group. Table 3-1. The amount of original variations of the 'Devi-' variables and 'Diff-' variables explained by the redundant variables of their opposite group. Original Redundant variable of the 'Devi-'group Redundant variable of the 'Diff-' group Variations 1st 2nd 3rd 4th 1st 2nd 3rd 4th 'Devi-' group 'Diff-'group 23.4% ' 4 . 9 0 % 1.90% 0.38% 46.37% 2.02% 1.55% 0.47% Table 3-2. Correlation coefficients (i.e., redundancy loadings) between the original variables with the first redundant variable of the 'Diff-' group. Correlation coefficients for the original Correlation coefficients for the original DiffLAT -0.803 DeviHT3 0.821 DiffLONG -0.147 DeviHTIO 0.637 DiffELEV -0.348 DeviHT20 0.605 DiffMAP 0.413 DeviDBH20 0.747 DiffMSP 0.057 DeviVOL20 0.560 DiffMAT 0.775 DiffMTCM 0.742 DiffMTWM 0.103 DiffNFFD 0.646 DiffFFP 0.416 DiffDD5 0.614 DiffDDO -0.816 (Note: Characters and values in bold represent the selected effective geoclimatic predictors) 91 3.4.2. Trends from the scatter plots The scatter plots of DeviVOL20 versus the five effective geoclimatic predictors (Figs. 3-1 to -5) indicated that the growth response depending on these predictors were quite similar, with quadratic trends that are concave down, except that the directions for DiffLAT and DiffDDO were opposite to the remaining predictors. This is because of the negative correlations of these two variables (DiffLAT and DiffDDO) with other three variables (Table HI-13 in Appendix Hi). Substantial volume growth variation at year 20 was found among the 11 test sites in that, the site means for VOL20 varied with good sites producing volumes that were ten more times higher than that for poor sites (see Chapt.l). The volume growth response related to local source performance (i.e., DeviVOL20), on the other hand, turned out to be consistent from site to site and based on one predictor or another. This is because Devi- variables integrated the variation from both test site and provenance origin, and thus removed the effects of Site and Provenance, as well as the G x E interactions. The effects from block within test site were also removed when determining local and ecdemic growth performances, based on provenance-by-site means. The consistency of the trends allows for making general predictions that can be applied to most of the coastal regions of British Columbia. 92 o "1 i — — i r cu O cu cu L_ O) cu T 3 O r— O Q O 1 1 1 7f • • 1 • • • _ • • • •• • *• i/ * "** * .• ••• • • • • • \ -i V •1 \ • 1 1 1 LO d o d LO d >. CO T> CD CU O) cu 2. o Q Q © 33 CN CO J .a > -a > 53 o 3 Q ^ • O, C3 t s Q 3 S Q £ <*> *5 « . Q ^ (021OA!/\9a)ui (0S1OA!Aaa)un ' ( a D h CD H-» o o g cn . cn Ov 1 1 1 • 4/ 1 _ • • • 7 1 / • • • • -• • / • 7y• • • -At* • • r» * • 1 • • • # % — • • . * -• • • • 1 1 1 cu T 3 as o cu 2> cu •o I o O LO o LO O LO o d d d c\i cn 3 cn CD > O CN O > > CD < H H o -4-» jo "HH ( H CD s O on T3 CD •a o o a o cr CD •a T 3 (02"IOA!A3a)ui CM <—i l m s •a 3 CD is 1 i y i i / * ~ • • • • •• IN # " V # J • • • i • * • • k»* > V v* * • • • • • — • • «tvV • • • >f • — • • • «\ 1 l *N i LO d o d LO O d T -• $ Q cu O 5. i5 o (021OA !AaQ)ui . 3.4.3. General predictions on each effective geoclimatic predictor Slight difference in volume response were observed between maritime (i.e., wet) sites (mean annual precipitation > 20.00mm, eight sites together ) and submaritime (i.e., less-wet) sites (mean annual precipitation < 2000mm, three sites together). The divergence was pronounced particularly when involving maximal transfer of southern provenances to northern areas (see Figs.IV-1 to -5 in Appendix IV). In less-wet sites, the range of northward transfer was wide enough to cause loss of growth superiority of southern provenances over local sources. However, this trend was not observed in wet sites. Without the less-wet sites, the prediction models that are based on DiffLAT, DiffMAT and DifiMTCM tended to be linear rather than quadratic. Sufficient precipitation somehow compensated unfavorable thermal-climatic conditions during winter for the southern provenances at northern outer coastal areas. However, if comparing the common range of geoclimatic distance of the seed transfers occurred between wet and less-wet sites, there were no substantial differences between these two site types. Growth response would be expected to be sub-optimal when the northward seed transfer exceeds certain range of geoclimatic distances no matter how wet a test site is. Besides, the experimental range and data availability did not allow for making separate predictions for these two site types. Therefore, I decided to pool the test sites together to make general predictions, while leaving the problem of the moisture differences among sites to the next section (3.4.4.) of this chapter, in which the volume response was related to both geoclimatic distances and site geoclimatic conditions. The general predictions in this section reflect the average volume-gain and -loss in seed transfer, and the results are applicable to the whole coastal BC area 94 (including maritime and submaritime regions) where the environment allows for successful planting with this species. Predictions were made on the five effective geoclimatic predictors (i.e., DiffLAT, DifiMAT, DiffJVITCM, DiffDDO and DiffNFFD), respectively. Values for the response variable (i.e., DeViVOL20) are at provenance-by-site mean level. Three outlier provenances were detected from the scatter plots. They are: Brookings (No. 3018, the southernmost provenance from Oregon) and Yakutat (No. 3021, the northernmost one from Alaska) tested at Nass River site, and Necanicum (No. 3012, from Washington) tested at Maroon Creek site (the harshest one of the 11 test sites). The outliers were eliminated from the model to exclude the extreme cases of seed transfer in the provenance trials. The results indicated that all the prediction models fit well to quadratic curves that are concave down (Table 3-3). This could be proved either by the significance levels of the parameter estimates and the model R 's, or from the residual plots (not presented here). In all the prediction models, both linear and quadratic parameter estimates were significantly different from zero (a = 0.05), which means that there were significant effects of these geoclimatic distance variables (linear and quadratic) on the variation of DeviVOL20. The gross model R 2 ranged from 0.308 to 0.476 (Table 3-3), indicating that about 31 to 48% raw variation in DeviVO120 were accounted for by these models (not purely by the predictors as the R 2 was not adjusted by the intercept that is set to zero). The residual means from these models ranged from -0.0037 to 0.0027, and none of which was significantly different from zero (Mest, Table 3-3), suggesting that the models represent unbiased predictions. 95 The predicted values for DeviVOL20 (in logarithmic values) were transformed into percent deviations of the growth response from local source performance, namely, DeviVOL20(%). The values for DeviVOL20(%) along with the transformed standard errors of the predicted means are listed in Table 3-4, in and comply with the ascending order of DiffLAT. The same prediction results were also visualized by plotting the predicted curves along with the standard errors of the mean predicted values plotted as vertical bars across the curves (Figs. 3-6 to-10). 96 Table 3-3. Estimated parameters for the prediction models along with the quality information of the models. (1) DeviVOL20 = / (DiffLAT, DiffLAT2) Factor Parameter Standard Error T for H 0: Pr > |T| Estimate of the Estimate Parameter = 0 DiffLAT -0.0800827549 0.00825244 -9.70 <0.0001 DiffLAT2 -0.0073923605 0.00147519 -5.01 O.0001 n = 217, Model R2 = 0.3075, Root MSE = 0.3275, Mean Residual = -0.0036690 (t-test for Ho: Mean Residual = 0 is 111 = 0.1646 < 1 0 05 = 1-97) (2) DeviVOL20 = / (DiffMAT, DiffMAT2) Factor Parameter Standard Error T for Ho: Pr > |T| Estimate of the Estimate Parameter = 0 DiffMAT 0.0931227597 0.00909669 10.24 <0.0001 DiffMAT2 -0.0149617111 0.00198997 -7.52 . <0.0001 n = 277, Model R2 = 0.3739, Root MSE = 0.3128, Mean Residual = -0.0026960 (t-test for HQ: Mean Residual = 0 is 111 = 0.1060 < 1 0 os = 1.97) (3) DeviVOL20 = / (DiffMTCM, DiffMTCM2) Parameter Standard Error T for H 0 : Pr > |T| Estimate of the Estimate Parameter = 0 0.0327584368 0.00377511 8.68 O.0001 -0.0025321835 0.00033773 -7.50 <0.0001 n=217, Model R2 = 0.3757, Root MSE = 0.3175, Mean Residual = 0.000485961 (t-test for HQ: Mean Residual = 0 is 111 = 0.0019 < 10.05 = L97) Factor DiffMTCM DiffMTCM2 (4) DeviVOL20 = /(DiffDDO, DiffDDO2) Factor Parameter Stamdard Error T for H 0 : Pr > |T| Estimate of the Estimate Parameter = 0 DiffDDO -0.0005775073 0.00005998 -9.63 <0.0001 DiffDDO2 -0.0000006145 0.00000008 -7.81 O.0001 n = 176, Model R2 = 0.4762, Root MSE = 0.2833, Mean Residual = 0.0027209 (t-test for HQ: Mean Residual = 0 is 111 = 0.0758 < 1 0 0 5 = 1.98) Factor DiffNFFD DiffNFFD2 Pr > |T| (5) DeviVOL20 = / (DiffNFFD, DiffNFFD2) Parameter Standard Error T for H 0 : Estimate of the Estimate Parameter = 0 0.0035561698 0.00039717 8.95 0.0001 -0.0000279873 0.00000393 -7.12 0.0001 n = 176, Model R2 = 0.3660, Root MSE = 0.3055, Mean Residual = -0.0010111 (t-test for HQ: Mean Residual = 0 is 111 = 0.0366 < 10.05 = 1.98) (Note: As the intercept of the model was set to zero, the model R2 was not adjusted for the mean.) 97 a o p «> • & H.a T 3 O N -a CD 6fl i i © 9 ^ > 8 Q e t, Q o o 43 Q S Q * Q CD <g > H "3 Q > CD 7 3 3 Q 1-1 M CD c a P H § CD CD CD J 3 > CD T3 CD IH CD -3 CD IH OH CD a u Q H C H o CD I CO CD ccS <D Q 'p U-i S3 o cm O CS > cu © c s -J Ii > © S 5^ 5 o CN - I tl > 1€ «w v^ •vt r n d i n O S O N O l O N oo vd in' i n •'t +1 +1 +1 +1 +1 O N h - O i n « * ci oi' o i n <N O N •vt oo CN •vt •vt •vt r n cn o i _; +1 +1 +1 +1 +1 +1 +1 +1 O N m i n •vt O N 00 © O l cn IT; cn o d +1 o o o r~- o o j vo oo T— i 1 CN o i o i o i o i +1 +1 +1 +1 +1 +1 +1 oo oo cn o l i n o vo CN VO 2 ^ i—> o - co CN o i cn o •vi-vo v a S § S r 5 S § 5 2 ? ? ? ? * « * v 5 v 5 v T i C 5 C i C i C i c j C 5 C 5 vq —t vp —; m ON 5) 5) oo od t > vo +| +| + | + | « t N OO - H r H CO O • t O O C j h H ^ r H i O M M v d i n i n - - t " ^ f O c n o i o i - - ^ +1 +1 +1 +1 +1 +1 +1 +1 +1 +1 cn vo oo —^  © o d o © o +1 +1 +1 +1 HH + •vt o vo cn o o o i m T j - T f T r ^ j - c n c N ' - H .. +1 o i vo O N c i n o l O o cn VO O N +1 +1 cn vd vS v » v » s« 5 « 5 « 5 v 5 C S c j c s c j c s c i C i C i C 5 C 5 C i a ^ » ( > c i c i C s i i n o l oo o j cn c n o i o i vo ON o i ^ ~ £ d d •vt v q oo ON - H S a i od vd r~ CN oo cn CN O l ^ O N m —I "vt •v i - i n ON * - 1 cn i n 00 ^ H •vt O N vd i n rn cn o i -H +1 +1 +1 +1 +1 +1 CN •<t © i-~ 00 •vt T-H o i r n rn o i , — ; d 1—1 i—1 *—* r — 1 t - H cn oo ^H' d +1 +1 •vt o -vt O ^ o d ° ° - H T t O l ob vd cn v; VI +l +l +1 +1 +.1 ±1 ^ $ vq O N o d cn t v . ^ ^ I vTfi, ON, Oo <N <N "-i *-< K V 5 i*i > »»( •n ' <N N v 5 o\ oo •-^ *»( »«H » « i K V - 3 l n N > r - j C N | ^ C i r s , ^ i oo i n o l o d d +1 +i +i *N >-< : >-H i n O N cn o l —< O N +1 cn cn "1. ^ O N 00 +1 +1 oo vq K oi t - . O N I-I vd vd +1 +1 +1 0 i-t vo 01 vd O N cn vo O N Oj vq o i n cn rn o i o i +1 +1 +1 +1 +1 +1 cn O l cn vq O N i n o i i n ir! TT' cn O N m d d oo "vt d +1 O N •vt O N c n r~; --H --t oo 0 ^ H i-^< o i o i o i +1 +1 +1 +1 +1 +1 01 O N oo oo O N O N o i n i—i r~ cn O N r _ H i—* CN o l cn cn vS OS v5 >ri * >T) vTi i r i C J K K vd sci vn «^ •n vS >ri v5 i r i c j cv" cv" rvj rvj «0 "n vs <5S vS 'n « IC) C J IC) C J 1^  vS I CN\ fN| CT) CT) • • I I I f"! r~; o c n vo v i v i i n i > oo cn o i +! +i +i +i c n oo i n m O N i n t—i f l « i r , N H N V O v t 00 CN VO ON ^ i ; ^ O N O d d O 00 cn i n "n v i vtf vn O S v s <n ON I 00 I < — 1 O l m i n 00 o •vt 00 O N 00 i > vd i n r n +1 +1 +1 +1 +1 +1 +1 +1 O l cn O N l - H O N OJ i n oo d ; cn cn T f ^t r n O l O l O l CN «N O l O l vS "n v?S «*n Oi "n vS "n oc K K vd <n <n •^ t O VO O l •vt O oo d d +1 +1 i n O N r~ r n o d +1 o d v 5 H , a i c i C i l r i v 5 l r i C l T3 - 3 c o o T f <n « Q 5 o CN m °) cn in vq 00 ON ON VOL ,4±1, t—< +1 CN +1 ,9±2. ,8±2. .6±2. CN +1 cn ,0+2. ,6+2. CN +1 .5+2. Devi w o T f CN i 00 CN ro m •en i o\ <n i cn T f T f o m i T f in r-in Q e CU NT pi. Tf 00 —; cn cn T}-' +1 +1 +1 CN in in cn ON in m in vo «s «a> va-cs *•< * » i "-i »»< i i i © N a e s N a t a e i e s t a c i c a c i N 5 i n « S ^ N 5 ^ N 5 i r i C i i f i C S ' T l ' n > > l ' > > n N C N o tx K oo h O N i / i h O i ' H r n T t m in <—< CN CN CN CN CN cn cn cn tn cn +1 +1 +1 +1 +1 +1 +1 T f O T f O A D 1 +1 +1 i in i> r~-o i n O N T f O N c n o o c N v o o T f CN CN CN c n c n T J - T J * in in A O N O i n NO i i i CK «a >s CN <n > "n O CN e n T f TJ- T f c n c n c n c n c n c n +1 +1 +1 +1 +1 +1 00 T f 00 ON NO O in ^ NO i-< NO r - H T f in m NO vo r--Q ^ c ic) a ic) > > «4-> irj VO A O I I I I I I T f 00 o © +1 +1 T f * OO fr)t~-;^ l /">ov.