{"@context":{"@language":"en","Affiliation":"http:\/\/vivoweb.org\/ontology\/core#departmentOrSchool","AggregatedSourceRepository":"http:\/\/www.europeana.eu\/schemas\/edm\/dataProvider","Campus":"https:\/\/open.library.ubc.ca\/terms#degreeCampus","Creator":"http:\/\/purl.org\/dc\/terms\/creator","DateAvailable":"http:\/\/purl.org\/dc\/terms\/issued","DateIssued":"http:\/\/purl.org\/dc\/terms\/issued","Degree":"http:\/\/vivoweb.org\/ontology\/core#relatedDegree","DegreeGrantor":"https:\/\/open.library.ubc.ca\/terms#degreeGrantor","Description":"http:\/\/purl.org\/dc\/terms\/description","DigitalResourceOriginalRecord":"http:\/\/www.europeana.eu\/schemas\/edm\/aggregatedCHO","FullText":"http:\/\/www.w3.org\/2009\/08\/skos-reference\/skos.html#note","Genre":"http:\/\/www.europeana.eu\/schemas\/edm\/hasType","IsShownAt":"http:\/\/www.europeana.eu\/schemas\/edm\/isShownAt","Language":"http:\/\/purl.org\/dc\/terms\/language","Program":"https:\/\/open.library.ubc.ca\/terms#degreeDiscipline","Provider":"http:\/\/www.europeana.eu\/schemas\/edm\/provider","Publisher":"http:\/\/purl.org\/dc\/terms\/publisher","Rights":"http:\/\/purl.org\/dc\/terms\/rights","ScholarlyLevel":"https:\/\/open.library.ubc.ca\/terms#scholarLevel","Title":"http:\/\/purl.org\/dc\/terms\/title","Type":"http:\/\/purl.org\/dc\/terms\/type","URI":"https:\/\/open.library.ubc.ca\/terms#identifierURI","SortDate":"http:\/\/purl.org\/dc\/terms\/date"},"Affiliation":[{"@value":"Forestry, Faculty of","@language":"en"}],"AggregatedSourceRepository":[{"@value":"DSpace","@language":"en"}],"Campus":[{"@value":"UBCV","@language":"en"}],"Creator":[{"@value":"Nault, Jason Ray","@language":"en"}],"DateAvailable":[{"@value":"2011-01-24T21:40:03Z","@language":"en"}],"DateIssued":[{"@value":"1989","@language":"en"}],"Degree":[{"@value":"Doctor of Philosophy - PhD","@language":"en"}],"DegreeGrantor":[{"@value":"University of British Columbia","@language":"en"}],"Description":[{"@value":"A new method is presented for rapid differentiation between coniferous woods commonly found in mixtures in major lumber producing regions of British Columbia. The species mixtures differentiated are the group known as \"spruce\/ pine\/ fir\" (SPF) containing white spruce (Picea glauca Voss), lodgepole pine (Pinus contorta Dougl.) and subalpine fir (Abies lasiocarpa Nutt.); the pair western larch (Larix occidentalis Nutt.) and Douglas-fir (Pseudotsuga menziesii Franco.); and the group of western hemlock (Tsuga heterophylla Sarg.), Sitka spruce (Picea sitchensis Carr.) and amabilis fir (Abies amabilis Dougl.).\r\nThe method entailed measuring the reflectance infrared spectrum of a sample set of small wood pieces at a resolution of two wave numbers, determining which wavelengths were useful for differentiating species through a combination of correlation analyses and principal component analyses and using measurements at these wavelengths to develop species models using discriminant analysis. These models were then used to classify a larger set of samples measured under the same conditions. This approach was used to classify both green wood samples and the same samples after freeze-drying.\r\nFor the SPF group the most effective overall classification model for the dry samples used 30 wavelengths and correctly classified 76% of samples, including 74% of heartwood samples and 89% of sapwood samples. For the green samples, the most effective sort used 10 wavelengths and correctly assigned species to 83% of green samples representing 84% of heartwood samples and 76% of sapwood samples.\r\nClassification was unsuccessful when the same classification parameters were applied to a matched set of extractive-free SPF samples, indicating that the sorting criteria are dependant upon the presence of extractive chemicals, both in heartwood and sapwood.\r\nThe same classification parameters applied to a SPF mixture from eastern Canada (black spruce (Picea mariana B.S.P.), white spruce (Picea qlauca Voss.), jack pine (Pinus banksiana Lamb.), red pine (Pinus resinosa Ait.), white pine (Pinus strobus L.) and balsam fir (Abies balsamea Hill) were less successful than for the western SPF mixture. This suggests that each species may have unique sorting criteria based upon the somewhat different extractive chemical complex present in its wood.\r\nFor the western larch\/ Douglas-fir, the most effective overall classification model used 18 wavelengths and classified 98% of dry samples correctly for both heartwood and sapwood. For green samples, the best sort used 12 wavelengths and correctly assigned species to 91% of green samples, representing 90% of heartwood samples and 91% of sapwood samples.\r\nFor the western hemlock\/ Sitka spruce\/ amabilis fir mixture, the most effective sort for the dry samples correctly classified 83% of the samples, 85% of heartwood samples and 56% of sapwood samples. Classification of the green samples proved difficult, with the best sort only 67% correct and using 15 wavelengths. However, if only western hemlock and Sitka spruce were sorted, the effectiveness rose to 82%.","@language":"en"}],"DigitalResourceOriginalRecord":[{"@value":"https:\/\/circle.library.ubc.ca\/rest\/handle\/2429\/30788?expand=metadata","@language":"en"}],"FullText":[{"@value":"DIFFERENTIATION OF SOME CANADIAN CONIFEROUS WOODS BY COMBINED DIFFUSE AND SPECULAR REFLECTANCE FOURIER TRANSFORM INFRARED SPECTROMETRY By JASON RAY NAULT B.Sc, The University of Winnipeg, 1976 M.Sc, The University of B r i t i s h Columbia, 1986 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES (DEPARTMENT OF FORESTRY) We accept t h i s report as conforming to the required standard. THE UNIVERSITY OF BRITISH COLUMBIA December 1989 \u00a9 Jason Ray Nault, 1989 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 The University of British Columbia Vancouver, Canada DE-6 (2\/88) ABSTRACT A new method i s presented for rapid d i f f e r e n t i a t i o n between coniferous woods commonly found i n mixtures i n major lumber producing regions of B r i t i s h Columbia. The species mixtures d i f f e r e n t i a t e d are the group known as \"spruce\/ pine\/ f i r \" (SPF) containing white spruce (Picea glauca Voss), lodgepole pine (Pinus contorta Dougl.) and subalpine f i r (Abies lasiocarpa Nutt.); the pair western larch (Larix o c c i d e n t a l i s Nutt.) and Douglas-fir (Pseudotsuga menziesii Franco.); and the group of western hemlock (Tsuga heterophylla Sarg.), Sitka spruce (Picea s i t c h e n s i s Carr.) and amabilis f i r (Abies amabilis Dougl.). The method entailed measuring the reflectance infrared spectrum of a sample set of small wood pieces at a re s o l u t i o n of two wave numbers, determining which wavelengths were useful for d i f f e r e n t i a t i n g species through a combination of c o r r e l a t i o n analyses and p r i n c i p a l component analyses and using measurements at these wavelengths to develop species models using discriminant analysis. These models were then used to c l a s s i f y a larger set of samples measured under the same conditions. This approach was used to c l a s s i f y both green wood samples and the same samples afte r freeze-drying. For the SPF group the most e f f e c t i v e o v e r a l l c l a s s i f i c a t i o n model for the dry samples used 3 0 wavelengths and c o r r e c t l y c l a s s i f i e d 76% of samples, including 74% of heartwood samples and 89% of sapwood samples. For the green samples, the most e f f e c t i v e sort used 10 wavelengths and c o r r e c t l y assigned species to 83% of green samples representing 84% of heartwood i i i samples and 76% of sapwood samples. C l a s s i f i c a t i o n was unsuccessful when the same c l a s s i f i c a t i o n parameters were applied to a matched set of e x t r a c t i v e - f r e e SPF samples, i n d i c a t i n g that the sorting c r i t e r i a are dependant upon the presence of extractive chemicals, both i n heartwood and sapwood. The same c l a s s i f i c a t i o n parameters applied to a SPF mixture from eastern Canada (black spruce (Picea mariana B.S.P.), white spruce (Picea qlauca Voss.), jack pine (Pinus banksiana Lamb.), red pine (Pinus resinosa A i t . ) , white pine (Pinus strobus L.) and balsam f i r (Abies balsamea H i l l ) were less successful than for the western SPF mixture. This suggests that each species may have unique sorting c r i t e r i a based upon the somewhat d i f f e r e n t extractive chemical complex present i n i t s wood. For the western larch\/ Douglas-fir, the most e f f e c t i v e o v e r a l l c l a s s i f i c a t i o n model used 18 wavelengths and c l a s s i f i e d 98% of dry samples corre c t l y for both heartwood and sapwood. For green samples, the best sort used 12 wavelengths and c o r r e c t l y assigned species to 91% of green samples, representing 90% of heartwood samples and 91% of sapwood samples. For the western hemlock\/ Sitka spruce\/ amabilis f i r mixture, the most e f f e c t i v e sort for the dry samples c o r r e c t l y c l a s s i f i e d 83% of the samples, 85% of heartwood samples and 56% of sapwood samples. C l a s s i f i c a t i o n of the green samples proved d i f f i c u l t , with the best sort only 67% correct and using 15 wavelengths. However, i f only western hemlock and Sitka spruce were sorted, the effectiveness rose to 82%. TABLE OF CONTENTS ABSTRACT i TABLE OF CONTENTS i LIST OF TABLES LIST OF FIGURES v i i ACKNOWLEDGEMENT X i INTRODUCTION LITERATURE REVIEW 3 METHODS 2 RESULTS AND DISCUSSION SPRUCE\/ PINE\/ FIR 2 DOUGLAS-FIR\/ WESTERN LARCH... A WESTERN HEMLOCK \/ SITKA SPRUCE\/ AMABILIS FIR..A CONCLUSIONS -LITERATURE CITED 5 V LIST OF TABLES TABLE 1- INFRARED ABSORPTION BAND ASSIGNMENTS FOR LIGNIN 57 TABLE 2. INFRARED ABSORPTION BAND ASSIGNMENTS FOR WOOD ...59 TABLE 3. ORIGIN AND SPECIES OF WOOD SAMPLES 61 TABLE 4. PRINCIPAL COMPONENT ANALYSIS OF MEAN CENTERED DRY SPF DATA FOR 7 00 SELECTED WAVELENGTHS FROM 60 SAMPLES 62 TABLE 5. PRINCIPAL COMPONENT VALUES FOR INDIVIDUAL SAMPLES OF SPF 63 TABLE 6. EIGENVECTORS ASSOCIATED WITH INDIVIDUAL WAVELENGTHS FOR SPF 64 TABLE 7. RESULTS FROM SORTING DRY SPF USING 30 WAVELENGTHS... 65 TABLE 8. RESULTS FROM SORTING WET SPF USING 30 WAVELENGTHS... 66 TABLE 9. RESULTS FROM SORTING WET EXTRACTIVE FREE SPF USING 30 WAVELENGTHS 66 TABLE 10. RESULTS FROM SORTING WET EASTERN SPF USING 3 0 WAVELENGTHS 67 TABLE 11. RESULTS FROM SORTING WET SPF USING 20 WAVELENGTHS... 68 TABLE 12. RESULTS FROM SORTING WET SPF USING TEN WAVELENGTHS..69 TABLE 13. RESULTS FROM SORTING WET EXTRACTIVE FREE SPF USING TEN WAVELENGTHS 69 TABLE 14. RESULTS FROM SORTING WET EASTERN SPF USING TEN WAVELENGTHS 70 TABLE 15. RESULTS FROM SORTING WET SPF USING SIX WAVELENGTHS..71 v i TABLE 16. PRINCIPAL COMPONENT ANALYSYIS OF MEAN CENTERED DRY WESTERN LARCH\/DOUGLAS-FIR DATA: A) FIRST 4 60 OF 931 SELECTED WAVELENGTHS FROM 20 SAMPLES 72 B) LAST 471 OF 931 SELECTED WAVELENGTHS FROM 20 SAMPLES 72 TABLE 17. PRINCIPAL COMPONENT ANALYSIS OF MEAN CENTERED DRY WESTERN LARCH\/DOUGLAS-FIR DATA USING 160 WAVELENGTHS SELECTED ON THE BASIS OF EARLIER PRINCIPAL COMPONENT ANALYSES 7 3 TABLE 18. RESULTS FROM SORTING DRY WESTERN LARCH\/DOUGLAS-FIR USING THE FIRST 18 OF 4 0 WAVELENGTHS 73 TABLE 19. RESULTS FROM SORTING DRY WESTERN LARCH\/DOUGLAS-FIR USING THE FIRST 38 OF 40 WAVELENGTHS 74 TABLE 20. RESULTS FROM SORTING WET WESTERN LARCH\/DOUGLAS-FIR USING THE FIRST 18 OF 40 WAVELENGTHS 74 TABLE 21. RESULTS FROM SORTING WET WESTERN LARCH\/DOUGLAS-FIR USING THE FIRST 14 OF 40 WAVELENGTHS 7 5 TABLE 22. RESULTS FROM SORTING WET WESTERN LARCH\/DOUGLAS-FIR USING THE SECOND 14 OF 4 0 WAVELENGTHS 75 TABLE 23. RESULTS FROM SORTING WET WESTERN LARCH\/DOUGLAS-FIR USING THE LAST 12 OF 40 WAVELENGTHS 76 v i i TABLE 24. PRINCIPAL COMPONENT ANALYSIS OF MEAN CENTERED DRY WESTERN HEMLOCK\/SITKA SPRUCE\/AMABILIS FIR DATA: A) FIRST 542 OF 1085 SELECTED WAVELENGTHS FROM 3 0 SAMPLES 77 B) LAST 543 OF 1085 SELECTED WAVELENGTHS FROM 3 0 SAMPLES 77 TABLE 25. PRINCIPAL COMPONENT ANALYSIS OF MEAN CENTERED DRY WESTERN HEMLOCK\/SITKA SPRUCE\/ AMABILIS FIR DATA USING 160 WAVELENGTHS SELECTED ON THE BASIS OF EARLIER PRINCIPAL COMPONENT ANALYSES 78 TABLE 26. RESULTS FROM SORTING DRY WESTERN HEMLOCK\/SITKA SPRUCE\/ AMABILIS FIR USING 19 WAVELENGTHS 78 TABLE 27. RESULTS FROM SORTING DRY WESTERN HEMLOCK\/SITKA SPRUCE\/ AMABILIS FIR USING TEN WAVELENGTHS 79 TABLE 28. RESULTS FROM SORTING WET WESTERN HEMLOCK\/SITKA SPRUCE\/ AMABILIS FIR USING 15 WAVELENGTHS 80 TABLE 29. RESULTS FROM SORTING WET WESTERN HEMLOCK\/SITKA SPRUCE\/ AMABILIS FIR USING TEN WAVELENGTHS 81 TABLE 30. RESULTS FROM SORTING WET WESTERN HEMLOCK\/SITKA SPRUCE USING 15 WAVELENGTHS 82 TABLE 31. RESULTS FROM SORTING WET WESTERN HEMLOCK\/SITKA SPRUCE USING TEN WAVELENGTHS 82 v i i i LIST OF FIGURES FIGURE 1. TYPICAL SINGLE-BEAM REFLECTANCE SPECTRUM FOR WET LODGEPOLE PINE WOOD 83 FIGURE 2. TYPICAL SINGLE-BEAM REFLECTANCE SPECTRUM FOR KBR PELLET 84 FIGURE 3. TYPICAL RATIOED REFLECTANCE SPECTRUM (KBR\/LODGEPOLE PINE HEARTWOOD) 85 FIGURE 4. REGION OF SPECTRUM USED FOR DIFFERENTIATION OF SPECIES (TYPICAL LODGEPOLE PINE SHOWN) 86 FIGURE 5. TYPICAL SPECTRA (n=5) FOR WHITE SPRUCE, LODGEPOLE PINE AND SUBALPINE FIR 87 FIGURE 6. TYPICAL NORMALIZED SPECTRA (n=5) FOR WHITE SPRUCE, LODGEPOLE PINE AND SUBALPINE FIR 88 FIGURE 7. AVERAGE SPECTRA \u2022 1 STANDARD DEVIATION (n=10) FOR WHITE SPRUCE, LODGEPOLE PINE AND SUBALPINE FIR 89 FIGURE 8. TYPICAL MEAN CENTERED REFLECTANCE SPECTRUM FOR WET LODGEPOLE PINE WOOD. 90 FIGURE 9. EIGENVALUES FOR FIRST 12 PRINCIPAL COMPONENTS OF WHITE SPRUCE\/ LODGEPOLE PINE\/ SUBALPINE FIR MIXTURE USING 700 SELECTED WAVELENGTHS 91 FIGURE 10. PRINCIPAL COMPONENT SCORES FOR FIRST AND SECOND PRINCIPAL COMPONENTS OF SAMPLES IN WHITE SPRUCE\/ LODGEPOLE PINE\/ SUBALPINE FIR MIX 92 i x FIGURE 11. PRINCIPAL COMPONENT SCORES FOR FIRST AND THIRD PRINCIPAL COMPONENTS OF SAMPLES IN WHITE SPRUCE\/ LODGEPOLE PINE\/ SUBALPINE FIR MIX 9 3 FIGURE 12. PRINCIPAL COMPONENT SCORES FOR FIRST AND FOURTH PRINCIPAL COMPONENTS OF SAMPLES IN WHITE SPRUCE\/ LODGEPOLE PINE\/ SUBALPINE FIR MIX 94 FIGURE 13. PRINCIPAL COMPONENT SCORES FOR FIRST AND FIFTH PRINCIPAL COMPONENTS OF SAMPLES IN WHITE SPRUCE\/ LODGEPOLE PINE\/ SUBALPINE FIR MIX 95 FIGURE 14. PRINCIPAL COMPONENT SCORES FOR SECOND AND THIRD PRINCIPAL COMPONENTS OF SAMPLES IN WHITE SPRUCE\/ LODGEPOLE PINE\/ SUBALPINE FIR MIX 96 FIGURE 15. PRINCIPAL COMPONENT SCORES FOR SECOND AND FOURTH PRINCIPAL COMPONENTS OF SAMPLES IN WHITE SPRUCE\/ LODGEPOLE PINE\/ SUBALPINE FIR MIX 97 FIGURE 16. PRINCIPAL COMPONENT SCORES FOR SECOND AND FIFTH PRINCIPAL COMPONENTS OF SAMPLES IN WHITE SPRUCE\/ LODGEPOLE PINE\/ SUBALPINE FIR MIX 98 FIGURE 17. PRINCIPAL COMPONENT SCORES FOR THIRD AND FOURTH PRINCIPAL COMPONENTS OF SAMPLES IN WHITE SPRUCE\/ LODGEPOLE PINE\/ SUBALPINE FIR MIX 99 FIGURE 18. PRINCIPAL COMPONENT SCORES FOR THIRD AND FIFTH PRINCIPAL COMPONENTS OF SAMPLES IN WHITE SPRUCE\/ LODGEPOLE PINE\/ SUBALPINE FIR MIX 100 FIGURE 19. PRINCIPAL COMPONENT SCORES FOR FOURTH AND FIFTH PRINCIPAL COMPONENTS OF SAMPLES IN WHITE SPRUCE\/ LODGEPOLE PINE\/ SUBALPINE FIR MIX 101 X FIGURE 20. TYPICAL MEAN CENTERED REFLECTANCE SPECTRUM FOR WET LODGEPOLE LODGEPOLE PINE WOOD SHOWING LOCATION OF 3 0 WAVELENGTHS USED FOR DISCRIMINANT ANALYSIS 102 FIGURE 21. EIGENVALUES FOR FIRST NINE PRINCIPAL COMPONENTS OF WESTERN LARCH\/DOUGLAS-FIR MIXTURE USING FIRST 460 OF 931 SELECTED WAVELENGTHS 103 FIGURE 22. EIGENVALUES FOR FIRST NINE PRINCIPAL COMPONENTS OF WESTERN LARCH\/DOUGLAS-FIR MIXTURE USING LAST 471 OF 931 SELECTED WAVELENGTHS 104 FIGURE 23. EIGENVALUES FOR FIRST NINE PRINCIPAL COMPONENTS OF WESTERN LARCH\/DOUGLAS-FIR MIXTURE USING 160 WAVELENGTHS SELECTED ON BASIS OF PREVIOUS ANALYSES 105 FIGURE 24. TYPICAL MEAN CENTERED REFLECTANCE SPECTRUM FOR WET DOUGLAS-FIR WOOD SHOWING LOCATION OF 40 WAVELENGTHS USED FOR DISCRIMINANT ANALYSIS 106 FIGURE 25. EIGENVALUES FOR FIRST TEN PRINCIPAL COMPONENTS OF WESTERN HEMLOCK\/ SITKA SPRUCE\/ AMABILIS FIR MIXTURE USING FIRST 542 OF 1085 SELECTED WAVELENGTHS 107 FIGURE 26. EIGENVALUES FOR FIRST TEN PRINCIPAL COMPONENTS OF WESTERN HEMLOCK\/SITKA SPRUCE\/AMABILIS FIR MIXTURE USING LAST 543 OF 1085 SELECTED WAVELENGTHS 108 x i FIGURE 27. EIGENVALUES FOR FIRST TEN PRINCIPAL COMPONENTS OF WESTERN HEMLOCK\/SITKA SPRUCE\/AMABILIS FIR MIXTURE USING 171 WAVELENGTHS SELECTED ON BASIS OF PREVIOUS ANALYSES.. 109 FIGURE 28. TYPICAL MEAN CENTERED REFLECTANCE SPECTRUM FOR DRY SITKA SPRUCE WOOD SHOWING LOCATION OF 2 0 WAVELENGTHS USED FOR DISCRIMINANT ANALYSIS 110 ACKNOWLEDGEMENT The author wishes to express his gratitude to a l l those people who assisted i n t h i s study. In p a r t i c u l a r , Dr. J. F. Manville, P a c i f i c Forestry Centre, Dr. E. P. Swan, Forintek Canada Corporation and Dr. J. W. Wilson, U.B.C. Faculty of Forestry have helped greatly. Dr. D. Veltkamp, University of Washington Chemometrics Group, provided useful insights into the data reduction techniques and J. S. Gonzalez, Forintek Canada Corporation, provided the anatomical i d e n t i f i c a t i o n on a l l samples studied. In addition, thanks are extended to Forintek Canada Corporation for funding and support and to the P a c i f i c Forestry Centre for providing access to the necessary equipment for the study. 