<nvooN^H ~* -—icNcNcNcncncnTf +1 +1 +1 +1 +1 +1 +1 •• • ' T f © m o oo cn oo CN t-^ i—i vo © . m CN CN cn cn o T f +1 +1 cn in T f od T f T f ^ C ^ N i ^ C i K N C J I r N C J ^ «5 ^ ^ CN CN) cn >f >" «n «^ (%) OSIO/W^Q (%) 031OA!A9a o o (%) 031OAIA3Q c o (%) 031OA!Aaa c o T3 CO s CO cd '</> a 0) o •3 cu CD o a> (-1 CD o c h- w < T3 c3 T3 b •4—1 H CZJ < cS • C H ( O oh 1—t c o Positive values for DeviVOL20(%) in Table 3-4 implies higher-than-local volume growth. Results show that the predicted higher-than-local performances were only associated with seed transfer from southern or mild sites to northern or harsh sites within certain ranges. This allows for drawing such a conclusion that in Sitka spruce, northward seed transfer is favored while southward transfer is unfavorable in volume growth as compared with using local seed source. From Table 3-4, the predicted ranges of northward seed transfer for a provenance remaining superior or at least equal to local source in term of VOL20 are as follows: DiffLAT = 0~ -10 ° N DiffMAT = 0 ~ 6 ° C DiffMTCM = 0 ~ 14°C DiffDDO = 0 - -900 degree days DiffNFFD = 0-90 days These northward seed transfer ranges should be considered as limits rather than best ranges in coastal BC, as the predictive models only accounted for about 40% variation in the volume growth response variable. In applying the predictions, one should check the ultimate growth-gain in a given situation and chose the best possible seed source for planting. Ultimately, one could expect a mean volume-gain over local source of about 24.2% that is associated with a northward seed transfer of 5.5° of latitude (i.e., DiffLAT = ^5.5°, see Table 3-4). That is, an average ultimate volume-gain of 24% at year 20 is expected by transferring seed 5.5 degrees of latitude north as compared with using local seed source (Table 3-4). Note, this is only applicable to seed transfer in the oblique southeast-to-northwest direction along the coastal line of BC. Common to all the predictive models is that, the predictions within the range of DeviVOL20(%) of -20 to 10% had smaller standard error than those outside this range (Figs. 3-6 101 to -10), which means the predictions are reliable within this range of volume response comparing to local performance. Predictions are least reliable when the transfer towards the upper limits of geoclimatic changes in northward seed transfer (i.e., DeviVOL20(%) > 20%). This was due to the divergent patterns of volume response at wet sites and less-wet sites for the upper limits (see Figs. IV-1 to -5 in Appendix IV). Therefore, applications of the results in Table 3-4 when approaching to the upper limits of northward seed transfer should be cautious. If standard error is considered an additional constraint in achieving growth-gain through seed transfer, northward seed sources should not exceed 3° of latitude (Fig. 3-6), 2.5°C mean annual temperature (Fig. 3-7), 2.5°C mean temperature of the coldest month (Fig. 3-8), 200 degree days below 0°C (Fig. 3-9), and 20 frost-free days (Fig. 3-10). These limits may be considered as average distance of transfer for the species in the whole region of coastal of BC. The above limits are generally in line with the current seed transfer guidelines for Sitka spruce in BC (BC MoF 1995). The predicted rate of volume-gain from northward seed transfer, and the distance of the transfer that allows for higher-than-local performance, differed from one predictor to another. For instance, the ultimate growth-gain was 24.2, 15.6, 13.1,14.5% and 5.1%, that is associated with geoclimatic change of DiffLAT - -5.5 °N, DiffMAT = 3 °C, DifiMTCM = 7 °C, DiffDDO = -500 ~ -450 degree days, and DiffNFFD = 40 days, respectively (Table3-4). In other words, predicted volume-gain from northward seed transfer for DiffLAT was most pronounced, and those for DifiMAT, DifiMTCM and DiffDDO (i.e., thermal-climatic changes) were similar and moderately high, but that for DiffNFFD (i.e., changes in growing season length) was least pronounced. 102 The reason for the observed differences of volume response to different geoclimatic distance variables can be explained by the biological bases of southern provenance outgrowing local source of northern planting area. When a southern provenance transferred to a northern area, the major geoclimatic change that is beneficial to seed transfer practice is the latitudinal change (i.e., DiffLAT), which introduces noticeable lengthened photoperiod of growing season (DiffDAY) and lengthened growing season (DiffNFFD) as well. The thermal-climatic changes from northward seed transfer are generally unfavorable for southern provenances. However, being a coastal species, Sitka spruce is highly sensitive to moisture conditions rather than to thermal-climatic conditions (Chapt.l). Therefore, as long as the thermal-climatic changes in northward seed transfer do not exceed the winter tolerance of a southern provenance (e.g., frost hardness), it can outgrow the local source of northern areas with higher photosynthesis capacity to make better use of lengthened photoperiod of the growing season and lengthened growing season, to offset the effects of lowered thermal-climates. Therefore, latitudinal change (i.e., DiffLAT, actually northwest-southeast oriented) is the primary factor accounting for the reason of southern provenance out-growing northern ones, and this could explain why the growth response to DiffLAT was most pronounced. The predicted DeviVOL20 on DiffMAT, DifiMTCM and DiffDDO were similar and moderately high, indicating the variations in these three Diff- variables were highly redundant on that of DiffLAT. This was also known from the strong correlations of these thermal-climatic change variables with DiffLAT (Table HI-13). : However the changes in NFFD was not so contingent with changes in latitude, so that the volume response due to DiffNFFD was much lower than those due to DiffLAT and thermal-climatic changes. The predicted volume-gain rate for DiffLAT is about five times high as that for 103 DiffivTFFD, suggests that higher-than-local growth performance in northward seed transfer is largely due to geographic changes rather than to climatic changes (since they are mostly unfavorable changes). The question will be addressed in detail in Chapter Four where the effects of photoperiod change is distinguished from thermal-climatic changes in latitudinal seed transfer. 3.4.4. Predictions pertaining to site conditions Although the prediction models are developed by now, it should to noticed that these predictions are still not operable for seed transfer practice. The extent to which the volume-gains can be achieved by northward seed transfer, compared with using local source, highly depends on the planting area's geoclimatic conditions (see below). The actual volume-gain or -loss could be greatly different from the predictions if the planting site is noticeably divergent from the average maritime condition of coastal BC areas. Therefore, to apply these models properly, one needs to look into the contour graphs relating the growth response (DeviVOL20) to both Diff- variables and the corresponding site geoclimatic variables (Figs. 3-11 to -16). Response surface analyses were performed by relating DeviVOL20 to Diff- predictors and the corresponding site geoclimatic variables, in addition with the number one site climatic factor, SMSP (Site Mean Summer Precipitation) as determined by the previous analyses (Section 1.4.3. in Chapt.l). The quality of these response surfaces on the five Diff- predictors and six site climatic variables are listed in Tables HI-l to -6 (see Appendix TTJ), respectively. The lack-of-fit tests for the six models were all not significant, which means all these response surfaces fit well to the second polynomial (see the 'lack-of-fit' F-tests in Tables ILT-1 to -6 in Appendix m). Variations of the response variable were noticeably accounted for by the surfaces (R2 = 0.33 ~ 104 0.50). Partial F-test for the cross products of Diff- predictors with site geoclimatic variables were all not significant at a = 0.05 level, indicating that the predicted general trends in last section are good within the experimental span. This is because the using of DeviVOL20 as response variable moved all the experimental effects, as indicated before (Section 3.4.2.). Canonical analyses of these quadratic response surfaces show that the growth response varied mainly along the axes of site geoclimatic variables, while geoclimatic distance variables 'modify' the rate of growth response to site geoclimatic gradients. This again emphasizes the dependency of growth response upon planting area's geoclimatic conditions. It is clear from these contour graphs that, the milder (or the more southern) the planting site is; the greater the range of northward seed transfer that allows for pursuing higher-than-local performance, and also the greater the amount of volume-gain can be achieved through northward seed transfer. For instance if by looking into Fig. 3-11, suppose at a planting site with latitude of 52° N, by using a seed source from 6° of latitude south of the site, one could expect an average volume-gain of 50% over local source, and a maximum of 60% volume-gain could be achieved by using a seed source approximately 8° of latitude south of the site. Note that this kind of high volume-gain is only theoretically achievable if northward seed transfer is strictly along the outer coast 'fog-belt' (Pojar, et al 1987). However, according to the same contour graph, suppose the planting site is located at 55° N, it is very unlikely to achieve any volume-gain because southern provenances could jeopardize from winter injuries at so high a latitude. Another example can be made by looking into Fig. 3-12, suppose the planting site has an average MAT (Mean Annual Temperature) of 8°C, by using seed source from a place that is warmer than the planting site by 2.2 or 4.0°C in MAT, one would expect an average volume-gain 105 of 20 or 40%, respectively. However, if the planting site has MAT at 6°C, then, to the maximum of 10% volume-gain could be achieved by using seed source from a place warmer than the planting site by approximate 5°C. Once more, if the planting site is even more colder, say, with MAT of 5°C, it is very unlikely that any volume-gain could be achieved comparing to local seed source performance. The same kind of projections can be applied to the remaining contours (i.e., Figs. 3-13 to -16). However, when applying these contours, one should not exceed the experimental span (Table HI-14 in Appendix HI) to make unrealistic extrapolation. At this point, one should also be aware of the fact that, the predictions for thermal-climatic variables were more reliable, though possibly less operable, than for the latitudes, because temperatures are less related to geographic orientation while latitudinal transfer in this species in BC is virtually southeast-to-northwest oriented (see Section 3.3.5). Again, the high volume-gains are only theoretically achievable if northward seed transfer is strictly along the outer coast 'fog-belt' (Pojar, et al 1987). Summer precipitation of test site is the most important factor affecting growth of Sitka spruce (Chapt.l), the volume response to DiffLAT and SMSP (Fig. 3-16) hence should be given particular attention in reforestation with this species. Based on this contour, high-than-local growth could be achieved only when the planting site has a minimum summer rainfall of 500 mm, approximately. Local source remained optimal when planting area has less than 500 mm summer rainfall. This emphasizes that the northward seed transfer of this species should be restricted in the fogbelt along coastal BC. From 500 to 700 mm SMSP for test site, there was a steady increase in volume-gain over local source by northward seed transfer, which is also summarized in Table 3-5. Results show that at a site with 500 to 600 mm summer rainfall, 106 northward seed transfer should be limited between 3 to 8° of latitude, while at a site with 600 to 700 mm summer rainfall, this limit could be expanded up to 12° of latitude. This shows high dependency of volume response on site moisture conditions. The present results somehow contradict with the current seed transfer guidelines of 2 to 4° of latitude in Sitka spruce set by BC MoF (BC MoF 1995; Ying 1997). Though the analytical results here support my projections, one should be aware of the limitations of this analytical approach as stated before (section 3.3.5.). Again, the projections made in the present study are focused on higher-than-local performances only. That is, the range of northward seed transfer allows for higher-than-local performance does not mean the range that allows southern provenance transferred north without suffering from winter injuries of northern planting areas. Another limitation for applying the contour graphs is that the planting areas climatic condition (say, summer precipitation) has to be known in advance before the seed transfer limit can be determined, which is not always realistic. It should also be noticed that volume growth was only evaluated in the present study. In real situations of forestry practice, matters could be more complicated in which the selection of provenances should also take wood quality, disease and pest resistance and many other aspects into account. Table3-5. Volume-gains in northward seed transfer that are conditional upon site summer precipitation. Site major climatic factor SMSP (mm) Northward seed transfer range and volume-gain DiffLAT (°C) DeviVOL20 (%) 500 ~600 -3~-8 0-4 600 ~700 -2--12 4-20 700 ~750 -3--12 20-40 >750 -6--12 >40 107 Contour of GainVOL20 (%) with DiffLAT & SLAT 49 50 51 52 53 54 55 56 SLAT (degree of latitude) Fig. 3-11. Contour graph reflecting the predictions of DeviVOL20 (%) on DiffLAT (= PLAT - SLAT) pertaining to site latitude conditions (SLAT). 108 Contour of DeviVOL20 (%) with DiffMAT & SMAT 5 H O o Q -5 SMAT (°C) Fig, 3-12. Contour graph reflecting the predictions of DeviVOL20 (%) on DiffMAT (= PMAT - SMAT) pertaining to site mean annual temperature conditions (SMAT). 