1 INTRODUCTION The lumber industry i n Canada has expressed a need f o r rapid methods to sort f r e s h l y cut lumber by species. This would enable the industry to take advantage of p a r t i c u l a r properties of each species, either for increased value of end products, or to solve problems associated with manufacture and handling of a p a r t i c u l a r species. For example, a mixture of white spruce (Picea qlauca Voss), lodgepole pine (Pinus contorta Dougl.) and subalpine f i r (Abies lasiocarpa Nutt.) i s common i n B r i t i s h Columbia (B.C.) i n t e r i o r sawmills and i s commonly referred to as spruce\/pine\/fir (SPF). The presence of white spruce i n t h i s mixture can cause problems i n pressure t r e a t i n g of the mix with preservatives. The a b i l i t y to delegate the white spruce i n the sawn mixture to other uses would r e s u l t i n a better treated product and hence more value. Likewise, separation of the subalpine f i r from t h i s same sawn mixture could r e s u l t i n better uniformity i n k i l n drying, since i t i s generally of much higher moisture content than the lodgepole pine or white spruce. In eastern Canada, a si m i l a r mix occurs (eastern SPF) with red spruce (Picea rubens Sarg.), black spruce (Picea mariana B.S.P.), red pine (Pinus resinosa A i t . ) , jack pine (Pinus banksiana Lamb.) and balsam f i r (Abies balsamea H i l l ) . Jack pine i s a preferred species for wood preservers, while separation of red spruce from red pine would be useful due to the premium pr i c e commanded by red pine lumber. In areas where western larch (Larix o c c i d e n t a l i s Nutt.) and 2 Douglas-fir (Pseudotsuqa menziesii Franco.) grow i n association, separation of the larch may be desired to take advantage of i t s s t i f f n e s s r e l a t i v e to Douglas-fir. Larch has been known to cause problems i n gluing for laminated beams, so separation would be an advantage for that application. As well, the two species have d i f f e r e n t average moisture contents, so separation would be an advantage i n k i l n drying. In the mixture of western hemlock (Tsuga heterophylla Sarg.) and amabilis f i r (Abies amabilis Dougl.) often c a l l e d hem-fir, the f i r i s sometimes subject to hidden pocket ro t . This rot i s not apparent when the wood i s green, but when the wood dries (often a f t e r i t i s incorporated into a product) the rot d i s c o l o r s and becomes v i s i b l e . Relegation of the f i r to uses where t h i s would not be v i s i b l e would be an asset. Sitka spruce (Picea s i t c h e n s i s Carr.) also may be harvested i n association with the hem-fir mix. Separation of these three-species to meet customer preferences may also be useful. In a l l species mixtures, customer preferences or higher pr i c e s f o r a p a r t i c u l a r species could be exploited by species separation, r e s u l t i n g i n a higher value product mix. Since such p r i c e advantages are market determined, and thus d i f f i c u l t to predict, a f l e x i b l e system for i d e n t i f i c a t i o n and separation would be of considerable advantage. A possible further application of species s o r t i n g would be i n the pulp and paper industry where i d e n t i f i c a t i o n of logs could be advantageous where certain species (for example Douglas-fir) can cause problems i n the paper or during pulping i t s e l f . 3 Perhaps the most r e l i a b l e method for i d e n t i f i c a t i o n of wood samples from various species i s by examination of anatomical features on both a gross and microscopic scale. By observing the presence or absence of cert a i n anatomical features, as well as t h e i r frequency and siz e , r e l i a b l e i d e n t i f i c a t i o n s can be made. A useful guide to t h i s type of i d e n t i f i c a t i o n for Canadian woods i s the book by S t r e l i s and Kennedy (1967). A trained technician w i l l take about f i v e minutes per sample for i d e n t i f i c a t i o n , i f the species group or geographical source of the sample i s known. Another means of species i d e n t i f i c a t i o n of wood samples i s by examination of t h e i r wood extractive chemicals. Extractives cover a large number of compounds of d i f f e r e n t classes which can be extracted from wood with polar and non-polar solvents, as well as some carbohydrates, carbohydrate derivatives and polymerized phenols extractable with small amounts of a l k a l i . Among the groups represented by extractive chemicals are terpenes, f a t s , f a t t y acids, waxes, alcohols, phenolic compounds, carbohydrates and others. Useful reviews of research i n t h i s area can be found i n H i l l i s (1962), Fengel and Wegener (1984) and H i l l i s (1987). These extractives vary considerably i n composition and concentration from species to species, tree to tree, heartwood to sapwood, growth r i n g to growth rin g and from one type of wood ti s s u e to another. Many of these extractive chemicals have been found to be ubiquitous among softwoods (alpha and beta-pinene, for example), while others are unique to a sing l e family or species (pinosylvin i n Pinaceae, for example). 4 Because these extractive chemicals can be species s p e c i f i c , much research has centered upon i d e n t i f y i n g them and t h e i r b i o l o g i c a l s i g n i f i c a n c e . Species i d e n t i f i c a t i o n has been attempted by means of s p e c i f i c extractives i n several ways. One method involves using reactions of unique extractives with ind i c a t o r chemicals to form coloured complexes, which then can be used to i d e n t i f y species. Examples of t h i s are works by Barton (1973) and M i l l e r et a l . (1985). Unfortunately, these reactions are as a rule too slow for on-line m i l l conditions, often taking up to 30 minutes. They can also give f a l s e r e s u l t s where a sample represents only sapwood, has been stored i n water or has been contaminated with bark or bark extractives. Another approach to wood i d e n t i f i c a t i o n by extractives i s through gas chromatography. Very small amounts of s p e c i f i c e xtractives can be detected by t h i s technique, although sample preparations and analyses are very time consuming. By coupling gas chromatography with other techniques such as i n f r a r e d spectroscopy or mass spectroscopy, t h i s can be a very powerful technique. Examples of works i n t h i s area are Swan (1966), and Manville and Tracey (1989). A s i m i l a r approach i s the use of ion mobility spectrometry to study v o l a t i l e compounds from the wood (Lawrence, 1989). While the speed of these analyses may meet requirements for a m i l l environment, sample introduction may be very d i f f i c u l t . Since the p r i n c i p a l end use for any species s o r t i n g method w i l l be i n a m i l l , several potential c h a r a c t e r i s t i c s of such systems must be considered. F i r s t , sorting of logs must be considered. One pot e n t i a l method would simply be to tag the 5 logs by species as they are f e l l e d and bucked, and sort them into separate p i l e s within the m i l l yard. Each p i l e of logs could then be cut into lumber at a d i f f e r e n t time, making species sor t i n g simple. The l o g i s t i c s of t h i s method, however, have prevented i t s use. F i r s t , a considerably larger sort yard would be required at most m i l l s . Second, since the m i l l s are generally geared to maximum production, and the logs tend to be d i f f e r e n t average sizes for each species, cutting one species at a time would r e s u l t i n a skewed size d i s t r i b u t i o n . This would make maximizing productivity d i f f i c u l t . An a l t e r n a t i v e to t h i s would e n t a i l i d e n t i f y i n g logs but not sor t i n g them, with the ind i v i d u a l pieces cut from each log followed through the m i l l , and kept separate when the lumber i s stacked. Again the l o g i s t i c s of following so many pieces through multiple cutting l i n e s makes t h i s system unattractive. The most a t t r a c t i v e method would be to i d e n t i f y the woods immediately p r i o r to an automatic bin sorter, which would then merely stack the lumber with one additional c r i t e r i o n over those currently i n use (such as size and length). Such a system would have to be able to rapidly i d e n t i f y species, as lumber would be presented at a rate of about one piece per second. Another factor to consider for such a system i s the m i l l environment, which i s harsh to t y p i c a l s c i e n t i f i c instruments (to say the l e a s t ) . A system would have to be robust enough to withstand the dust, temperature variations, moisture and vi b r a t i o n s t y p i c a l i n a sawmill, and s t i l l be s e n s i t i v e and fast enough f o r the i d e n t i f i c a t i o n needed. Non-invasive methods of determining wood species seem to 6 o f f e r the most p r a c t i c a l solutions to t h i s problem. Optical scanning systems are p a r t i c u l a r l y appealing from t h i s standpoint. Placing the actual a n a l y t i c a l instrument some distance from the m i l l environment should be possible f o r these systems with the use of f i b e r optics to carry measurements to the instrument. This has considerable advantage over methods which would involve actual physical contact or sampling of the