109 Contour of DeviVOL20 (%) with DiffMTCM & SMTCM -12 -10 -8 -6 -4 -2 0 2 SMTCM (°C) Fig. 3-13. Contour graph reflecting the predictions of DeviVOL20 (%) on DiffMTCM (= PMTCM -SMTCM) pertaining to site mean coldest month temperature conditions (SMTCM). 110 Contour of DeviVOL20 •(%) with DiffDDO & SDDO 200 300 400 500 600 700 800 900 1000 SDDO (degree day) Fig. 3-14. Contour graph reflecting the predictions of DeviVOL20 (%) on DiffDDO (= PDDO - SDDO) pertaining to the amount of winter coldness of the planting sites (SDDO). Ill Contour of DeviVOL20 (%) with DiffNFFD & SNFFD 140 160 180 200 220 240 260 280 300 320 SNFFD (day) Fig. 3-15. Contour graph reflecting the predictions of DeviVOL20 (%) on DiffNFFD (= PNFFD -SNFFD) pertaining to lengths of annual frost free period of the planting sites (SNFFD). 112 Contour of DeviVOL20 (%) with DiffLAT & SMSP CD •a ca CD cu CD 0 TJ Q -5 H -10 300 400 500 600 SMSP (mm) ' 700 Fig. 3-16. Contour graph reflecting the predictions of DeviVOL20 (%) on DiffLAT (= PLAT - SLAT) pertaining to lengths of mean summer precipitation of the planting sites (SMSP). 3.4.5. Contours assisting the guide of seed transfer in BC An alternative way of defining the range of suitable seed source under given site geoclimatic conditions is by plotting contour graphs relating growth vigor with pairs of site and provenance geoclimatic variables that are influential on the growth vigor. The method was 113 proposed by Kung and Clausen (1983). In this study, the growth vigor was represented by volume growth (VOL20), and the two dimensional independent variables were the effective geoclimatic factor for provenance origin and for test site. Response surface were constructed by using VOL20 (at provenance-by-site mean level) as the dependent variable, site and the corresponding provenance geoclimatic factors as the two independent variables (i.e., SLAT with PLAT, SMAT with PMAT, and etc.), using the SAS RSREG procedure. The five effective geoclimatic variables (i.e., LAT, MAT, MTCM, DDO, NFFD) were determined by the previous redundancy analysis (see section 3.4.1.), while MSP is adopted because SMSP is the predominant climate factor for test site, as determined by previous analyses on the climatic sensitivity in Sitka spruce, (see section 1.4.3. for detail). The models were all set to the second polynomial to comply with the previous predictions and response surface analyses. The contour graphs from these models are presented as Figs. 3-17 to -22, and the quality information of these models are listed in Tables HI-7 to -12 (see Appendix III). The results indicate that the effects for the six pairs of geoclimate factors on the volume productivity were all highly significant (p < 0.0001) based on partial F-tests. This agrees with the previous analysis of variance result that the effects of Site and Provenance were both highly significant on the growth measurements (Chapt.l). Variations of the response variable are well accounted for by the models for MTCM, NFFD and MSP (R2 ranging from 0.40 to 0.43), but not so well accounted for by those for LAT, MAT and DDO (R2 ranging from 0.18 to 0.29). Linear effects of these geoclimatic factors were all significant. The quadratic effects and the provenance-by-site interactions were not always significant among these factors, implying that milder site is generally more favorable for volume growth. Significant provenance-by-site 114 interactions (i.e., G x E interactions) were detected as PMTCM x SMTCM and PNFFD x SNFFD, which proved that winter coldness and length of growing season were the two major causes of the G x E interactions in growth of the provenances. However, as expected from the previously mentioned bias sources (i.e., location of test sites and weevil damage), the quadratic smoothing procedure of the response surfaces generated significant portion of 'lack-of-fit' error in total errors for all these models (see 'Lack-of-fit' tests in Tables HI-7 to -12 in Appendix HI). This emphasizes the descriptive rather than inferential nature of the contours from these models, which could only be used in assistance with the previous general predictions and response surface analyses. 115 Contour of VOL20 (dm3) vs SLAT & PLAT 50 51 52 53 54 55 56 SLAT (degree of latitude) Fig. 3-17. Contour graph of VOL20 (dm3) on SLAT (site latitude) and PLAT (provenance origin's latitude). 116 Contour of VOL20 (dm3) vs SMAT & PMAT 3 4 5 6 7 8 9 SMAT (°C) Fig. 3-18. Contour graph of VOL20 (dm3) on SMAT (Site Mean Annual Temperature) and PMAT (Provenance origin's Mean Annual Temperature). 117 Contour of VOL20 (dm3) vs SMTCM & PMTCM -12 -10 - 8 - 6 - 4 - 2 0 2 SMTCM (°C) Fig. 3-19. Contour graph of VOL20 (dm3) on SMTCM (Site Mean Temperature of the Coldest Month) and PMTCM (Provenance origin's Mean Temperature of the Coldest Month). 118 Contour of VOL20 (dm3) vs SDDO & PDDO 200 400 , 600 800 1000 SDDO (degree day) Fig. 3-20. Contour graph of VOL20 (dm3) on SDDO (Site annual accumulated Degree Days below 0°C) and PDDO (Provenance origin's annual accumulated Degree Days below 0°C). 119 Contour of VOL20 (dm3) vs SNFFD & PNFFD 140 160 180 200 220 240 260 280 300 320 SNFFD (day) Fig. 3-21. Contour graph of VOL20 (dm3) on SNFFD (Site annual Number of Frost Free Days) and PDD0 (Provenance origin's annual Number of Frost Free Days). 120 Contour of VOL20 (dm3) vs SMSP & PMSP 300 400 500 600 700 SMSP (mm) Fig. 3-22. Contour graph of VOL20 (dm3) on SMSP (Site mean' Summer Precipitation) and PMSP (Provenance origin's mean Summer Precipitation). Observing Figs. 3-1.7 to 3-22 gives the impression that the directions of the contours line almost paralleled (or obliquely paralleled) along with site geoclimatic gradients, which implies that the selection of planting site is more important than the selection of provenance in Sitka spruce. One maximum and one minimum of the volume production were found corresponding to 121 LAT and DDO, respectively. They were located at SLAT = 52°15" N and PLAT = 45°45" N for LAT (peak), and SDDO = 613 degree days and PDDO = 573 degree days for DDO (valley), respectively (see Table HI-7 and -10 in Appendix III). However, the peak found in LAT (Fig. 3-17) should not be considered a real maximum. It occurred due to excellent growth at the test sites on Queen Charlotte Islands (i.e., Holberg, Rennel Sound, and those on Graham Island), but exceptionally poor growth at the more southern test site (i.e., Head Bay) due to drastic weevil impacts (see Chapt.2). The other remaining contours had no maxima or minima, but exhibited saddle shapes, which means southern provenances grew well in milder and/or southern areas but did not fare well in northern and/or harsh areas. This is caused by the G x E interactions in growth performance of the provenances. From these remaining contours (i.e., Figs.3-18, -19, -21 and -22), one could still perceive that the possible maxima of volume production lie beyond the test range at a direction pointing to milder and/or moister areas of planting site and of provenance origins. If taking VOL20 = 150 dm as a high level of volume growth, the suitable ranges of planting site and provenance origin for this volume productivity could be defined based on the contour graphs of this section. These ranges are summarized in Table 3-6. It is clear from this table that the ranges for planting site selection are much narrower than those for provenance selection. Especially in MSP, high volume productivity is exclusively associated with high site moisture condition, with a minimum summer precipitation requirement of 670 mm, approximately. For the relative importance of planting site selection over provenance selection, this moisture criterion should be considered the number one site factor for reforestation with 122 Sitka spruce. In application, one should be aware of the limitations of this graphic approach as stated before (Sections 3.3.4. and 3.3.5.), keeping in mind that the limits presented here are by-and-large due to lack-of-fit error in the contour constructions. In seed transfer practice, one needs to think about the given site geoclimatic conditions comprehensively while taking the best possible advantage of northward seed transfer. Table 3-6. Suitable geoclimatic ranges at BC for planting areas along with provenance origins if VOL20 = 150 dm is the level of individual tree volume growth at year 20 to be achieved. Influential geoclimatic factor LAT 1 MAT MTCM DDO NFFD MSP Planting site condition 51°~53°N >8°C > 1 °C < 80 dd2 > 260 days > 670 mm Prov. origin condition <51°30"N > 5 °C >-7°C < 100-dd > 170 days indifferent 1. The range for SLAT is least reliable due to the biases from test site locations and weevil damage as stated before. 2. dd = degree days (below 0°C, in this table). 3.5. Conclusions 1. General predictive models have been developed by relating volume growth response to geoclimatic distances between provenance origin and planting site (i.e., DiffLAT, DiffMAT, DiffMTCM, DiffDDO and DiffNFFD) to predict the average volume growth of ecdemic provenances relative to local seed source of the planting site. Volume response predicted was most pronounced for DiffLAT, less pronounced for DiffMAT, DiffMTCM and DiffDDO, but least pronounced for DiffNFFD. Predictive results proved that northward seed transfer is favored for this species, and latitudinal change is the major beneficial change to northward 123 seed transfer practice. Predictions are reliable within the range of -20 to 10% in volume-gain over local source after 20 years from planting. An average ultimate volume-gain was predicted in association with a 5.5 ° of latitude transfer from southeast to northwest along the coast line of BC. Results largely support the current seed transfer guidelines in this species in BC, but also indicate possible wider limits for planting at maritime areas. 2. The range and extent of northward seed transfer which allows for higher-than-local performance are subject to planting area's geoclimatic conditions: the milder (or the more southern) the planting site is, the greater the range of northward seed transfer that allows for pursuing higher-than-local performance, and also the greater the amount of volume-gain can be achieved through the transfer. High dependency of volume response on site moisture condition was found that, a minimum of 500 mm summer rainfall (SMSP) was required to achieve higher-than-local growth performance. In outer coastal areas with high precipitation (e.g., SMSP > 700 mm), about 40% of volume-gain could be achieved by northward seed transfer up to 12 ° of latitude. 3. The geoclimatic ranges of suitable planting site and provenance origin were defined for high volume production of this species by a series of contours, constructed by relating the volume growth to both site and provenance geoclimatic conditions on each predictive factors, respectively. Results indicated that the selection of planting area is much more important than selection of provenance origin. A minimum of 670 mm summer precipitation at planting site is required for high volume productivity of the species. 124 3.6. References BC Ministry of Forests. 1995. British Columbia Forest Practice Code: Seed and Vegatative Material Guidebook. Government of British Columbia, Ministry of Forests, Victoria, Canada. Beuker, E. 1994. Long-term effects of temperature on the wood production of Pinus sylvestris L.. and Picea abies (L.) Karst. in old provenance experiments. Scand. J. For. Res. 9: 34-45. Box, G.E.P., Draper, N.R. 1987. Empirical Model-Building and Response Surfaces. John Wiley and Sons, Inc. New York. Burley, J. 1966. Genetic variation in seedling development of Sitka spruce, Picea sitchensis (Bong.) Carr.. Planta, 99: 283-289. Campbell, R.K., and F.C. Sorenson. 1978. Effect of test environment on expression of clines and on delimitation of seed zones in Douglas-fir. Theor. Appl. Genet. 51: 233-246. Daniel, C. and F. S. Wood. 1980. Fitting Equations to Data. (2nd edition). John Willy and Sons. pp.458. Daubenmire, R. 1968. Some geographic variations in Picea sitchensis and their ecological interpretation. Can. J. Bot. 46: 787-798. Draper, N.R. and Smith, H. 1966. Applied regression analysis. John Wiley and Sons Inc., New York. Falkenhagen, E.R. 1978. Parent tree variation in Sitka spruce provenances, an example of geographic variation. Silvae Genet. 27: 24-29. Hormann, R.K. 1987. North American tree species in Europe. J. Forestry. 85: 27-32. • Jarvis, N.J. and Mullins, C.E. 1987 Modelling the effects of drought on the growth of Sitka spruce in Ilingworth, K. 1978a. Sitka spruce provenance trials three years after planting in British Columbia. In: Proc. IUFRO Joint meeting of Working Parties, Douglas-fir, lodgepole pine, Sitka spruce and true firs. Vol.2 Pp 311-326. Vancouver, B.C. Canada. . 1978b. Study of lodgepole pine genotype-environment interaction in British Columbia. Proc. IUFRO joint meeting. Vancouver, Canada. Vol. 2. B.C. MoF, Victoria, B.C. Pp.151-158. Kovats, M. 1977. Estimating juvenile tree volumes for provenance and projeny testing. Can. J. For. Res. 7: 335-342. Kung, F.H. and Clausen, K.E. 1983. Graphic solution in relating seed sources and planting sites for white ash plantations. Silvae Genet. 33: 46-53. Lines, R. 1973. Sitka spruce IUFRO collection. Rep. For. Res. 42-45. MacSiurtain, M.P. 1,981. Distribution, management, variability and economics of Sitka spruce in coastal British Columbia. M.Sc. thesis. Fac. For., UBC, Vancouver, B.C., Canada. 256p. Malcol, D.C. 1987. Some ecological aspects of Sitka spruce. In: Henderson, D.M. and Faulkner, R. (eds) Proc. Royal Soc. Edinburgh, 93B. Pp. 85-92. Edinburgh, Scotland. Matyas, C. and Yeatman, C.W. 1994. Effect of geographic transfer on growth and survival of jack pine (Pinus banksiana Lamb.) populations. Silvae Genet. 41: 370-376. Mergen, F., J. Burley, and G.M. Furniyal. 1974. Provenance-temperature interaction in four conifer species. Silv. Genet. 23: 200-210. Pojar, J., K. Klinka and D.V. Meidinger. 1987. Biogeoclimatic ecosystem classification in British Columbia. For. Eco. Manag. 22: 119-154. 125 Raymond, CA. and Lindgren, D. 1990. Genetic flexibility — A model for determining the range of suitable environments for a seed source. Silvae Genet. 39: 112~120. Rehfeldt, G.E. 1983. Adaptations of Pinus contorta populations to heterogeneous environments in northern Idaho. Can. J. For. Res. 13: 405-411. 1995. Genetic variation, climate models and the ecological genetics of Larix occidentalis. Forest Ecology and Management. 78: 21-37. SAS Institute Inc. 1990. SAS/STAT User's Guide, Version 6, 4th edition. Vol .1 (pp 943) and Vol. 2 (pp 846). Cary, NC, USA. Schmidtling, R.C. 1993. Use of provenance tests to predict response to climate change: loblolly pine and Norway spruce. Tree Physiology. 14: 805-817. Skroppa, T. and Johnson, O. 1994. The genetic response of plant populations to a changing environment: the case for non-Mendelian processes. In: Boyle, T.J.B. and Boyle, C.E.B. (eds) Biodiversity, temperate ecosystems and global change. Springer-Verlag, Berlin. Pp 183-199. Wareing, P.F. 1956. Photoperiodism in woody plants. Annu. Rev. Plant. Physiol. 7: 191-214. Wetherill, G.B. 1986. Regression Analysis with Applications. Chapman and Hall Ltd., New York. Wollenberg, A.L. 1977. Redundancy analysis — An alternative for canonical correlation analysis. Psychometrika 42(2): 207 -219. Worrell, R. and Malcolm, D.C. 1990. Productivity of Sitka spruce in northern Britain; 1. The effects of elevation and climate. 2. Prediction from site factors. Forestry 63: 105-118. Ying, C.C. 1991. Genetic resistance to the white pine weevil in Sitka spruce. Research Notes No. 106, Research Branch, Ministry of Forests B.C., Victoria, B.C., Canada. Ying, C.C. 1997. Effects of site, provenance, and provenance and site interaction in Sitka spruce in coastal British Columbia. Forest Genetics. 4(2): 99-112. 126 4. Sitka spruce IUFRO provenance trials in British Columbia: old experiment new approach Abstract: The 20-year growth data from Sitka spruce provenance trials in British Columbia were used to simulate volume growth response to rapid thermal-climatic changes after adjusting out the effects of photoperiod change in latitudinal seed transfer. The predictive models are biased and with low precision due to many limitations of this approach. Results predicted that the advantage this species can take from global warming is not substantial if there is not a precipitation increase accompanying global warming trend. The predicted ultimate volume-gain from thermal-climatic changes is 2.3%(±1.3%) on average that is associated with a 1.5°C increase in mean annual temperature, or 4.3%(+2.2%) with a 300-degree-day's decrease in annual accumulated degree days below 0°C which is the amount of winter coldness. If global warming brings about 50-day's increase of frost free days per annum, a 4.4%(±1.9%) volume-gain from the lengthened growing season would be expected. The volume response to elevated winter temperature is predicted to be less pronounced, with just a 1.6%(±1.1%) volume-gain that could be expected from a 3°C increase in monthly mean temperature of January. The study also suggests that volume growth of Sitka spruce could respond more quickly and linearly to an increase in precipitation compared to rapid thermal-climatic changes. Dependency of the volume response to thermal-climatic change upon site summer moisture condition was analyzed. Results show that changes in mean annual temperature could result in positive effect on volume growth only when there is enough summer rainfall at the planting site (i.e., SMSP > 500mm). The 127 higher summer precipitation a planting site has, the greater volume-gain could be expected at that site from elevated thermal-climatic conditions, and the wider range for this species growing at that site to benefit from global warming scenario. At maritime areas with more than 700 mm summer precipitation, up to 20% volume-gain was projected from an increase in mean annual temperature by 5 °C. Keywords: global warming; growth response; provenance trial; Sitka spruce {Picea sitchensis (Bong.) Carr.). 4.1. Introduction It is becoming widely acknowledged that we are entering a period of climate change at an unprecedented rate. The global mean annual temperature has been projected to increase by 2.5°C from 1989 to 2050 as a result of the greenhouse effect (Schneider 1989). If this global warming scenario is true, its impact would be more pronounced at higher than lower latitudes areas. For instance, a winter temperature increase of 7°C and a summer temperature increase of 4°C were projected for British Columbia region by Canadian Climate Program Board (CCPB 1991). Although a great deal of uncertainty still surrounds these, projections, most meteorological data forewarn about the global warming trend. The real uncertainty seems to be related to the level of warming and how it will affect the amount and distribution of precipitation (Ledig and Kitzmiller 1992). 128 Assuming that rapid climate change is taking place, there is concern about how climate change will affect tree growth and survival. Trees, with their long life spans, are less able to respond by migration and genetic selection in a relatively short period of time. To date, tree responses to the expected rapid climate change are largely unknown. The responses in growth rate and its directions (i.e., negative or positive) were addressed by physiologists with growth chamber experiments. These models are based on extrapolation from short term trials conducted on seedlings under artificial settings. There is a lack of experiments with mature trees under natural conditions with temperature fluctuations and biological effects retained intact. This is vital, long-term growth responses of trees are different from those in seedlings, and trees grown under natural conditions may be different from those seedlings grown in growth chambers. An effective way to measure the response of a tree species to climate change is to establish a long-term experiment where trees of known origins and genetic background are planted in many climatically different environments. The growth of the trees in the experiment is measured periodically, preferably well past reproductive maturity. Coincidentally, this kind of experiment has been conducted by foresters for more than 200 years, under the name of provenance trial (Langlet, 1971). This idea was recently advocated by the Finnish scholar, Koski (1989). With widening recognition of the global warming trend, the issue aroused much interests and several studies have been reported (Matyas 1994; Schmidtling 1994; Beuker 1994 and 1996). The main advantage of this new approach is that many old provenance trials with most of the commercial tree species are already in place with data available, so that one can easily make the best possible use of them within the limitations set by the experimental designs. 129 Sitka spruce, Picea sitchensis (Bong.) Carr., a fast growing coastal conifer native to North America, occupies a long, narrow strip along the Pacific coast spanning over 22 degrees of latitude (Daubenmire 1968). Its high growth rate, great stumpage and wood quality made it a recommended species for reforestation in coastal areas of British Columbia (BC) where the white pine weevil (Pissodes strobi (Peck)) threat is low (Ying 1997). In order to exploit the potential of genetic superiority of ecdemic (non-local) provenances over local seed source in growth and screen for weevil resistant provenance, the Research Branch of British Columbia Ministry of Forests (MoF) launched a long-term project of Sitka spruce provenance trials in early 1970s which included 43 IUFRO Sitka spruce provenances, collected along the coast from Oregon coast to south Alaska, tested at 11 test sites of coastal BC areas (Illingworth 1978; Ying 1991 and 1997; Chapt. 1). The main factor considered in these trials is latitudinal seed transfer, with changes in elevation and edaphic conditions as secondary (For details of the locations of provenance origin and test site, see Tables 1-1 and -2 in Chapt. 1). Growth of individual trees in these trials were measured periodically over the first 20 years since planting. These growth data, along with the geoclimatic data for provenance origins and test sites, provided a good opportunity to simulate growth response to rapid climate changes, if the effects of photoperiod change from latitudinal transfer are eliminated. The objective of this chapter is to simulate growth responses of Sitka spruce to the potential of rapid climate changes (focusing on global warming trend) in term of volume growth at the 20th year after planting. As latitudinal trend in growth of the provenances by year 20 has been proved to be relatively stable compared to early height growth (Chapt. 1), the projections made in this study can be considered as long-term growth response. 130 There are quite a few limitations, however, for using provenance trial data to predict tree growth response to rapid climatic changes. First, provenance trials were not designed for this simulation purpose. That is, symmetric latitudinal and longitudinal strictures and random sampling for test site location and provenance origin are rare, and consequently, fine gradients of climatic change are rare in old provenance trials. Second, it is possible that changes in growth rate could be over-estimated since climate change is a gradual process in any given location, while in provenance trials the climate change associated with long-distance transfer is immediate. Third, latitudinal seed transfer results in both temperature and photoperiod changes, which means the simulation may over-estimate the growth response due to temperature change only. In this study, although the latitudinal effects were accounted for by adjusting out the effects of photoperiod changes in growing season, this procedure also removed certain effects of thermal-climatic changes as well, because the thermal-climatic change variables are closely correlated with latitudinal change and thus with photoperiod change (see Table III-13 in Appendix III). Therefore, it is almost impossible to remove the effects of photoperiod change from latitudinal seed transfer while retaining the effects of thermal-climatic changes intact. Thus, the predictions in this study could under-estimate the effects of thermal-climatic changes only. Forth, the predictions only focus on climatic changes between test site and seed origin, but in fact, other environmental changes, e.g., soil differences between test site and origin and among test sites, could also affect the tree's growth response noticeably. Another big limitation of this study is due to the fact that the range of provenance origins exceeds the range of test sites considerably (see Fig. 1-1 in Chapt. 1), which means, it is impractical to derive the autochthonous growth performances of the provenances at their origin places. Therefore, the growth response presented 131 in this study is relative to local performance rather than to autochthonous performance of a provenance itself. This is vital, because high-than-local growth is generally associated with northward see transfer (see Chapt.3), that is, positive growth response is associated with lowered thermal-climatic conditions for the provenances transferred north, which certainly can not be true for the growth response relative to autochthounous performance. Therefore, in this study, the thermal-climatic changes were given reversed signs between test site and provenance origin, i.e., the climatic differences were defined by subtracting the values for provenance origin by the values for test site, in stead of vise versa. How much bias having resulted from this manipulation is largely unknown, but this is the best I can do to approach the research purpose. The use of growth response relative to local than to autochthounous performance could also result in under-estimations when seed is transferred south while over-estimations when seed is transferred north, if considering that the local performance is usually higher than sources north of the site but lower than sources south of the site in this species. At this point, contradictorily, one might also want to argue that the predictions would be over-estimated when seed is transferred south but under-estimated when seed is transferred north, if accounting for the fact that the autochthonous performance of a southern provenance is generally greater than local ones but that of a northern provenance is generally poorer than local ones. Finally, the predictions made are under the assumption that trees are genetically well adapted to new climatic conditions, which could also bring about over-estimation of the actual growth response. In conclusion of the above limitations, the predictions made in this chapter should be considered descriptive rather than inferential. I expect effects resulting in under-estimation are to some extent offset by those over-estimating response. Thus, the predictions made here are probably as accurate as extrapolations 132 based on growth chamber data, while taking the advantages of available data and the fact that the trees are grown in natural environments. 