wood. Spectrometry ( u l t r a v i o l e t , v i s i b l e or infrared) has been a t r a d i t i o n a l choice for non-destructive analysis of compounds. Of these i n f r a r e d (IR) spectrometry i s an e s p e c i a l l y powerful technique i n chemical analyses of compounds, both for i d e n t i f i c a t i o n and s t r u c t u r a l elucidation. The general p r i n c i p l e of IR spectrometry i s that molecules s e l e c t i v e l y absorb d i f f e r e n t wavelengths and amounts of infrared r a d i a t i o n according to which chemical functional groups are present, t h e i r p osition i n the molecule and the concentration of the molecule i n the sample compartment of the spectrometer. By studying what amounts are absorbed at which frequencies, predictions can be made about which functional groups are present, the o v e r a l l structure of the molecule and the concentration of these molecules i n a sample. There are many methods for c o l l e c t i n g IR spectra, ranging from very simple through exotic techniques. One of the simplest techniques i s transmission IR, where the IR l i g h t i s passed through the sample, and absorption i n t e n s i t y at various wavelengths measured. By comparing the i n t e n s i t y of the beam that has passed through a sample with the i n t e n s i t y of a beam 7 that has not, a percent transmittance can be calculated. Several techniques have been used f o r transmission IR. One method involves placing a l i q u i d sample i n an IR transparent c e l l and simply passing the IR beam through the sample. By increasing or decreasing the l i g h t path through the c e l l s u i t a b l e spectra can be measured. A l t e r n a t i v e l y , a s o l i d sample may be dissolved i n a solvent and measured i n t h i s way. An advantage of t h i s technique i s i t s s i m p l i c i t y . A major disadvantage i s that for solute samples, the spectrum of the solvent used may block the spectrum of the sample. One method of measuring s o l i d samples by transmission i s to f i n e l y grind them, mix them with an inorganic halide (KBr and KC1 are common) and press the mixture into a t h i n p e l l e t . The beam i s then passed through t h i s p e l l e t . Since the inorganic halides have high transmittance i n the mid-IR region, they have minimal interference with the sample spectrum. Where the s o l i d sample i s transparent i t s e l f , or can be made t h i n enough to allow transmittance, the spectrum can be measured d i r e c t l y . Another method used to c o l l e c t IR spectra i s to measure the i n t e n s i t y of an IR beam re f l e c t e d off the sample. This can be done i n a number of ways. One method i s attenuated t o t a l r e f l e c t i o n (ATR). In ATR, a single r e f l e c t i o n prism and varying angles of incident l i g h t are used to produce the desired spectrum by r e f l e c t i n g the l i g h t off the surface of the sample. This method can be greatly improved by using a m u l t i - r e f l e c t i o n prism. By placing the sample i n close contact with the prism, the energy that escapes temporarily from the prism i s 8 s e l e c t i v e l y absorbed by the sample. By allowing multiple r e f l e c t i o n s , a much larger absorbance i s allowed, r e s u l t i n g i n better spectra. This technique i s known as multiple i n t e r n a l r e f l e c t i o n (MIR). An advantage of MIR i s that i t allows spectra to be measured on o p t i c a l l y opaque samples. I t also allows spectra to be measured on very thick samples. A disadvantage i s that the method i s dependant upon the area and e f f i c i e n c y of contact between the sample and the prism. This makes e f f e c t i v e measurement d i f f i c u l t on uneven surfaces, and complicates sample preparations. Another method of measuring r e f l e c t i o n IR spectra i s through d i f f u s e reflectance. In di f f u s e reflectance the sample i s i r r a d i a t e d with IR, which i s then d i f f u s e l y scattered i n a l l d i r e c t i o n s . Part of t h i s scattered l i g h t i s c o l l e c t e d and directed to the detector by a special o p t i c a l arrangement. The r a t i o of t h i s spectrum to a background spectrum where the d i f f u s e r e f l e c t i o n i s measured from a reference object (such as a KBr p e l l e t ) gives the dif f u s e reflectance spectrum (DRS). The major advantage of DRS techniques i s that they allow IR measurements on samples i n an unaltered state. The major disadvantage i s that the DRS signal i s very weak, thus d i f f i c u l t to measure. This disadvantage has been larg e l y resolved by the use of Fourier transform infrared (FTIR) spectroscopy. In t r a d i t i o n a l IR instruments the IR spectrum i s determined by sequential measurements at each wavelength. This means that the q u a l i t y of the recorded spectrum i s determined by the speed at which the wavelengths are changed and the range of 9 wavelengths measured at any one instant (determined by a s l i t width on the incident or transmitted side of the sample). The small s l i t width required for high resolution, combined with the sequential scanning of wavelengths mean that a high r e s o l u t i o n spectrum takes a long time to measure. In a FTIR instrument, a l l wavelengths are scanned at once (multiplex advantage), and no s l i t i s required so the IR source strength i s not diminished (throughput advantage). The mechanics and mathematics of t h i s are beyond the scope of t h i s t h e s i s , but can be found i n any recent IR spectroscopy text. In short, the recorded spectrum i s a time domain spectrum which i s mathematically transformed (Fourier transformation) into the frequency domain usually used in IR spectrometry. Through the multiplex and throughput advantages the instrument has a large improvement i n the signal to noise r a t i o , making possible higher r e s o l u t i o n and measurement of smaller signals. Analyses by IR techniques have been mainly q u a l i t a t i v e i n nature and usually performed upon p u r i f i e d substances. This r e s t r i c t i o n has been due to v a r i a b i l i t y between spectra and the r e l a t i v e l y low resolution available u n t i l recently. Recent advances i n equipment technology, notably computers, have combined to make FTIR spectrometers extremely fast, accurate and r e l a t i v e l y inexpensive. The modern FTIR instrument can acquire spectra with far greater precision than was possible with t r a d i t i o n a l grating instruments and i n much less time. The increased s e n s i t i v i t y available from modern FTIR instruments and the use of computers for data analyses have expanded the use of IR to new areas. These new areas include increased use of FTIR 10 spectroscopy for quantitative analyses and the study of molecules i n mixtures rather than i n pure states. The increased throughput and signal to noise r a t i o of FTIR instruments have also made Diffuse Reflectance FTIR (DRIFT) and other forms of reflectance spectroscopy r e l a t i v e l y easy to perform, opening whole new f i e l d s of study. Applying IR spectroscopy to wood, we can make several general observations. Since wood i s a mixture of many d i f f e r e n t compounds, we expect the IR spectrum to be a complex mixture of contributions from these many in d i v i d u a l components. The compounds with the largest concentrations ( c e l l u l o s e , l i g n i n , hemicellulose) are expected to dominate. Superimposed upon these w i l l be the contribution from water (highly variable, depending upon wood moisture content) and f i n a l l y minor contributions from a l l the various extractive compounds. I t i s hypothesized that by measuring the reflectance spectrum of wood using FTIR the contribution of the extractive compounds can be detected. Differences i n the wood extractive chemical composition between species can thus be detected and can form the basis for a rapid means of species i d e n t i f i c a t i o n . However, given the low concentrations of species s p e c i f i c e x t r a c tive chemicals, and t h e i r i n c l u s i o n within the strongly absorbing matrix of wood tissue, very small differences are expected. Detection of these differences would thus require high r e s o l u t i o n spectra, and a sophisticated algorithm for i n t e r p r e t i n g the spectra. Prediction of which areas of the spectrum w i l l be useful w i l l be limited because we do not know i n advance which extractive compounds w i l l be detected within 11 the wood, nor which s p e c i f i c extractive chemicals w i l l be of most use for species i d e n t i f i c a t i o n . Thus, we w i l l have to analyze i n minute d e t a i l a large portion of the wood IR spectrum, which presents a formidable task i n data analyses. Regarding methodology for interpreting spectra, the a v a i l a b i l i t y of modern computer-controlled chemical instrumentation has resulted i n the generation of massive amounts of data. I t can be very d i f f i c u l t to obtain useful information from the tremendous volume of data generated by these instruments. This s i t u a t i o n has given r i s e to a new area of s t a t i s t i c a l analysis termed \"chemometrics\". Chemometrics i s e s s e n t i a l l y a set of methods for obtaining chemical information from data through processes such as c a l i b r a t i o n , resolution, pattern recognition, etc. Applications of chemometrics encompass the entire f i e l d of chemistry. Examples of chemometrics use range from discrimination of\"bee species or hybrids (Lavine et a l . , 1988) to determining the baking q u a l i t y of wheat f l o u r s i n bread (Devaux et a l . , 1987). One of the most powerful tools for chemometrics i s the procedure of factor analysis. Factor analysis attempts to explain c o r r e l a t i o n s between a large set of variables through a small number of underlying factors, with the purpose of reducing the dimensionality of the data while preserving the majority of the o r i g i n a l information. The basic p r i n c i p l e s of factor analysis are explained well i n several texts (Malinowski and Howery, 1980, Sharaf et a l . , 1986, Massart et a l . . 