4.2. Materials and Methods Data from three series of Sitka spruce provenance trials in coastal BC (supplied by the Research Branch of MoF), which include the test of 43 Sitka spruce IUFRO provenances at 11 sites, were used in this study (Ying 1991 & 1997; also see Chapt.l for details). The experiments have completely randomized block design, with 4, 5, 6 or 9 blocks at different test sites. Within a block each provenance tested is represented by a 9-tree-row plot. Not all provenances tested at all sites, thus there are 220 provenance-by-site means for the simulation process (see Chapt.3). For long-tern simulation, the growth measurement selected for analysis is volume in the most recently measured year, i.e., the 20th year after planting. Individual tree volume was calculated by Kovas' volume function (1977) and termed as VOL20. Volume growth response is expressed in logarithmic values for the ratio of an ecdemic provenance's growth performance over the local performance where the ecdemic one was tested, and symbolized as DeviVOL20. The estimation of the ecdemic provenance's growth performances and the local performances were described in Chapter Three. Macro-climatic data were obtained from the nearest weather station to a test site as well as to a provenance origin place. With prefix "S-" added for all site climatic variables, and "P-" 133 for all provenance climatic variables, the acronyms of the six climatic variables that describe thermal-climatic, photoperiod and summer moisture condition are as follows: MAT = Mean Annual Temperature (°C) MTCM = Mean Temperature of the Coldest Month (i.e., January) (°C) DDO = annual accumulated Degree Days below 0°C (°C) NFFD = annual Number of Frost Free Days (day) DAY = accumulated daylength of a growing season (April ~ October) (in hours, calculated as a function of latitude) MSP = Mean Summer Precipitation (mm) (May ~ September) The climatic differences between provenance origin and test site were obtained by subtracting the values for provenance origins by those for test sites where the provenances tested, correspondingly. These climatic difference variables were named with a "Diff-" prefix, i.e., "DiffMAT", "DiffMTCM", and etc. For instance, DiffMAT = PMAT - SMAT. In order to eliminate the effect of photoperiod changes from the gross effect of latitudinal seed transfer, curvilinear regression (quadratic) was performed on DeviVOL20 with DiffDAY, a variable that defines the photoperiod difference between provenance origin and test site in growing season. The residual (the observed value subtracting the predicted value) from this regression model is the net growth response in VOL20 mainly to thermal-climatic changes. It was therefore termed as "NDeviVOL20" (Net growth Deviation in VOL20), and was used as the growth response variable in the subsequent simulations for rapid climate changes. Thus, the only difference between the simulation models in this chapter and those in Chapter Three is that the observed volume response (DeviVOL20) is substituted by NDeviVOL20 to distinguish the effects of thermal-climatic changes from photoperiod change in latitudinal seed transfer. 134 The response in NDeviVOL20 to thermal-climatic changes were quantified by curvilinear regressions relating NDeviVOL20 to the thermal Diff- variables, respectively. The prediction models were set up as quadratic functions of the Diff- variables, respectively. The predicted values were transformed back from logarithmic values into percent deviation of volume growth relative to the growth under current temperature conditions (i.e., local performance) for reporting results. The ranges of thermal-climatic changes that did not cause growth-loss on average were also defined by the mean predicted values for NDeviVOL20. Growth response to different moisture conditions of planting site was also explored by examining the volume productivity (VOL20, transformed into logarithmic value) of the 11 frequently tested provenances (see Appendix II) to test site summer precipitation (SMSP). Dependency of the predicted growth response to thermal-climatic changes on test site summer moisture condition was analyzed by two-dimensional response surface analysis using second degree polynomials. DiffMAT was selected as the major thermal-climatic factor while SMSP as the most influential site moisture index in this response analysis. Results were presented in a contour graph corresponding to the smoothed surface. All data analyses were performed with SAS procedures (SAS Inc. 1990). Growth response variables were transformed into natural logarithmic values in all the analyses to approach normal distribution, but interpreted in the original units in reporting. To avoid scale problems with growth and climatic variables, response surface analysis was based on standardized data (i.e., subtracting the mean and then dividing by the standard deviation of that mean). All tests of significance are valid under the assumptions of normality and homogeneous 135 variance of the response variables across different levels of experimental effects and climatic gradients. The significance criterion was set at a = 0.05 level unless otherwise specified. 4.3. Results and Discussions 4.3.1. Growth response to thermal-climatic changes only Net volume growth response to thermal-climatic changes were obtained from the curvilinear (quadratic) regression of DeviVO120 with DiffDAY. The regression model is highly significant (p < 0.0001), with relatively high model R 2 (coefficient of determination) as 0.296. The residuals from this regression model were used as the net volume response variable, i.e., NDeyiVOL20, for the subsequent simulations. The effect of photoperiod change was eliminated from the original volume response to both thermal-climatic changes and photoperiod change from latitudinal seed transfer. This is evident from the residual plot of NDeviVOL20 with DiffDAY (Fig. 4-1), — no discernible trend left for the growth response to DiffDAY. However, as mention before, the removal of photoperiod effects, i.e. latitudinal effects, also eliminated the effects from thermal-climatic changes to some extents, because the latter ones are contingent with, or say redundant on the changes in latitudes. This could be seen from the pair-wise correlations between DiffDAY (or DiffLAT) and Diff- thermal variables (Table III-13 in Appendix III), and was also discussed in Chapter Three (Section 3.4.3). The correlation coefficients between these variables ranged from 0.65 to 0.79, and were statistically significant (p < 0.0001). Therefore, the use of NDeviVOL20 as the response variable for the subsequent simulation will under-estimate the growth response due to thermal-climatic changes only. 136 Scatter plots were used to observe the growth response trend by plotting NDeviVOL20 (in natural logarithmic values) versus the thermal-climatic change variables, respectively (Figs. 4-2 to -5). One can see from these plots the quadratic response trends of the net growth response to all these thermal-climatic changes. However, the trends are quite flat, which means that the growth did not respond highly to thermal-climatic changes between seed source and test site. This agrees with the previous results that Sitka spruce is highly sensitive to moisture conditions, not to thermal conditions in later growth (Chapt.l). However, as stated before, the response to thermal-climatic change effects were depleted by the removal of photoperiod changes, so that the. flat response curves are rather expected. There were large amount of variations of NDeviVOL20 surround these quadratic trends, suggesting that growth response of individual provenance could be greatly different from the mean predicted growth response, that is, the predictions could not be precise, though statistically unbiased prediction can be achieved. Regression analyses proved that the volume responses fit well to the quadratic curves for the four thermal-climatic change variables separately (Table 4-1). The prediction models are all highly significant (p < 0.0001), but the values of model R2's are relatively low (ranging from 0.11 to 0.24, and in the fact that these R2's are not adjusted for the null intercept). Low coefficients of determination imply that a great portion of variation in the volume response was not accounted for by the prediction models. This is partly attributable to the adjustment for DiffDAY which removed the effects of thermal-climatic changes to certain extent. As a result, the precision of the predictions are low, especially below the peak region of the response curves. However, the residual means for the prediction models are not significantly from zero at a = 0.01 level (Mest on the residual means in Table 4-1), which means the predictions are statistically unbiased. 137 J uAJ i i CO cu 33 o D O 13 CN C J CO O o s @ . Q g 3 cn -o £ | ^ O CD O _a s CO cn > >N ^ .2 =3 J 5 O _o "a, CD •^ f ±i '5 M 00 OZIOAIAarjN E .2 5-O •r- CN c -o an o >L2 FD OA! -o > a De 5 CD -*-» Z CO -o. 3 , 4> C+H oo o i -CD O J2 a. > o 'CDS E 00 CD _3 j>N catt CO > ical oo o CO in thm ladr; •4 an — ' C T 00 E 00 CD •.••••ivl 021OA!AaQN i — r r r CD O 0) CD D) CD •a in P i I— o b co'i" 13 C CO u f—I o CN O > CD Q 3 T3 3 CD s-, +-» 13 CD OSIOAIAeaN a. CD CD oo 00 00 E o o o S _>N "c0 o -o CO 3 cr CD 00 CI (031OA!A8QN=) lenpisay Table 4-1. Estimated parameters for the prediction models relating NDeviVOL20 to the four thermal-climatic change variables, respectively, along with the quality information for these models. (1) Ln(NDeviVOL20) = /(DiffMAT, DiffMAT2) Factor Parameter St. Error of the T for Ho: Pr > |T| Estimate Estimate Parameter = 0 DiffMAT 0.0293827710 0.00908803 3.23 0.0014 DiffMAT2 -0.0093562460 0.00198807 -4.71 O.000.1 Model R2 = 0.1086, Root MSE = 0.3125, n = 212, Residual Mean = 0.0408 (t-testfor H0: Residual Mean = 0 is | T | = 1.90 < J m s = J-97)i:m . (2) Ln(NDeviVOL20) = / (DiffMTCM, DiffMTCM2) Factor Parameter St. Error of the T for Ho: Pr > |T| Estimate Estimate Parameter = 0 DiffMTCM 0.0103435873 0.00365795 2.83 .0.0051 DiffMTCM2 -0.0016755210 0.00032725 -5.12 O.0001 Model R2 = 0.1361, Root MSE = 0.3077, n = 212, Residual Mean = 0.0529 (t-test for Hp: Residual Mean = 0 is T 0 M S = 2.60 > | T \ = 2.50 > Tp^s = 1.97) (3) Ln(NDeviVOL20) = / (DiffDDO, DiffDDO2) Factor Parameter St. Error of the T for Ho : Pr > |T| Estimate Estimate Parameter = 0 DiffDDO -0.0002696064 0.00006080 -443 <0.0001 DiffDDO2 -0.0000004297 0.00000008 -5.21 O.0001 Model R2 = 0.2403, Root MSE = 0.2 796, n = 173, Residual Mean = 0.0455 (t-testfor H0: Residual Mean = 0 is T0M5 = 2.60 > | T] = 2.14> Tomi = 1.97) (4) Ln(NDeviVOL20) = / (DiffNFFD, DiffNFFD2) Factor Parameter St. Error of the T for Ho: Pr>|T| Estimate Estimate Parameter = 0 DiffNFFD 0.0017662612 0.00038191 4.62 <0.0001 DiffNFFD2 -0.0000181254 0.00000378 -4.79 O.0001 Model R2 = 0.1620, Root MSE = 0.2937, n = 173, Residual Mean = 0.02779 , (t-test for H0: Residual Mean = 0 is \ T | = 1.24 < T0M25 = 1.97) 139 2 Of the four prediction models, the model based on DiffDDO has the highest R value (Table 4-1). This is in accordance with the previous results that, it is the difference in winter coldness ('harshness') between provenance origin and test site that contributed the most to the growth deviations of ecdemic provenances from local source among all the thermal-climate differences (Table 3-2 in Chapt.3). Using these prediction models, the volume responses to rapid climatic changes are quantified within the experimental span (see Table III-14 in Appendix III) and presented in Table 4-2, along with the standard errors of the mean predicted values. These predicted response show that the volume-gain from thermal-climatic changes alone was rather small when compared to that with both thermal-climatic changes and photoperiod changes considered (see Section 3.4.2., Chpat.3). For instance, in the previous prediction for DiffMAT, a 1.5°C difference of mean annual temperature between provenance origin and planting site was predicted to result in an average volume-gain of 11.2% over local source (Table 3-5 in Chapt.3), but now will only result in an average volume-gain of 2.3% when the effects of photoperiod change were eliminated (Table 4-2). However, the predictions made here are very likely to underestimate the growth response to thermal-climatic changes. As stated before, the effects of thermal-climatic changes are largely redundant on the effects of photoperiod change. The removal of photoperiod change effects also removed the effects of thermal-climatic changes to certain extent. Therefore, these predictions are considered biased. Nevertheless, unlike the predictions for the other Diff-variables, the predictions for DiffNFFD were almost unaffected by the adjustment of photoperiod change. For instance, in the previous prediction for DiffNFFD, a 50-day increase of annual number of frost free days (NFFD) was predicted to bring in an average of 4.9% volume-gain 140 (Table3-5, Chapt.3). Similar prediction was found that upon a 50-day increase of NFFD, a 4.4% volume-gain was predicted when the effects of photoperiod change were eliminated (Table 4-2). This indicate that the effects ,of changes in growing season length were almost not redundant on that of photoperiod changes (also see Section 3.4.3.), and thus, the prediction for DiffNFFD could be considered more reliable than for those temperature change variables. The predicted ultimate volume-gain from thermal-climatic changes is 2.3%(±1.3%) on average that is associated with a 1.5°C increase in MAT, or 4.3%(±2.2%) with a 300-degree-day's decrease in DDO which is the amount of winter coldness, or 4.4%(±1:9%) that is associated with a 50-day's increase in NFFD (table 4-2). The volume response to elevated winter temperature is predicted to be less pronounced, with just a 1.6%(+1.1%) volume-gain that could be expected from a 3°C increase in MTCM. These results seem suggesting that the advantage Sitka spruce could take from global warming (thermal effect only) is very limited. Instead, either rapid warming or 'cooling' trend is predicted to be more likely to bring in volume-loss rather than -gain in this species. Evidently this is because an increase in temperature affects growth positively only within the physiological and ecological tolerance limits of the species (Matyas 1994). The ranges for Sitka spruce to buffer rapid thermal-climatic changes without suffering volume-loss are defined according to the mean predicted values (Table 4-2) as follows: DiffMAT =0-3.0 °C, DiffMTCM =0-6.0 °C, DiffDDO = -600 - 0 degree days, which is not far from the projections of global warming trend (Harrington 1987; CCPB 1991) for Brititsh Columbia region. That is, probably we do not have to worry about the impact of global 141 warming scenario on Sitka spruce at least in BC, neither can we expect big 'bonus' of stumpage increase of this species from elevated thermal-climate conditions only, if global warming does advent in the next century. The present projections are similar to those of Beuker (1994), Matyas (1994), and Schmidtling (1993) who all demonstrated that tree growth responses to thermal-climatic changes are quadratic, though the rates of growth-gain from elevated thermal-climatic conditions are different from case to case. The major difference between the current projections with Sitka spruce and other projections with other species is that, while the above mentioned authors predicted high growth response of other tree species to changes of annual temperature sum above 5 °C (i.e., the amount of warmth), I found that Sitka spruce did not respond significantly to changes in the amount of warmth (p = 0.6367, R 2 = 0.0053), but respond to changes in the amount of coldness significantly (e.g., with DDO, p < 0.0001 and R 2 = 0.2403, Table 4-1). The difference is mainly attributable to this species' unique coastal nature. Being a coastal species, Sitka spruce is highly sensitive to moisture, not temperature, conditions (Chapt. 1). As long as a provenance of this species grows in an area without severe winter injury, its growth responds to moisture abundance rather than to the warmth of the area, and thus we can not detect the growth response to changes in warmth but to changes in coldness (e.g. DDO and MTCM). This draws concerns on another bias source of the current projections that, the predictions did not take the changes.in moisture conditions into account which should have more pronounced influence on the species, and which, according to the CCPB predictions, is likely to occur with a decrease of up to 10% summer precipitation in coastal BC areas by the middle of the next century (CCPB 1991). 142 -a OJ CD 60 G c "5, T3 C cj C '5b CD CD c C CD > o G CD CD CD -O cn CD 00 C " c-J o c3 B i "3 CD E o ccS ov O cd v? O CS -J O > Q fa to > CD CD > , Q 4= ^ CN CD J> cn 43 C o a a.— cn tn I CD O c ccS c CD > O •3 ol CD CD > 43 o > -o CD CN i CCS Cd vo r- VO CO CO OS oo •-J CN O VO wo o .7±5. ON i CO +1 CO CO +1 CO +1 CN +1 l> +j m t— CO co CO d CO vd CN CN CN ON wo CN wo CN VO CN 9 +i +i wo r-os vd o\ OS r— co d d +1 +1 CN OS o o +1 +1 © vq d H N r i Vi CN r- ^ © rH +1 +1 oo r— © © . V) OS +1 CN t> •o r> CN CN +1 +1 CN v> •-t vo CN co +1 VO rn CN CO •* +1 +1 +1 CNi vi H d I N CN CN OS WO v i +1 oo Tf wo •-f .—< CO Os wo d 00 +1 +1 +1 +1CN CO vo © d CO vd d <—1 '—1 <—1 CN © © © C=> (=> 2 £ 2 ® °> . . . • • © © © © © © t> VO Tl Tf CO CN ° , , _ , © © © © © © © © © ' 7 ° — cNro--tinvor-oooN o o o o o o o o © i - H c N - O T r v i v o t > OS oo VO CO ON wo © ^ t 00 CN VO o oo VO "sf OS "Ct *—< OS vq •"Cf CO CO co CO CN CN CN H ,—4* +1 +1 +1 +1 +1 +1 +1 +1 +1 +1 + l oq oo OS © CO 1/0 OO' CN t-; od wi CN d •-cf ^ H 00 vd CO CO CO CO CO CN CN CN '—1 '—1 r—l © CO vo 00 © - f 00 VI C<1 — < - - H d d d +1 +1 +1 +1 +1 p On OS Tf O S VO Tf co >-i © © © © © © O O O O O O O © © — © v > © V ) © v ) © v ) © v > © v ) © v ) © S c c r~ r-vovow j i r j ' - f " * -o*o<NCN ' H — © 00 in Tf 00 Tf CN o CO VO OS CO 1—1 Vi o Ti © Tl © © d d © CN CN CO CO -tf +1 +1 +1 +1 +1 +1 +1 +1 +1 +1 +1 +1 +1 © CN ro CO CO © "1 00' 00 © c4 CO co --t --* CO CN © 00 o o CN OO CO (— f— OO OO OS +1 +1 +1+1 +1 +j +1 +j v o c N O v o o o o P " ^ 0 0 . d ( N c n v i ^ 2 2 2 oo CN r- r-c m wo - H \q CN r-w-i vd vd © . © © © © © © © © © • © © © © o o o © ~ in o in o v i o o v i o ifi o v i a K o K CNICOCO-^^Wlf tVOVOt^l^oOOOOS © © K •" w — — n © © ov 2 — — i r- © © CN r— VO © CN CO CO •-cf •-cf wo 00 wo CN OS uo CN oo © vq CN f^ CO iri >n St CO CO CO CN CN +1 +1 +1 +1 +1 +1 •+l +1 +1 +1 9±1 6±1 5±1 7±( co l> *-' VO wo CO CO WO 9±1 6±1 5±1 7±( t-^  ^ f d co © l> "Cf 1—( 00 1 vd 1 • CN 1 1 CO 1 CO 1 CO 1 CN 1 CN • CN 1 1 1 -7 1 o © . ro i> VO Os ITl w o\ Tf OS OS wo o d OS © wi vq © CN CN CO CO "sf wo +1 +1 +1 +1 +1 © CN. CN wo —< CN vd 06 VD 0 CN vq ,—™^ r~ CN r-w-j vd vd +1 +1 +1 +1 +1 00 00 •On CN CO VO. OS CO ' 1 - H ' — 1 ' — 1 CN r~ o r~ © 00 —vq os co wo 00 00 co On On +1 +1 +1 +1 +1 wo. © wo ^ vo _ 6 cn.h CN CN CO CO CO V) Tt CO CN — © - oo r-vo i ^ T f f o <N-< O T _ r l,'fO 'Tt in vo r- a l o , O H N « 1 ' l f i « h - O O N O « CN VO ^ f 00 ^ f 00 CN wo wo CN t~~ CN vq '—1 wo vd vd wo wo •<* r f CO CO CN +1 +1 +1 +1 +1 +1 +1 +1 +1 "Cf OS wo wo wo 00 ^ f ^ f © wo CO OS w-i CN •-Jf 1 CO 1 CO 1 CN 1 CN 1 1 1 1 vo wo vo •* O © i n o o f N o o v o v o - ^ S S S ^ ©•^•OOCOI>CNt>cooOTj-_--. d d d ' H r t N t i n w - i ; 1 ' ) CN O © +1 +1 +1 +1 cNcooor^ocNOcor^vJ-^fcNCN-voco < > « f ? ^ © F H r 4 M N — © ^ r ? ^ : o ? ^ ^ . S VO I - H 'Cf CN 00 f- Tf © vq CN r-; ' — ' wo w-i vd t— t> 00 00 Os Os +1 +1 +1 +1 +1 +1 +1 +1 Tf 00 wo CO Tf wo On ^ H 00 CN vd d 00 —* CN CN CO CO CO V) © Ti © V) © vd vd vi vi Tf TT V J ® V ) © V ) © V ) o Tl © Tl ©• wi wi © V ) © V ) © V ) © V ) © V; O Ti r4r4roco 'Tf-^fvivivdvdt^r> © Ti © 00 00 ON CO -3-4.3.2. Growth response to thermal-climatic change pertaining to site moisture condition High volume growth response to moisture gradient was observed by plotting the provenance-by-site means for VOL20 of the 11 frequently tested provenances (see Appendix II for Chapt.l) to the amount of summer precipitation (i.e. SMSP) of the 11 test sites (Fig. 4-6). This plot indicate that VOL20 was much higher at SMSP > 600 mm level''than at SMSP = 400 mm level. The mean for Ln(VOL20) at SMSP > 600 mm is around 4.7 (i.e., VOL20 = exp(4.7) « 110 dm3) and that at SMSP> 400 mm is around 3.3 (i.e., VOL20 = exp(3.3) *-27 dm3). That is, the increase of SMSP from 400 mm to 600 mm brought in about four times higher volume growth at year 20. The moisture response is so astonishing that it suggests reforestation with Sitka spruce should be applied to areas where at least 600 mm summer rainfall is available (also see Table 3-6, Chapt.3). However, at the four weevil attacked sites, there were no perceivable moisture trends in VOL20 (Fig. 4-6). This could possibly due to the fact that there were not enough weeviled sites to observe the moisture response trend at weeviled sites. The distribution pattern of weeviled sites versus unweeviled sites in relation to the amount of SMSP also suggests that, drier sites are more risky to weevil attack as compared to moister sites. 144 6 i § I 4 c _ l o CN 3 h 5 x Yes © No weevil o 2 L 200 300 400  500 SMSP (mm) 600 700 800 Fig. 4-6. Scatter plot of the provenance-by-site means for VOL20 (in logarithmic values) of 11 frequently tested provenances at the 11 test sites to the amount of site summer precipitation (SMSP). Due to experimental constraints, namely, lack of sufficient and even number of provenances tested at a variety of moisture conditions, lack of fine moisture gradient for test sites within the experimental span, and significant weevil (Pissodes strobi (Peck)) damage occurred at four out of the 11 test sites (see Chapt. 2), I was unable to predict the growth response solely to moisture fluctuations in this study. However, the dependency of the volume response (to thermal-climatic changes) upon site moisture conditions can still be analyzed without high precision. Followed the idea in Chapter Three, the net growth response (NDeviVO120) to thermal-climatic change (represented by DiffMAT only) was related to the number one site climatic factor (i.e., SMSP), using response surface analysis. The response surface model is set to the second polynomial and is highly significant (p < 0.0001) with a model R 2 of 0.147 (see Table V-1 in Appendix V). However, the partial F-test for SMSP shows that generally the effects of 145 SMSP were not significant (p = 0.2552). This seems ironical, but is largely due to the drastic weevil influence at four out of the 11 test sites that might suppressed the tree's moisture sensitivity. Nevertheless, the contour graph of this response surface (Fig. 4-7) can still serve the general discussion purpose of this section: This contour graph clearly shows the dependency of the volume response to the thermal-climatic change upon site moisture conditions. Changes in MAT could result in positive effect only when there was enough summer precipitation at the test site (i.e., SMSP > 500 mm). As SMSP increasing from 500 to 750 mm, there was an increase in volume-gain to temperature rise, in general. The predicted net volume-gains to DiffMAT that are conditional on SMSP gradient are listed in Table 4-3, derived from this contour graph. These predictive results show that the higher the summer precipitation a planting site has, the greater the volume-gain could be expected at that site from elevated thermal-climatic conditions, and the wider the range of global warming trend that is beneficial to the tree growth at that site. Surprisingly when comparing with the projections in previous section of this chapter, at maritime ares with a minimum SMSP of 700 mm, up to 20% volume-gain was projected from the rise of the mean annual temperature by 5 °C, approximately (Table 4-3). However, one should be advised the descriptive rather than inferential nature of the above predictions, due to the limitations as stated before (Section 4.1. and 4.2.) and to the biases from the significant lack-of-fit error in this contour construction. Particularly, one should be aware of the limitation that the predictions made here did not take soil conditions into account. The amount of SMSP is not equal to the amount of available soil moisture, which is the actual affecting agent of moisture on volume production, and which is determined not only by precipitation, but also by soil properties, e.g., depth, texture, slope angle and position as well as 146 soil nutrition contents. Unfortunately, the available information for this study prevents more precise prediction to many environmental variations. Table 4-3. Net expectable volume-gains to changes in mean annual temperature (DiffMAT) that are conditional upon site summer precipitation (SMSP). * Site major climatic factor Net volume-gain to DiffMAT SMSP (mm) DiffMAT .(PC) NDeviVOL20 (%)• 500 -650 "0 - 4 0 -4 650-700 -2-7 0 - 10 700-750, -1 -0.5 or 5-6 10-20 700-750 0.5-5 > 20 147 . Contour of NDeviVOL20 (%) with DiffMAT & SMSP 350 400 450 500 550 600 650 700 750 SMSP (mm) Fig. 4-7 Contour of the quadratically smoothed response surface of NDeviVOL20 (%) to changes in mean annual temperature (DiffMAT) over site summer precipitation (SMSP) gradient. 4 . 4 . Conclusions 1. Quadratic volume growth responses were detected to rapid thermal-climatic changes when the effects of photoperiod changes were accounted for. Predictive models were set up separately 148 for the thermal-climatic change variables,' namely, DiffMAT, DiffMTCM, DiffDDO and DiffNFFD. The predictions were unbiased and with low precision because of the removal of photoperiod change which also removed the effects of thermal-climatic changes to certain extent due to the fact that, the former is highly correlated with the latter ones. Based on the predictive results, the advantage that Sitka spruce could take from global warming alone was not substantial. The ultimate volume-gain from thermal-climatic change was 2.3% on average that is associated with a 1.5°C increase in MAT, or 4.3% with a 300-degree-day's decrease in DDO which is the amount of winter coldness. If global warming brings about 50-day's increase of frost free days per annum, a volume-gain of 4.4% would be expected. The volume response to elevated winter temperature is predicted to be less pronounced, to the maximum of only 1.6% volume-gain that could be expected from a 3°C increase of MTCM. The study also suggests that volume growth of this species could respond more rapidly and linearly to changes in precipitation than to rapid thermal-climatic changes. High dependency of the volume response to thermal-climatic change upon site summer precipitation was found and.analyzed. Results show that changes in MAT could result in positive effect only when there was enough summer precipitation at the planting site (i.e., SMSP > 500 mm). The higher the summer precipitation a planting site has, the greater the volume-gain could be expected at that site from elevated thermal-climatic conditions, and the wider the range for this species growing at that site to benefit from global warming scenario. At maritime areas with a minimum SMSP of 700 mm, up to 20% volume-gain was predicted from an increase of MAT by 5°C, approximately. 