1988) and are beyond the scope of t h i s thesis. The f i r s t step i n factor analysis i s to organize the data 12 into a matrix of samples and variables. The variables may or may not be weighted by d i f f e r e n t factors, depending upon the ultimate purpose of the factor analysis. Various forms of weighting are possible, including normalizing the data, mean centering, variance weighting and Fisher weighting (Sharaf et a l . . 1986). Once the data i s thus organized, the f i r s t step i n factor analysis i s p r i n c i p a l factor analysis (PFA), also known as p r i n c i p a l component analysis (PCA) or eigenvector analysis. If we s t a r t with a sample matrix of 'm' samples by 'n' variables, PCA w i l l express the o r i g i n a l data matrix i n terms of a l i n e a r combination of orthogonal (independent) vectors known as p r i n c i p a l components (PC) or eigenvectors. Each PC explains part of the v a r i a t i o n within the data, with the f i r s t PC explaining the largest part, followed by the second PC, down to * n' PCs. The minimum number of PCs needed to account for v a r i a t i o n s i n the data i s known as the necessary number of independent vectors, 'k', to describe the dataset. Where 'k ^ ' n 1 , the dimensionality of the data i s reduced. In pr a c t i c e , *k' PCs w i l l probably not account for 100% of the v a r i a t i o n within the dataset, due to experimental noise, but 'k' can be chosen where i t explains most of the v a r i a t i o n . Within each PC, each of our \u2022 n 1 variables i s assigned a loading, depending upon i t s importance within that PC. Small p o s i t i v e or negative loadings indicate a weak r e l a t i o n s h i p between the variable and the vector, while large p o s i t i v e or negative loadings indicate a strong r e l a t i o n s h i p . As well, each of our 'm* samples i s given a score for each 13 PC, r e l a t e d to i t s weight within that PC. I f we examine the scores f o r the in d i v i d u a l samples, we can get an idea of the re l a t i o n s h i p s between the samples. In our case, where we are attempting to assign species to our samples, we hope that the samples w i l l thus be grouped together by species. Thus, PCA gives us the number of p r i n c i p a l factors within our dataset, the r e l a t i v e importance of the variables within each PC, and the relationships within the sample set. In many cases t h i s i s a l l the information required for a chemical problem, but i n other cases we want to gain p r a c t i c a l i n sight into the nature of the PCs. To do t h i s the abstract solution furnished by PCA must be transformed into a more p r a c t i c a l s o l u t i o n . Several methods of obtaining more p r a c t i c a l information are av a i l a b l e . One method i s factor r o t a t i o n (Malinowski and Howery, 1980, Sharaf et a l . . 1986, Massart et a l . , 1988) to further reduce the dimensionality of the data and emphasize cl u s t e r i n g s . As f o r methods of c l a s s i f y i n g samples, several choices are av a i l a b l e . One method i s s t r i c t l y to use factor scores as determined i n PFA (or a rotation of PFA) coupled with some measure of f i t to assign a sample to a group. A sophisticated example of t h i s technique i s found i n simple modeling of class analogy (SIMCA) where a PC model i s f i t to each category, and a confidence envelope constructed around the model to contain the data points. This provides a set of parameters that characterize each category and forms a basis for assigning unknown samples to a category. It also provides a method for 14 detection of o u t l i e r s , and can provide a basis to predict external properties (Sharaf et a l . , 1986, Massart et a l . . 1988). A common system for c l a s s i f y i n g objects i s the family of methods known as discriminant analysis, covered i n many texts (Hope, 1968, Klecka, 1980, Johnson and Wichern, 1982, Massart et a l . . 1988). In discriminant analysis, a set of variables i s measured for each object, and t h i s set of variables i s used to derive a discriminant function. The discriminant function can take many forms, the simplest of which i s a l i n e a r combination of the variab l e s . The basic requirements of discriminant analysis are: 1. Two or more groups exi s t which we presume to be mutually exclusive; 2. These groups d i f f e r with respect to several v a r i a b l e s simultaneously; 3. The number of data cases should exceed -the number of variables by more than two; 4. No variable may be a lin e a r combination df other discriminating variables; and 5. P e r f e c t l y correlated variables cannot be used simultaneously. Other r e s t r i c t i o n s may apply, depending upon which discriminant technique i s used. Once the discriminant function i s developed, i t may be used i n two ways. The f i r s t use i s i n studying r e l a t i o n s h i p s within and between groups. The second use i s to c l a s s i f y unknown samples into one of the groups. Note that the assumption i s made that the unknown belongs within one of the groups 15 discriminated. To summarize, the hypothesis to be tested i n t h i s work i s that reflectance FTIR spectra of wood samples may be used to separate woods of d i f f e r e n t tree species. Differences may be based upon differences i n the wood extractive chemical composition between species. Chemometrics techniques w i l l be used to a s s i s t i n the analyses of the data. 16 LITERATURE REVIEW IR spectroscopy has been applied to wood and pulp chemistry studies f o r a considerable time. The early studies u t i l i z e d more p r i m i t i v e spectrometers than are avai l a b l e now, severely l i m i t i n g the q u a l i t y of the spectra and hence the u t i l i t y of studies i n t h i s f i e l d . More recent works have taken advantage of improvements i n IR spectroscopy made possible by FTIR, where increased si g n a l to noise r a t i o and signal strength have made improved resolution and faster scanning possible. The majority of the work published on IR spectroscopy of wood or wood components has focused upon the pulp and paper industry, where quantitative determination of l i g n i n within the pulp and monitoring progress of the pulping process are the main uses. Very l i t t l e work has been published upon wood extractive chemicals examined i n s i t u . probably due to the extreme d i f f i c u l t i e s i n analyzing small amounts of extractives within the complex chemical matrix of wood s t r u c t u r a l components. Kolboe and E l l e f s e n (1962) attempted to see i f such l i g n i n preparations as Braun's \"native l i g n i n \" and m i l l e d wood l i g n i n represented unaltered l i g n i n as i t would be found i n whole wood. They u t i l i z e d Norway spruce (Picea abies Karst.) preparations of h o l o c e l l u l o s e and ether extracted ground wood pressed into KBr p e l l e t s and recorded t h e i r transmission IR spectra. By subtracting the holocellulose spectrum from the wood spectrum they obtained a spectrum of \"protolignin\", which was assumed to represent l i g n i n as i t exists within the wood structure. This \" p r o t o l i g n i n \" spectrum was then compared to spectra of Braun's 17 \"native\" l i g n i n and milled wood l i g n i n . The three spectra were found to be q u a l i t a t i v e l y s i m i l a r . They u t i l i z e d the l i g n i n absorbance at 1515 wavenumber (cm - 1) (based upon the benzene r i n g content of l i g n i n ) to estimate the l i g n i n