149 3. Despite of the many limitations of the predictions, provenance trials remained at present the only available mean of generating long-term growth data of a species grown in rapidly changed climate conditions under natural environments, and thus are a unique resource for simulating mature tree growth responses to rapid climate changes and evaluating the genetic and physiological flexibility of the species in buffering rapid climate changes. 4.5. References Beuker, E. 1994. Long-term effects of temperature on the wood production of Pinus sylvestris L. and Picea abies (L.) Karst. in old provenance experiments. Scand. J. For. Res. 9: 34-45. Beuker, E. 1996. Implications of climate adaptability in provenance trials with Scots pine and Norway spruce in Finland for possible effects of climate warming. D.Sc.(Agr. And For.) thesis, Faculty of Forestry, University of Joensuu, Finland. Box, G.E.P., Draper, N.R. 1987. Empirical Model-Building and Response Surfaces. John Wiley & Sons, Inc. New York. Cannell, M.G.R., Grace, J. and Booth, A. 1989. Possible impacts of climatic warming on trees and forests in the United Kingdom: a review. Forestry 62: 337-364. Harrington, J.B. 1987. Climatic changes: a review of causes. Can. J. For. Res., 17: 1313-1339. Harding, L.E. 1992. Atmospheric Changes in British Columbia. In: Biodiversity in Brititsh Columbia. (Chapt.24). BC Ministry of Environment, Lands, and Parks; BC Ministry of Forests. Pp 323-341. Illingworth, K. 1978. Sitka spruce provenance trials three years after planting in British Columbia. In: Proceedings of IUFRO Joint Meeting of Working Parties, Douglas-fir, lodgepole pine, Sitka spruce and true firs. Vol.2. Pp. 311-326. 1978. Vancouver, Canada. B.C. Ministry of Forests. • Kellomaki, S. and Kolstrom, M. 1995. The influence of climate change on the productivity of Scots pine, Norway spruce, Pendula birch and Pubescent birch in southern and northern Finland. For. Eco. AndManag. 65:201-217. Kendall, M.G. and Stuart, A. 1967. The Advanced Theory of Statistics: Vol2. Inference and Relationship. (2nd ed). Hafner Publishing Company, New York. Koski, V. 1989. Siemensiirrot ja ilmastoon sopeutuminen. Metsantutkimuslaitoksen tiedonantaja. 328: 20-37. (in Finnish) Kovats (1977). Estimating juvenile tree volumes for provenance and projeny testing. Can. J. For. Res. 7: 335-342. Langlet, O. 1971. Two hundred years of genecology. Taxon. 20: 653-722. 150 Larsen, J.B. 1991. Breeding for physiological adaptability in order to counteract an expected increase in environmental heterogeneity. For. Tree Imp. 23: 5-9. Ledig, F.T. and Kitzmiller, J.H. 1992. Genetic strategies for reforestation in the face of global climate change. For. Eco. And Man. 50: 153-169. Matyas, C. 1994. Modeling climate change effects with provenance test data. Tree Physiol. 14; 797-804. SAS Institute Inc. 1990. SAS/STAT User's Guide, Version 6 (4th edition). Vol .1 (pp 943) & Vol. 2 (pp 846). Cary, NC, USA. Schmidtling, R.C. 1994. Use of provenance tests to predict response to climate change: loblolly pine and Norway spruce. Tree Physiol. 14: 805-817. Schneider, S.H.,1989. The greenhouse effect: science and policy. Science. 243:771-781. Skroppa, T. and Johnson, O. 1994. The genetic response of plant populations to a changing environment: the case for non-Mendelian processes. In: Boyle, T.J.B. and Boyle, C.E.B. (eds) Biodiversity, temperate ecosystems and global change. Springer-Verlag, Berlin, pp 183-199. Ying, C.C. 1991. Genetic resistance to the white pine weevil in Sitka spruce. Research Notes No. 106, Research Branch, Ministry of Forests B.C., Victoria, B.C., Canada. Ying, C.C. 1997. Effects of site, provenance, and provenance and site interaction in Sitka spruce in coastal British Columbia. Forest Genetics. 4(2): 99-112. 151 RECOMMENDATIONS 1. Sitka spruce populations are differentiated by photoperiod and winter temperature regimes. Physiological studies are recommended to evaluate the photosynthesis capacity and cold hardiness of southern Sitka spruce provenances, and to screen for relatively cold resistant southern provenance(s) with high photosynthesis capacity for planting at coastal British Columbia. 2. Three provenances, i.e., Kitwanga, Hoquiam and Big Qualicum, deserve further studies on the mechanisms of induced weevil resistance and/of tolerance, and the genetic bases accounting for these properties. 3. To ensure high volume productivity, the planting of Sitka spruce should be restricted to low weevil-hazard areas with a minimum of 600mm summer precipitation. 4. Northward seed transfer is favored when planting this species in coastal BC. The warmer and moister the planting site is, the farther seed can be transferred. A transfer of 6 to 8° of latitude could be applied to most of the maritime favorable areas. 152 APPENDICES 153 o CQ a 00 13 'E <o o . c a c > o o 3 S3 C/5 o u oo CD O 00 'oo C/5 CD GO cc e "5. u w CO CO CD >n O CN L-> o oo CD CD CO -4-» C+H o N ? 1) l-c •C o ccj H oo cu GO !Z2 CZ5 Q u u NW, 4? o o —• T f Tr o o o © © © o TT VO C O ,—1 o r-- oo 00 t-» © © o". t-H CN OO Tf VO Tf O CN Tf 1—< CN cn VO O m CN O OO © O i—' od i n r~-o> CN T f r -m oo o o i n 00 Tf Tf cn cn CN Tf T f cn vo h i n O N c n s. « T f i n r - oo © ^ OO O N O N T—< ,—i cn cn i n cn O N oo cn V O cn T f T f T f •—; CN © cn i oo i > i T f T f i n i n vo CN O N CN i n i n V O CN V O T f r-- O N T f T f i n CN r - O N cn r-» O N r-^  Ov vo 00 © oo 00 cn cn cn © ir, o i—i C N u s-« «• et CU CU CU >< >H >* T f IT) 02 u a SS fl CU > o u a TS CU •<-> Cfl CU fl CU 3 cu .1-cu J3 fl CU bxi fl s as a CU a a < 00 _CD x> .2 'C cd > CD E "o CD +H '55 Z o oo £ o <+h -a CD on CD CD §1 CD CD 4= O D4 o o ss a .1 c - -I cu CD I i T3 u. crj CD 42 o SS u CS o ••^  u ca u CQ u CQ fa cu o CS 6 © >/o m ro TT 00 "/-> CN PH O Z Q r— ro PH ° Z Q lO NO (-- o © ' © Q PH a, Z UH CN 00 r~ o © © -1 £ | O o ON >/-> OS Tf in TT o r o ro >/~) Tf Tf CN — © © © 00 >/-l CN Tf — —I © © ' © cu v-> © Tf r->o CN o © ^ s PH Q Z Q Tf 00 NO © Q tin CL PH E Z £ V I CN NO — oo KO Tf © oo Tf C O o o ro CN © ' © r- oo Tf © © © Oh © 2 Q ro f-Tf o © © Cu ~ 00 ?* IT) — © Tf — — © © © — ON NO O Q fe £ Z PH 1/0 oo ro CN o ON Tf C N N O O r o M n co CN CN ~ O © ' © 2 H < Q n o ro © >/-) ro < < ON NO >/-> CN H CH < < © c~ r- — © ' © — r» oo © ©' © H PH 0O © ro Tf © NO ON Tf N O o r o NO CN ro NO r— Tf —i o ©' © CH £N 00 > 2 £ 00 CN Tf - H © © ' • 00 Tf Tf - H © © ' Q PH Q . fe £ Z PH m ro in ^H © © Z 2 N S IT) O r o CN © Tf Tf CN © ' © ' © ' CH ^ © S S Q ON — ro CN © ©' 00 NO ro — o CH — < > 2 £ ro Tf u-i - H CH H < < — © ro CN tv5 H ro — Tf CN oo H CN 00 Tf — ©' ©' 00 CN Tf — oo ON CH £ £ £ S z ON © CH £ 22 £ s z CN ~ OH £ £2 £ S Z NO © Tf CN © O ON O NO —' -a c co o c -a a CD o e — ro <N o © * £ < £ 2 z CN 00 •/•> CN © O >0 CN © ©' CN oo ON NO NO lO © © © © ON ON oo NO CN r~ no • Tf 1 — 1 oo FS © CN oo © ro ro ^H CO ,— 0 0 o o o © ro Tf Tf iri 1/0 in 1/0 </-> ON T f © • C N T f T f T f r o O o © O r o r o r o r o o CN J O > o CN H 33 so H DC > < o u BS fa u cs to as o o cs to o CS to as u O u cs to CU o K E H top z i n oo i n CN CO CN © d d d ss ^ o\ — r-- — Oh £ 22 to S Z Oh £ S Z t— — CO CN d d Oh £ SS to S Z o i/o r- — to £ ss £ S Z oo o vo ov H^ i n H^ d d d d SS to CN O CO CN d d oo i n WO d d d d to © cu © (/) Q 00 Q s Q Q VO o d © VO VO wo wo 00 oo wo CN CN CN o o wo 00 wo wo O Tf m CN © © cn m Appendix III. Supporting information for the response surface analyses in Chapter Three. Table III-1. Technical report of the response surface analysis for variable DeviVOL20 with DiffLAT and SLAT (from SAS RSREG procedure). ..Quadratic- Response Surface f o r V a r i a b l e DeviVO120 with DiffLAT. and SLAT Regression DF Type I SS .R-Square F-Ratio Prob > F Linear 2 9 .248407 0 .2797 4 5 . 4 6 0 . o :. 0000 ' Quadratic 2 2 .176159 0 ,0658 10 .697 0.0000. Crossproduct 1 0 .279667 0 .0085 2 . 7 4 9 - 0 .0988 ' T o t a l Regress . 5 1.1.704233 0 .3540 2 3 . 0 1 3 0 .0000 Residual DF SS . Mean Square F-Ratio Prob > F Lack of F i t 195 18 .315903 0 .093928 ' ' 0 .463 0 .9911 Pure E r r o r ' 15 3 .045338 0 .203023 T o t a l E r r o r 210 2 1 . 3 6 1 2 4 1 0 .101720 Factor DF. . S-S. Mean Square . ' F-Ratio Prob > F DiffLAT 3 . 11 .403337 3 .801112 37 .368 • 0..0000 SLAT- - 3 1.830416.: 0 .610139 5 . 998 0 . 0006 Canonical A n a l y s i s of Response Surface(based - on standardized data) Eigenvalues Eigenvectors DiffLAT SLAT 0. 187604 -0 .527740 0 .208164 0. 97.8094 0 .978094 -0 .208164 157 Table III-2. Technical report of the response surface analysis for variable DeviVOL20 with DiffMAT and SMAT (from SAS RSREG procedure). Quadratic Response Surface f o r V a r i a b l e DeviVO120 with DiffMAT and SMAT Regression Linear Quadratic Crossproduct. T o t a l Regress DF 2 2 1 5 Type I SS 11.009293 1.875821 0.252602 13.137716 R-Square 0.3437 0.0586 • 0.0079 0 .4101 F-Ratio 60.016 10.226 2 . 754 28.647 Prob > 0.0000 0 . 0001 '• 0. 0985 "0.0000 Residual Lack of F i t Pure E r r o r T o t a l E r r o r DF 200 6 , 20.6 SS • •  •  18 .184086 0.710241 '18.894327 Mean Square • 0.090920. 0.118374 0.091720 F-Ratio 0.768 Prob > •0.7422 Factor DiffMAT. SMAT-DF • 3 3 SS 13. 07.9519 1,656048 Mean. Square 4 . 359840 0.552016 F-Ratio 47.534 6. 018 Prob > 0.0000 0.0006 Canonical A n a l y s i s of. Response Surface (based-on standardized data) Eigenvalues 0.274555 • -0.40.7817 ' , Eigenvectors DiffMAT • SMAT"'-0.311957 0.950096 0/950096 -0.311957 158 Table III-3. Technical report of the response surface analysis for variable DeviVOL20 with DiffMTCM and SMTCM (from SAS RSREG procedure). Quadratic Response Surface f o r V a r i a b l e DeviVO120 with DiffMTCM and SMTCM Regression Linear Quadratic Crossproduct T o t a l Regress DF 2 2 1 5 Type I SS 10.695283 1.240033 0.303758 12.239073 R-Square 0.3339' 0.0387 0.0095 0.3821 F-Ratio 55.657 6. 453 3.161 25 .476 Prob > F 0.0000 0.0019 0.0769 0.0000 Residual Lack of F i t Pure E r r o r T o t a l E r r o r DF 200 6 206 SS 19.099996 0. 692975 19.792970 Mean Square 0.095500 0.115496 0.096082 F-Ratio 0. 827 •Prob > F 0. 6971 Factor SMTCM DiffMTCM DF 3 3 SS 1. 381291 12.196499 Mean Square 0.460430 4 . 065500 F-Ratio 4 .792 42 . 313 Prob > F 0.0030 0.0000 Canonical A n a l y s i s of Response Surface(based on standardized data) Eigenvalues 0.418391 -0.281142 Eigenvectors SMTCM DiffMTCM 0.894314 0.447440 -0.447440 0.894314 159 Table III-4. Technical report of the response surface analysis for variable DeviVOL20 with DiffDDO and SDDO (from SAS RSREG procedure). Quadratic Response Surface f o r V a r i a b l e DeviVO120 with DiffDDO and SDDO Regression Linear Quadratic Crossproduct T o t a l Regress DF 2 2 1 5 Type I SS 10.890052 0.931075 0.243221 12.064348 R-Square 0.4521 0.0387 0.0101 0.5008 F-Ratio 75.621 6.465 3.378 33. 510' Prob > F 0. 0000. . 0.0020 0.0679 0.0000 Residual Lack -of F i t Pure E r r o r T o t a l E r r o r DF 161 6 167 SS 11. 359404 0. 665326 12 . 024730 Mean Square 0.070555 0.110888 0.072004-F-Ratio 0. 636 Prob > F 0.8414 Factor SDDO . DiffDDO DF 3 3 SS 0.304234 11.462213 Mean Square 0.101411 3.820738 F-Ratio 1.408 53.063 Prob > F 0.2422 0.0000 Canonical A n a l y s i s of Response Surface (based on standardized data) Eigenvalues 0.371984 -0.222090 Eigenvectors SDDO DiffDDO 0.816891 • 0.576792 -0.576792 0.816891 160 Table III-5. Technical report of the response surface analysis for variable DeviVOL20 with DiffNFFD and SNFFD (from SAS RSREG procedure).. Quadratic Response Surface f o r V a r i a b l e DeviVOL20 with DiffNFFD and SNFFD Regression DF Type I SS R-Square F-Ratio Prob > F • Linear ' 2 8.314372 0.3452 52.345 0.0000 Quadratic 2 2.248079 0.0933 14.153 0.0000 Crossproduct 1 0.263548 0.0109 3.318 0.0703 T o t a l Regress 5 10.825999 0.4494 27 .263 0.0000 Residual DF SS Mean Square F-Ratio Prob > F Lack of F i t 161 12 473063 0 .077472 0 . 588 0.8757 Pure E r r o r 6 0 790016 0 .131669 To t a l E r r o r 167 13 263078 0 . 079420 " Factor ' DF SS Mean Square F-Ratio Prob > F SNFFD 3 2 691800 0 .897267 11.298 0.0000 DiffNFFD 3 10 587899 3 .529300 • 44.439 0.0000 Can o n i c a l . A n a l y s i s of' Response Surface (based on standardized data) Eigenvectors Eigenvalues SNFFD DiffNFFD 0.193687 0.972673 • -0.232180 • ' -0.893776 0.232180 0.972673 161 Table III-6. Technical report of the response surface analysis for variable DeviVOL20 with DiffLAT and SMSP (from SAS RSREG procedure). Q u a d r a t i c Response S u r f a c e f o r V a r i a b l e Dev iVOL20 w i t h D i f f L A T and SMSP R e g r e s s i o n L i n e a r Q u a d r a t i c C r o s s p r o d u c t T o t a l R e g r e s s DF 2 2 1 5 Type I SS 8 .394406 1. 616696 0. 931924 10 .943026 R - S q u a r e 0 . 2539 0 .0489 0 .0282 0 .3310 F - R a t i o 3 9 . 8 4 2 7 . 673 8 .846 2 0 . 7 7 6 P rob > F 0 .0000 0 .0006 0 .0033 0 . 0000 R e s i d u a l L a c k o f F i t Pu re E r r o r T o t a l E r r o r DF 195 15 210 Type I SS 19 .077110 3 .045338 22 .122448 Mean Squa re 0 .097831 0 .203023 0 .105345 F - R a t i o 0 .482 P rob > F 0 .9874 F a c t o r D i f f L A T SMSP DF Type I . S S 10 .845739 1 .069209 Mean Squa re 3 . 