content of the spruce wood, and attained a value of 28 to 29%, which was i n agreement with accepted values. They also attempted to assign functional groups within the l i g n i n to spectral features (summarized i n Table 1). M i c h e l l et a l . (1965) placed wood cross sections 2Oil thick and t h i n paper sheets prepared from pulps between layers of KC1 and pressed them into p e l l e t s . Transmission IR spectra were measured from 5000 to 4 50 cm - 1. They found that spectra of t h i n paper sheets were comparable i n qu a l i t y to those of t h i n sections of wood, and superior to p e l l e t i z e d m i l l e d wood spectra. Sheets were prepared from pulp samples taken at various stages of the pulping process and from d i f f e r e n t pulping processes. Spectra were then obtained from these sheets and subtracted from the spectrum of the whole wood. They found that the progressive removal of some substances gave r i s e to p o s i t i v e differences i n the spectra (more transmission) and that the r e s u l t i n g increase i n the concentrations of other substances l e f t within the sheets gave r i s e to negative differences (less transmission) when the same weight of material was scanned. This was overcome by reducing the weight of the sheets by the same proportion as losses i n o v e r a l l weight during pulping. They then used the progressive changes i n the difference spectra to make inferences about the chemical changes taking place while 18 pulping. They also assigned functional groups to spectral features of l i g n i n (Table 1) and wood (Table 2). No mention was made about whether or not the wood was extr a c t i v e free, or about contributions of extractive chemicals to the whole wood spectrum. Marton and Sparks (1967) used multiple i n t e r n a l r e f l e c t i o n (MIR) IR spectroscopy to measure spectra of various pulps, l i g n i n preparations and l o b l o l l y pine (Pinus taeda L.) wood meal. They presented representative spectra of l i g n i n preparations, wood and pulps and assigned functional groups to spec t r a l features found i n those spectra. Their assignments for l i g n i n and wood are summarized i n Table 1 and Table 2, resp e c t i v e l y . They demonstrated that l i g n i n could be recognized i n a carbohydrate (holocellulose) matrix through c h a r a c t e r i s t i c absorbances at 1595, 1510 and 1265 cm - 1, although the band near 1600 cm 1 could be obscured by water contained i n the sample. Certain carbohydrate absorbances (1310 and 893 cm - 1) were observed to be ins e n s i t i v e to l i g n i n content. By using the l i g n i n absorbance at 1510 cm - 1, and the i n s e n s i t i v e carbohydrate absorbances as in t e r n a l standards they were able to estimate l i g n i n content i n the handsheets. They found good c o r r e l a t i o n between MIR l i g n i n and Kappa No. and acetyl bromide l i g n i n values when l i g n i n content exceeded 4% by weight of the samples. Again, no mention was made about whether or not the wood was extractive free, or about contributions of extractive chemicals to the whole wood spectrum. Sarkanen et a l . (1967a) measured transmission IR spectra of 19 model l i g n i n molecules i n carbon te t r a c h l o r i d e s o l u t i o n . They found guaiacyl-based compounds could be distinguished e a s i l y from syringyl-based compounds even i n complex models. They also attempted band assignments of the major absorption bands of both kinds of l i g n i n (summarized i n Table 1). Sarkanen et a l . (1967b) studied transmission IR spectra of KBr p e l l e t s prepared with milled l i g n i n s from Douglas-fir, western redcedar (Thuja p l i c a t a Don), western larch, ponderosa pine (Pinus ponderosa Laws.), western hemlock and Himalayan pine ( G r i f f i t h i excelsa) (no authority given) and found l i t t l e d i f f e r e n c e between the \"softwood\" l i g n i n s studied. Some small differences were attributed to v a n i l l a t e or acetate units c h a r a c t e r i s t i c of some species. They concluded that the softwood l i g n i n s studied were nearly i d e n t i c a l . Sarkanen et a l . (1967c) then used the same technique to measure IR spectra of milled l i g n i n s from several \"hardwood\" species (eastern cottonwood (Populus deltoides B a r t r . ) , bigleaf maple (Acer macrophvllum Pursh), cascara buckthorn (Rhamnus purshiana DC.) and madrone (Arbutus menziesii Pursh). They found more pronounced v a r i a t i o n than for \"softwoods\" and that major maxima correlated to methoxyl:carbon r a t i o s . Chow (1972) studied transmission IR spectra from 4000 to 700 cm 1 of t h i n s l i c e s of heartwood and sapwood from four coniferous species sandwiched between two s i l v e r chloride plates. The four species studied were Douglas-fir, grand f i r (Abies qrandis L i n d l . ) , western hemlock and white spruce. He found that the spectra varied from sapwood to heartwood and from earlywood to latewood. The spectral differences i n a l l cases 20 were i n r e l a t i v e magnitudes of absorbances, never i n presence or absence of p a r t i c u l a r bands. By comparing the spectra of extracted samples and unextracted samples (noting that since the wood samples had to be water saturated to microtome th i n sample s l i c e s , some loss of water soluble extractives was unavoidable) he showed that the spectra were influenced by the presence of extractable materials, e s p e c i a l l y i n the region from 1300 to 1800 cm - 1. Again only r e l a t i v e i n t e n s i t i e s of the bands changed, although i n some cases the changes were dramatic. He assigned functional groups to spectral features of the woods (Table 2). A gradual decrease of the band at 173 0 cm - 1 was observed from stem p i t h to periphery and he attributed t h i s to autohydrolysis of the woody ti s s u e . He suggested that species i d e n t i f i c a t i o n by IR spectra would be hampered by t h i s gradual change i n - the heartwood spectra, and the fact that heartwood extractive content varies across the heartwood, and up the tree. He believed that species could be determined from the IR spectra based on s t r u c t u r a l elements of the wood ( l i g n i n and hemicelluloses) but that t h i s i d e n t i f i c a t i o n would be limited to sapwood tissu e s , which showed the l e a s t \"interference\" from extractive chemicals. However, no attempt was made to v e r i f y t h i s theory. Saad et a l \u2022 (1980) studied l i g n i n s recovered from soda black l i q u o r s of bagasse by measuring transmission IR spectra of the compounds i n KBr p e l l e t s . They attempted band assignments (summarized i n Table 1). They also attempted to determine l i g n i n content of pulps from t h e i r spectra with l i m i t e d success, 21 e s p e c i a l l y at low (8-12%) l i g n i n contents. Obst (1982) studied various c e l l types from white oak (Quercus alba L.) chemimechanical, sulphite semichemical and k r a f t pulps, using transmission IR spectra of preparations i n KBr p e l l e t s . M i l l e d wood l i g n i n from l o b l o l l y pine was found to be t y p i c a l of softwood l i g n i n s , with a strong guaiacyl absorbance at 12 70 cm - 1 and no syringyl absorbance at 13 3 0 cm\"1. I t also gave two weak bands at 855 and 815 cm\"1. The oak c e l l s a l l demonstrated t y p i c a l hardwood l i g n i n absorbances, with the t y p i c a l absorbance of syringyl l i g n i n at 1330 cm - 1. The weak guaiacyl bands at 855 and 815 cm - 1 were replaced by a singl e weak band at 835 cm - 1. He found t y p i c a l s y r i n g y l absorbances i n a l l spectra, with t y p i c a l syringyl\/guaiacyl r a t i o s i n a l l c e l l types and no s i g n i f i c a n t differences between spectra of the c e l l types studied. He concluded that the l i g n i n i n a l l the c e l l types studied was the same. This was a contradiction of e a r l i e r work by Goring (1970) who found d i f f e r i n g r a t i o s of s y r i n g y l and guiacyl units i n l i g n i n from f i b e r s , vessels, and ray c e l l s i n white b i r c h (Betula papyrifera Marsh.) using u l t r a v i o l e t spectroscopy. Schultz et a l . (1985) digested sweetgum (Liquidambar s t y r a c i f l u a L.) and white oak wood chips to varying degrees, ground and mixed them with KC1 and determined DRIFT spectra on the KC1 p e l l e t s . They selected absorbances based upon known spectra of c e l l u l o s e and l i g n i n of mixed hardwoods, then used a stepwise regression to y i e l d equations for D-glucose, L-xylose and l i g n i n . The r e s u l t i n g equations had f i v e variables for D-glucose, three for L-xylose and f i v e f o r l i g n i n , with 22 R-squared values of 0.921, 0.973 and 0.949 res p e c t i v e l y . No differences were found between sweetgum and white oak. Faix (1986) studied 44 l i g n i n polymer