615246 • 0 .356403 F - R a t i o 34 .318 3 . 3 8 3 . P rob > F 0 .0000 0 .0191 C a n o n i c a l A n a l y s i s o f Response S u r f a c e ( b a s e d on s t a n d a r d i z e d d a t a ) E i g e n v a l u e s E i g e n v e c t o r s D i f f L A T SMSP 0 .092387 -0 .596366 -0 .242795 0 .970078 0 . 970078 0 .242795 S t a t i o n a r y p o i n t i s a s a d d l e p o i n t . 162 Table III-7. Technical report of the response surface analysis for variable VOL20 with SLAT and PLAT (from SAS RSREG procedure). Quadratic Response Surface, f o r V a r i a b l e VOL20 with SLAT and PLAT Regression DF Type I SS R2 F-Ratio Prob > F Linear 2 29438 0 0214 2 . 965 0 0537 Quadratic 2 276531 • 0 2009 27 . 854 0 0000 Crossproduct 1 3271 0 0024 0 . 659 0 4178 Tot a l Regress 5 309240 0 2247 12 .459 0 0000 Residual DF Type I SS Mean Square F-Ratio Prob > F Lack of F i t 203 1052091 5182 715227 4 . 104 0 0043 Pure E r r o r 12 15156 1262 994935 Tot a l E r r o r 215 1067247 4963 940141 Factor SLAT PLAT DF 3 3 Type I SS 234344 76860 Mean Square 78115 25620 F-Ratio 15.736 5.161 Prob > F .0.0000 0.0018 Canonical A n a l y s i s of the- Response Surface(based on standardized data) C r i t i c a l Value. Factor Coded Uncoded SLAT -0.073844 52.238876 PLAT -0.597312 45.726544 Pr e d i c t e d value at s t a t i o n a r y point 170.610482 (maximum) Eigenvectors Eigenvalues SLAT . PLAT -36.106074 0.107107 0.994248 -127.291792 0.994248 -0.107107 163 Table III-8. Technical report of the response surface analysis for variable VOL20 with SMAT and PMAT (from SAS RSREG procedure). Q u a d r a t i c R e s p o n s e S u r f a c e f o r V a r i a b l e V O L 2 0 w i t h SMAT a n d PMAT R e g r e s s i o n L i n e a r Q u a d r a t i c C r o s s p r o d u c t T o t a l R e g r e s s DF 2 2 1 5 T y p e I SS 3 4 8 9 2 7 2 0 4 6 7 1 1 8 1 7 3 8 1 2 1 2 R^  0 . 2 6 0 9 0 1 5 3 0 . 0 0 8 8 0 . 2 8 5 1 0 F - R a t i o 38 . 505 2 . 2 5 9 2 . 608 1 6 . 8 2 7 P r o b > F 0 . 0 0 0 0 0 . 1 0 7 0 0 . 1 0 7 8 0 . 0 0 0 0 R e s i d u a l L a c k o f F i t P u r e E r r o r T o t a l E r r o r DF 207 4 211 T y p e I SS 954834 1 2 0 2 . 8 3 6 1 9 4 9 5 6 0 3 6 Mean S q u a r e 4 6 1 2 . 7 2 2 9 4 4 3 0 0 . 7 0 9 0 4 9 4 5 3 0 . 9 7 8 6 0 5 F - R a t i o 1 5 . 3 3 9 P r o b > F 0 . 0 0 7 9 F a c t o r SMAT PMAT •  DF 3 3 T y p e I' SS 3 0 5 6 5 4 7 7 1 2 6 Mean S q u a r e 1 0 1 8 8 5 2 5 7 0 9 F - R a t i o 2 2 . 486-5. 674 P r o b > F 0 . 0000 -. ,0 . 0 0 0 9 C a n o n i c a l A n a l y s i s o f R e s p o n s e S u r f a c e ( b a s e d o n s t a n d a r d i z e d d a t a ) C r i t i c a l V a l u e F a c t o r C o d e d U n c o d e d SMAT . - 1 . 6 5 2 9 5 5 1 . 1 0 6 4 3 1 PMAT - 0 . 3 9 2 3 6 5 5 . 3 1 6 6 5 0 P r e d i c t e d v a l u e a t s t a t i o n a r y p o i n t 1 8 . 6 4 8 3 3 2 ( s a d d l e ) E i g e n v a l u e s 22 . 1 4 4 3 5 7 - 4 5 . 5 6 0 3 9 6 E i g e n v e c t o r s SMAT PMAT 0 . 9 7 2 3 0 2 0 . 2 3 3 7 2 8 - 0 . 2 3 3 7 2 8 0 . 9 7 2 3 0 2 164 Table III-9. Technical report of the response surface analysis for variable VOL20 with SMTCM and PMTCM (from SAS RSREG procedure). Q u a d r a t i c R e s p o n s e S u r f a c e f o r V a r i a b l e V O L 2 0 w i t h SMTCM a n d PMTCM R e g r e s s i o n DF T y p e I SS R 2 F - R a t i o P r o b > F L i n e a r 2 393339 0 2941 54 . 010 0 . 0000 Q u a d r a t i c 2 156064 o' 1167 2 1 . 4 2 9 0 . 0000 C r o s s p r o d u c t 1 19514 0 0146 5 .359 0 . 0216 T o t a l R e g r e s s 5 568917 0 4254 31 .247 0 . 0000 R e s i d u a l L a c k o f F i t P u r e E r r o r T o t a l E r r o r DF 207 4 211 T y p e I SS 767103 1227 . 993637 768331 Mean S q u a r e 3705 .810573 306 .998409 3641 .378115 F - R a t i o 1 2 . 0 7 1 P r o b > F 0 . 0 1 2 4 F a c t o r SMTCM PMTCM DF 3 3 T y p e I SS 5 1 2 3 1 5 8 3 1 8 5 Mean S q u a r e 170772 27728 F - R a t i o 46 .898 7 . 615 P r o b > F 0 . 0000 0 . 0 0 0 1 C a n o n i c a l A n a l y s i s o f R e s p o n s e S u r f a c e ( b a s e d o n c o d e d d a t a ) C r i t i c a l V a l u e . F a c t o r C o d e d U n c o d e d SMTCM - 0 . 2 8 6 0 7 2 - 7 . 3 9 5 6 6 7 . PMTCM - 0 . 0 3 8 0 7 4 - 2 . 1 8 8 3 5 3 P r e d i c t e d v a l u e a t s t a t i o n a r y p o i n t 13 .346541 ( s a d d l e ) E i g e n v e c t o r s E i g e n v a l u e s SMTCM PMTCM •98.591957 0 .981028 0 .193866 - 1 7 . 7 7 2 8 2 7 - 0 . 1 9 3 8 6 6 0 .981028 165 Table III-10. Technical report of the response surface analysis for variable VOL20 with SDDO and PDDO (from SAS RSREG procedure). Q u a d r a t i c R e s p o n s e S u r f a c e f o r V a r i a b l e V O L 2 0 w i t h SDDO a n d PDDO R e g r e s s i o n DF T y p e I SS R 2 F - R a t i o P r o b > F L i n e a r 2 • 1 4 8 9 3 8 0 . 1369 14 . 000 0 0 0 0 0 Q u a d r a t i c 2 4 1 8 4 0 0 . 0 3 8 5 3 . 933 0 0214 C r o s s p r o d u c t 1 8 5 9 4 . 5 7 4 9 1 6 •0 . 0 0 7 9 1. 616 0 2055 T o t a l R e g r e s s 5 1 9 9 3 7 2 •0 . 1 8 3 3 7 . 496 0 0 0 0 0 R e s i d u a l DF T y p e I SS Mean S q u a r e F - R a t i o P r o b > F L a c k o f F i t 163 . 8 8 7 7 4 9 5 4 4 6 . 3 1 4 0 9 4 , 37 . 8 1 7 0 0014 P u r e E r r o r 4 ' 576 144 . 0 1 7 6 8 0 ' T o t a l E r r o r 167 8 8 8 3 2 5 . 5 3 1 9 . 3 1 2 9 8 2 F a c t o r DF T y p e I SS Mean S q u a r e F - R a t i o P r o b > F SDDO 3 1 3 6 2 9 5 4 5 4 3 2 8 . 541 0 0 0 0 0 PDDO 3 8 6 7 9 1 2 8 9 3 0 5 . 439 0 0014 C a n o n i c a l A n a l y s i s o f t h e R e s p o n s e S u r f a c e ( b a s e d o n s t a n d a r d i z e d d a t a ) C r i t i c a l V a l u e F a c t o r C o d e d U n c o d e d SDDO 0 . 1 4 2 0 6 7 6 1 3 . 3 1 2 2 8 4 PDDO 0 . 2 6 5 4 4 8 5 7 3 . 1 2 4 3 3 1 P r e d i c t e d v a l u e a t s t a t i o n a r y p o i n t 8 . 0 4 1 7 1 1 (min imum) E i g e n v e c t o r s E i g e n v a l u e s ' SDDO PDDO 7 6 . 5 3 8 1 8 7 0 . 9 7 5 1 9 4 0 . 2 2 1 3 5 2 7 . 0 2 5 7 1 6 - 0 . 2 2 1 3 5 2 0 . 9 7 5 1 9 4 166 Table III-11. Technical report of the response surface analysis for variable VOL20 with SNFFD and PNFFD (from SAS RSREG procedure). R e s p o n s e S u r f a c e f o r V a r i a b l e V O L 2 0 w i t h SNFFD a n d PNFFD R e g r e s s i o n DF T y p e I SS R 2 F - R a t i o P r o b > F L i n e a r 2 3 9 2 5 2 5 0 . 3 6 0 9 5 0 . 099 0 0 0 0 0 Q u a d r a t i c 2 2 3 8 7 1 0 . 0 2 1 9 3. 047 0 0502 C r o s s p r o d u c t 1 1 7 0 8 1 0 . 0157 4 . 360 0 0 3 8 3 T o t a l R e g r e s s 5 4 3 3 4 7 8 0 . 3985 22 . 130 0 0 0 0 0 R e s i d u a l DF T y p e I SS Mean S q u a r e F - R a t i o P r o b > F L a c k o f F i t . 163 6 5 2 9 3 8 4005 . 7 5 4 3 3 4 •• 12 . 505 0 0 1 1 6 P u r e E r r o r 4 1 2 8 1 . 3 6 5 2 6 7 320 . 3 4 1 3 1 7 T o t a l E r r o r 167 6 5 4 2 1 9 3917 . 4 8 0 9 6 8 F a c t o r DF T y p e I SS Mean S q u a r e F - R a t i o P r o b > F SNFFD 3 3 6 2 1 6 9 1 2 0 7 2 3 3 0 . 817 0 0 0 0 0 PNFFD 3 8 5 2 7 4 2 8 4 2 5 7 . 256 0 0001 C a n o n i c a l A n a l y s i s o f R e s p o n s e S u r f a c e ( b a s e d o n s t a n d a r d i z e d d a t a ) C r i t i c a l V a l u e F a c t o r C o d e d U n c o d e d S N F F D 0 . 1 0 0 5 6 9 2 4 0 . 6 5 4 6 4 9 PNFFD - 2 . 2 5 2 7 8 0 6 0 . 3 8 1 7 1 5 P r e d i c t e d v a l u e a t s t a t i o n a r y p o i n t 5 2 . 4 4 7 7 4 8 ( s a d d l e ) E i g e n v a l u e s 5 8 . 2 2 9 4 4 4 - 5 . 3 0 3 3 8 7 E i g e n v e c t o r s SNFFD PNFFD 0 . 8 9 7 0 2 1 0 . 4 4 1 9 8 7 - 0 . 4 4 1 9 8 7 0 . 8 9 7 0 2 1 167 Table III-12. Technical report of the response surface analysis for variable VOL20 with SMSP and PMSP (from SAS RSREG procedure). Q u a d r a t i c R e s p o n s e S u r f a c e f o r V a r i a b l e V O L 2 0 w i t h SMSP a n d PMSP R e g r e s s i o n DF T y p e I SS- R- S q u a r e F- R a t i o " P r o b > F L i n e a r 2 5 2 6 4 1 6 ' 0 . 3 9 3 7 69 . 4 4 6 0 0 0 0 0 ' . Q u a d r a t i c 2 . 1 0 9 8 0 0 . 0082 1 . 4 4 8 0 2 3 7 3 - " .. --C r o s s p r o d u c t 1 1 3 6 . 9 9 8 9 7 4 0 . 0001 0 . 0361 0 84 94 • '" • T o t a l R e g r e s s 5 5 3 7 5 3 2 . 0 . 4 0 2 0 28 . 365 0 OO'OO R e s i d u a l • DF T.ype I SS M e a n S q u a r e • F - R a t i o • P r o b > 'F L a c k o f F i t P u r e E r r o r T o t a l E r r o r 211 0 211 7 9 9 7 1 6 ' ' 3 7 9 0 . 1 2 3 6 8 3 i n e s t i m a b l e - i n e s t i m a b l e 0 ' i n e s t i m a b l e i n e s t i m a b l e 7 9 9 7 1 6 3 7 9 0 . 1 2 3 6 8 3 . F a c t o r SMSP PMSP DF 3 3. T y p e I SS M e a n S q u a r e F - R a t i o P r o b > F 5 2 9 2 9 6 - 1 7 6 4 3 2 4 6 . 5 5 0 0 . 0 0 0 0 1 0 3 5 7 34.52 0 . 9 1 1 0 . 4 3 6 6 • C a n o n i c a l A n a l y s i s o f R e s p o n s e S u r f a c e ( b a s e d o n s t a n d a r d i z e d d a t a ) C r i t i c a l V a l u e . F a c t o r C o d e d U n c o d e d SMSP - 4 . 4 2 6 1 6 5 - 4 8 8 . 0 1 7 9 6 6 ' PMSP' " 0 . 0 2 2 8 0 6 . 8 9 2 . 8 3 6 8 4 0 P r e d i c t e d v a l u e a t s t a t i o n a r y p o i n t - 9 8 . 8 8 5 2 0 3 -E i g e n v e c t o r s E i g e n v a l u e s SMSP - PMSP 1 0 . 0 3 8 5 0 1 0 . 9 9 9 0 6 1 - 0 . 0 4 3 3 1 8 - 2 4 . 9 0 5 5 4 5 0 . 0 4 3 3 1 8 0 . 9 9 9 0 6 1 S t a t i o n a r y p o i n t i s a s a d d l e p o i n t . 168 o Q to to. Q to to 2 H H to to > to a z o •f3 S3 . o o o in § oo § P © P © CN 0 s o © \ rr © T± o ^ o ^ o 3- o ^ o "0 © d d P d P d © OS o o OS o oo © p d d , d d T — o CN os m o CN o © CN o SO p d d d d d d °°. o . O o o m >n vo <— © M 2 o OS ^ o P d ° ' d o o ° S ^ C N O 0 0 O t — O ^ S O ^ O " v O N O ^ O M O ^ O t O . p o s p r - o ^ r o ^ o v q © ° d d d d d d d p d p d vo CN OS d .000) VO CN (000 vo vo oo 000) CN (000 •<t r» so o O o d o d o d ° S o «=> £ <=> o °° o d p d p d ' vo Os ^v in oo •*t vt o CO i—i CO o 00 o OV o CO o CN o CN o o. © o CO p d o in o <* © d <d d d i. d d d d © © o °° © i-JN OS-CO o ON © p o ^ © ~H o P d d d o o VO © in © in © VO © © © ON o m © >n © © © 00 © O 00 © o l— © H^ © oo © in © in © © CO VO. © p CO © d d d © • d © • d d d © © © . © v—y •—y o © o o o © fN © » .© © 2! © S o 3 © ^ © t^ © P d p> d OS © o so T 1 O CO vq d p © © o CO vq © ,—s © © © • .—1 Ov © r—i m © © vo © CN © CN © OS © © © p- in © © d © © d d d © © © d d it m © .429) r-CN .065) CO (000 d • © d © © © m o m CN o £• CO © © P © P CO , © Ov © oo t> © —< m " © t^ O <N © P © © © © CN © d © gs © ^ © P O CN © o © P P SO t- ^ CN © ©• ' _ © o ^ © © d ' o ^ B ^ ^ ^ S o S t - m © © P O O P C N T O P C N C - O © ° d P o ' P o P © © © © © co © o. © P o d co o ^ © rt o in © oo o p © d © os © vo © vo © vo © © Ov © © OS © © © © © © z O f3 > to J to C/3 H u 2 Q to to-5 - t o to. fe i/V a o Q e o ib D SH •3 _>v 1 C 00 o c BO .S ' '33 -c lw CD O CJ C o CJ t: o • o CJ > o 42 03 CJ •a <+* o S C3 42 O OS VO Table III-14. Geoclimatic distance ranges of the seed transfer used in the three series of Stika spruce provenance trials in BC, i.e., the experimental span. Geoclimatic distance Range used DiffLAT -12.5-9.0 °N DiffLONG -10.5-11.5 °N DiffJELEV -557 - 627 m DiffMAT -6.5 - 9.0 °C DiffMTCM -15~21 .5°C ' DiffMTWM -4.4-6.3 °C DiffMAP -3200 - 3020 mm DiffMSP -555 - 1155 mm DiffNFFD -122 - 162 day DiffFFP -104- 153 day DiffDD5 -695 - 860 degree day DiffDDO -1032-810 degree day DiffDAY -208-213 hour 170 J l J L o CN O > > CD Q o Q C+-c S CD J3 •*-» O O i- CD -a .ti cd 00 3 cr u co 5 -3 . , oo 3. on h -3 "D CD CD O 00 (OZ10A!ASQ)un 60 cd -a on cd KJ' nU •O « CD *H CD CD tab £ CD l_ -o ,o CN o > cd -a D [in a a CD 00 CD O O 6 o oo cd j - . > "3 3 > cr > CD 1 / 1 oo CD (03nOA!iaa)ui ;—• O C C pLi -X cd cd r-(oznoAiAaa)un cu JS -** fa ° eu CO C A B -2 o a a , co « 1. © - I ^ 0D la •Si CO a a cu CU co co CU CU <U to fl — " fl 9 3 ~ .9 -5 Z i4 co _-• cu "B cu S3 S Cw ,eu ate CU fl o u o cu cu o H •? a -CD a S oo S-, CD > 1-cS CD -3 O o a.. CD g o .Si 00 g ' cd > 60 cd +-» 3 » cr £ CD oo -3 co *J • CD — ed 3 cd CD .2? s 2 S Appendix V. Technical report on the response surface analyses in Chapter Four. Table V - l . Technical report of the response surface analysis for variable NDeviVOL20 (in logarithmic value) with DiffMAT and SMSP (from SAS RSREG procedure). • '• - ' R e s p o n s e S u r f a c e f o r V a r i a b l e ' DAYRES ID w i t h D i f f M A T a n d SMSP R e g r e s s i o n DF T y p e I SS R - S q u a r e F - R a t i o P r o b > F L i n e a r 2 0 . 8 5 7 8 3 4 0 . 0 3 7 3 4 . 5 0 3 0 . 0122 Q u a d r a t i c 2 2 . 3 8 2 5 1 9 0 . 1 0 3 6 12 . 506 0 . 0 0 0 0 C r o s s p r o d u c t 1 0 . 1 4 0 4 9 3 0 . 0 0 6 1 1 . 4 7 5 0 . 2 2 6 0 T o t a l R e g r e s s 5 3 . 3 8 0 8 4 6 0 . 1 4 7 0 7 . 0 9 9 0 . 0 0 0 0 R e s i d u a l DF T y p e I SS Mean S q u a r e F - R a t i o , P r o b > F L a c k o f F i t 200 1 8 . 9 1 5 2 8 3 0 . 0 9 4 5 7 6 . 0 . 8 0 3 0 . 7 1 5 6 P u r e E r r o r 6 0 . 7 0 7 0 2 3 0 . 1 1 7 8 3 7 T o t a l E r r o r 206 1 9 . 6 2 2 3 0 6 0 . 0 9 5 2 5 4 F a c t o r SMSP D i f f M A T DF 3 T y p e I SS 0 . 3 8 9 4 6 7 3 . 2 1 9 8 8 8 Mean S q u a r e 0 . 1 2 9 8 2 2 1. 0 7 3 2 9 6 F - R a t i o I. 363 I I . 268 P r o b > F 0 . 2 5 5 2 0 . 0 0 0 0 C a n o n i c a l A n a l y s i s o f R e s p o n s e S u r f a c e ( b a s e d o n s t a n d a r d i z e d d a t a ) C r i t i c a l V a l u e F a c t o r C o d e d U n c o d e d SMSP - 0 . 3 5 2 7 1 5 4 4 8 . 8 7 5 6 2 9 D i f f M A T 0 . 0 4 9 5 8 6 1 . 1 5 7 0 1 9 P r e d i c t e d v a l u e a t s t a t i o n a r y p o i n t 0 . 0 2 5 9 8 1 E i g e n v e c t o r s E i g e n v a l u e s SMSP D i f f M A T 0 . 1 2 2 5 7 7 . 0 . 9 8 9 8 6 2 0 . 1 4 2 0 3 4 - 0 . 5 2 2 6 3 1 . - 0 . 1 4 2 0 3 4 0 . 9 8 9 8 6 2 S t a t i o n a r y p o i n t i s a s a d d l e p o i n t . 172 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.831.1-0075297/manifest

Comment

Related Items