models by FTIR using transmission spectra of the l i g n i n model compounds i n KBr p e l l e t s . He found good quantitative r e l a t i o n s h i p s between absorbances i n the spectra obtained and s t r u c t u r a l properties of the l i g n i n s . Gurnagul et a l . (1986) used FTIR photoacoustic spectroscopy (PAS) to obtain spectra on bleached k r a f t papers prepared from a singl e sample of pulp subjected to varying degrees of beating. They observed i n t e n s i t y changes which they a t t r i b u t e d to surface area differences of the papers, which were rel a t e d to the degree of beating. They concluded that quantitative FTIR of paper must take sheet structure into account. Schultz and Glasser (1986) studied l i g n i n s from various sources using DRIFT with KBr p e l l e t s . Peak assignments for l i g n i n were made based upon previous works (summarized i n Table 1). They then established empirical quantitative r e l a t i o n s h i p s between FTIR spectral information and some s t r u c t u r a l features i n l i g n i n s . These included methoxyl content, aromatic H content, phenolic hydroxyl content, guaiacyl\/syringyl r a t i o s , \"hydrolysis r a t i o \" (a measure of the degree of a l k y l - a r y l i n t e r - u n i t ether linkages) and \"condensation r a t i o \" (a measure of the content of carbon-carbon i n t e r - u n i t bonds). DRIFT spectra were found superior to transmission spectra of KBr p e l l e t s i n some aspects. Berben et a l . (1987) developed a method fo r estimating l i g n i n i n unbleached pulp using DRIFT. They found a l i n e a r r e l a t i o n s h i p between sample absorbance i n the 1510 cm band and Kappa No. or Klason l i g n i n content ( e s s e n t i a l l y the same method as was used by Kolboe and E l l e f s e n (1962) using transmission IR). The r e l a t i o n s h i p held for several types of pulp over a wide range (1-19%) of l i g n i n contents and for both \"hardwood\" and \"softwood\" pulps. Grandmaison et a l . (1987) performed DRIFT analyses on chars produced by s u p e r c r i t i c a l gas extraction of trembling aspen wood. The chars were ground and DRIFT spectra recorded for the powders. General peak assignments were made for the wood samples (summarized i n Table 2). No mention was made about whether or not the wood was extractive free, or about contributions of extractive chemicals to the whole wood spectrum. By comparing progressive changes i n the spectra they found that DRIFT and thermogravimetry gave complementary information on loss of components at increasing - temperatures of extraction. Abbot et a l . (1988) u t i l i z e d FTIR spectra on samples of klason l i g n i n , hemicellulose and c e l l u l o s e components prepared from extractive free wheat straw, kenaf stalk s , oak and pine. They s e l e c t i v e l y s o l u b i l i z e d portions of each component and studied the spectral differences of the remaining material to approximate i n s i t u spectra for these components. They found that the difference spectra attained for l i g n i n agreed well with published spectra of mil l e d wood l i g n i n , and that the spectral subtraction method generally gave good approximations of i n s i t u spectra for the compounds studied. Durig et a l . (1988) used transmission FTIR of peat samples 24 pressed i n KBr p e l l e t s to study decomposition of peat i n bogs. They found general differences between peat types, but much v a r i a b i l i t y within types due to varying botanical o r i g i n s of the material and i t s degree of decay. They also found that c e l l u l o s e content decreased as decay progressed. Faix (1988) presented a review of progress to date on p r a c t i c a l applications of FTIR i n wood science. He described progress on determination of l i g n i n content i n woody materials, Kappa No. i n pulp, degree of substitution i n c e l l u l o s e d e r i v a t i v e s , and monitoring d e l i g n i f i c a t i o n by FTIR of pulping l i q u o r s . Kuo et a l . (1988) found that FTIR-PAS gave spectra comparable to transmission spectra of t h i n r a d i a l sections of wood from ponderosa pine, redwood (Sequoia sempervirens Endl.) and red oak (Ouercus rubra L.). Some differences were noted and these were attr i b u t e d to differences i n l i g n i n composition among the three species and the higher xylan content i n red oak. They compared spectra obtained from oblique and transverse sections of ponderosa pine, showing some differences i n the spectra. They at t r i b u t e d these differences to m i c r o f i b r i l o r i e n t a t i o n i n the samples. They also studied decayed eastern cottonwood samples and found that progressive loss of l i g n i n to decay could be detected i n the spectra. They found that very small samples could be analyzed and concluded that PAS could overcome l i m i t a t i o n s of DRIFT, KBr p e l l e t transmission and ATR methods. Michel l (1988a) compared ATR FTIR and transmission FTIR on t h i n sections of radiata pine (Pinus radiata D. Don) wood treated with 30% hydrogen peroxide solution to study surface 25 changes. He found that ATR gave comparable r e s u l t s to absorbance spectra of the t h i n samples and that the samples were more robust, making sample preparation and handling easier. M i c h e l l (1988b) used FTIR difference spectra of f i b e r s e i t h e r pressed i n KBr p e l l e t s or into t h i n sheets to study chemical changes i n eucalyptus (Eucalyptus reqnans F. Muell.) wood during pulping. He found evidence that d i f f e r e n t materials dis s o l v e i n progressive pulping steps. Wood (1988) analyzed charcoal prepared from seven hardwoods and one softwood using DRIFT. He found minor q u a l i t a t i v e differences between the spectra of the d i f f e r e n t charcoals. Owen and Thomas (1989) u t i l i z e d DRIFT spectra to d i f f e r e n t i a t e \"hardwoods\" from \"softwoods\" i n a study involving 24 species. A l l wood spectra exhibited the same basic features. These features included the broad OH stretch around 3400 cm - 1, prominent CH stretch absorption at about 2900 cirT1 and a strong, broad absorption envelope from 1750 to 1000 cm - 1 with many sharp and d i s c r e t e absorptions. One absorption at 174 3 cm\"1 i n balsa (Ochroma pyrimmidale Sw.) and 1737 cm - 1 i n redwood was assigned to stretching of carbonyl groups which occur abundantly within a l l the polymer constituents of wood. The majority of these carbonyl groups were assumed to be i n the branched hemicellulose polymers, with other polymers and extractive compounds contributing. This absorption was deemed to be i n d i c a t i v e of the concentration of the c e l l u l o s i c component of the wood. A l l softwoods were found to have t h i s band at lower than 1739 cm\"1, and most hardwoods had t h i s band at higher than 1738 cm\"1. This s h i f t was thought 26 to be caused by the r e l a t i v e l y higher concentration of l i g n i n i n softwoods. Absorptions at 1600 and 1510 cm - 1 were deemed to be caused by v i b r a t i o n s of the l i g n i n , associated with the aromatic r i n g deformation modes. The absorption around 1510 cm - 1 occurred above 1509 cm - 1 for a l l softwoods, and below 1509 cm - 1 f o r most hardwoods. This differerence was attributed to the d i f f e r e n t l i g n i n components i n hardwood (syringyl and guaiacyl) as compared to softwoods (guaiacyl only). They used three absorption frequencies to characterize the woods by l i g n i n content and components, h o l o c e l l u l o s e content and r a t i o of l i g n i n to holocellulose. Their c l a s s i f i c a t i o n method was generally successful, although problems were encountered when l i g n i n content was higher than usual i n a piece of wood, or extractive content was high. They also found that t h e i r DRIFT spectra obtained from small pieces of whole wood were superior to transmission spectra of t h i n s l i c e s and transmission spectra from KBr p e l l e t s using ground wood. As regards other techniques for species i d e n t i f i c a t i o n of wood, Swan (1966) presented a tentative method for d i f f e r e n t i a t i o n of Canadian pines by g a s - l i q u i d chromatography (GLC) of steam v o l a t i l e extractives. D i f f e r e n t i a t i o n of western hemlock and amabilis f i r was attempted by two dimensional paper chromatography of phenolic extractives, GLC of steam v o l a t i l e extractives and t h i n layer chromatography (TLC) of l i p o p h i l i c e x tractives. Of the three methods, TLC was the most successful, with 136 of 137 samples i d e n t i f i e d c o r r e c t l y . 27 Barton (1973) reviewed various chemical t e s t s u s e f u l for separation of Canadian woods. Included were a t e s t f o r : 1. D i f f e r e n t i a t i o n of Douglas-fir from western hemlock, true f i r s , spruces and western redcedar by detection of dihydroquercetin; 2. D i f f e r e n t i a t i o n of pine and spruce wood through non-v o l a t i l e phenolic compounds present i n pine heartwood; 3. D i f f e r e n t i a t i o n of amabilis f i r from subalpine f i r based on sesqiterpenes present in subalpine f i r ; 4. D i f f e r e n t i a t i o n of sapwood and heartwood of western hemlock based on leucoanthocyanidins present i n the sapwood; and 5. Iron contamination on wood. Lawrence (1989) demonstrated that ion mobility spectrometry (IMS) was a useful method for d i f f e r e n t i a t i n g between several species of wood and for detecting s p e c i f i c chemicals i n wood. He succe s s f u l l y separated several samples of black spruce, jack pine and balsam f i r by subjecting sawdust or small s l i v e r s of sample to a c a r r i e r gas, heating them and recording the IMS of the r e s u l t i n g e f f l u e n t . Bacteria infested red oak and the presence of pentachlorophenol i n western hemlock wood chips could also be detected by t h i s method. To summarize the l i t e r a t u r e , only a few researchers have studied IR spectra of whole wood. Kolboe and E l l e f s e n (1962) studied only extracted wood s l i c e s and used transmission IR. Mic h e l l et a l . (1965) used transmission IR on wood s l i c e s , but made no mention of whether the wood was extracted or not. Marton and Sparks (1967) used transmission spectra of wood meal i n KBr, and also make no mention of extractives. Chow (1972) i n 28 h i s exhaustive study using transmission through t h i n s l i c e s of wood made passing references to the e f f e c t of wood extractives. However, he considered that they would i n t e r f e r e with any species i d e n t i f i c a t i o n scheme. Kuo et a l . (1988) used FTIR PAS on t h i n wood sections, and did not mention wood extractives or t h e i r possible contributions to the spectra. Owen and Thomas (1989) used DRIFT to c l a s s i f y small wood pieces into hardwoods and softwoods. They mention that t h e i r c l a s s i f i c a t i o n was confused by pieces with higher than normal amounts of extractive chemicals. No c i t e d author has made an attempt to more thoroughly investigate the e f f e c t s of extractive chemicals i n s i t u on the IR spectrum of wood. 29 METHODS Samples of freshly cut green lumber were obtained for lodgepole pine, Sitka spruce, white or Engleman spruce, western larch, Douglas-fir, western hemlock, amabilis f i r and subalpine f i r from various areas of B.C. Samples of jack pine, red pine, black spruce, white spruce and balsam f i r were obtained from Forintek's eastern laboratory i n Ottawa. Numbers and source of these samples are noted i n Table 3. I t was decided to make preparation and scanning of samples approximate conditions as they may be found under m i l l conditions, i n order to ensure that the method was robust enough to meet i n d u s t r i a l uses, as follows. Several t h i n (5mm) s l i c e s were cut from one face of each board sample using a bandsaw, without regard to r i n g o r i e n t a t i o n or grain angle. No further preparation of the sample faces was done p r i o r to scanning, thereby introducing a degree of v a r i a t i o n from surface smoothness, which i s unavoidable i n a m i l l s i t u a t i o n s . These s l i c e s were placed into i n d i v i d u a l p l a s t i c bags and kept frozen u n t i l used (as f a r as possible, samples were kept at the same moisture content as when received, hopefully preserving the variable moisture content a c t u a l l y found i n f r e s h l y cut lumber). One s l i c e for each sample was also cut to be used for microscopic i d e n t i f i c a t i o n of the species by a wood i d e n t i f i c a t i o n expert at Forintek. This i d e n t i f i c a t i o n was performed by J. Gonzalez, using a combination of gross and microscopic anatomical features. 30 Immediately p r i o r to IR scanning, a piece of wood small enough to f i t the reflectance accessory of the spectrometer was cut from one s l i c e for each sample using a 9 mm diameter c i r c u l a r punch. In samples where sapwood could be discerned a separate sample disk was taken representing sapwood only. For extractive free samples, another disk was taken from the same s l i c e as the o r i g i n a l sample(s), as close to the f i r s t disk as possible. The disk was soxhlet extracted sequentially, f i r s t i n cyclohexane:ethanol (2:1), then ethanol and f i n a l l y water, with the sample being extracted for 2 4 hours i n each solvent and a i r - d r i e d between each extraction. This i s e s s e n t i a l l y TAPPI standard method T12 os-7 5 revised by using cyclohexane i n place of benzene as suggested by Fengel and Przyklenk (1983). A Nicolet model SX2 0 FTIR spectrometer with a Spectra-Tech d i f f u s e reflectance accessory was used for c o l l e c t i o n of a l l spectra. Each sample was scanned 100 times from 4850 to 400 \u20141 \u20141 cm at a resolution of two cm . The spectrometer performed an averaging of these scans, then performed a Fourier transform on the average. A l l samples were scanned at room temperature i n an open sample compartment, f i r s t in the green and then a f t e r freeze-drying for 24 hours. No attempt was made to orient the samples i n any p a r t i c u l a r fashion with regards to grain or growth r i n g orientation. The open sample compartment was to approximate m i l l conditions, where i t would be impractical to obtain the moisture-free, nitrogen purged sample compartment normally used for FTIR scanning. The spectra were stored on the Nicolet data system, then transferred to a VAX computer for data 31 analysis. In t o t a l 740 samples were scanned i n the green condition, 738 freeze dried and 264 extracted. Because an FTIR spectrometer gives a single beam spectrum that i s uncorrected for background, source c h a r a c t e r i s t i c s and detector s e n s i t i v i t y (Figure 1), each absorption spectra was ratioed against the reflectance spectrum of a KBr p e l l e t run on the same day (Figure 2) using the formula: A(i) = Ab(i)\/Am(i) where: A(i) = signal r a t i o at wavenumber ( i ) ; Ab(i) = single beam KBr signal at wavenumber ( i ) ; and Am(i) = single beam sample signal at wavenumber ( i ) . This corrected each spectrum for d a i l y background absorption o r i g i n a t i n g from carbon dioxide and water vapor i n the a i r and d a i l y v a r i a t i o n s i n source intensity and detector s e n s i t i v i t y . The r e s u l t i n g spectra consisted of 4615 data points each, representing ratioed signal strengths at evenly spaced points from 4850 cm - 1 to 400 cm - 1, with each i n t e r v a l being 0.964 cm - 1 (Figure 3). A l l s t a t i s t i c a l analyses were performed on a VAX mainframe computer using the s t a t i s t i c s package SAS. 32 RESULTS AND DISCUSSION This study concentrated on the spectral region from 2500 to 400 cm - 1 (Figure 4) because i t i s i n t h i s region that the most c h a r a c t e r i s t i c \" f i n g e r p r i n t s \" of the wood extractive chemicals are found (see also Table 2) and interference from highly v a r i a b l e moisture content i s minimized. This region from each ratioe d spectrum was submitted to various forms of data ana l y s i s . SPRUCE\/PINE\/ FIR Because the largest single group of samples was the white spruce\/ lodgepole pine\/ subalpine f i r mixture common to i n t e r i o r regions of B.C., t h i s sample set was used to t e s t several p o t e n t i a l techniques for species sorting. Spectra were f i r s t v i s u a l l y compared to determine i f there were any obvious differences between spectra of d i f f e r e n t species. While differences were noted, these differences were not consistent enough from piece to piece within a species to make accurate i d e n t i f i c a t i o n possible. Figure 5 shows a comparison of f i v e t y p i c a l spectra for each of the three species. To adjust for differences i n signal magnitude and o f f s e t of the spectra, the spectra were then normalized using the formula: NA(i) = (A(i)-Aav)\/SD where: NA(i) = normalized absorbance at wavenumber ( i ) ; A(i) = measured absorbance at wavenumber ( i ) ; Aav = average absorbance over entire spectrum; and 33 ( A ( i ) - Aav) 2 n-1 (the standard deviation of entire spectrum). This forced the same average magnitude and standard deviation on each spectrum, making mathematical comparisons possible (Figure 6). Again, v i s u a l comparison of the spectra was not useful for species i d e n t i f i c a t i o n , so t h i s l i n e of attack was abandoned. Next, the average and standard deviation of the absorbance at each wavenumber i n the normalized spectra was calculated for each species. The average +\/- one standard deviation was plot t e d for each wavenumber of each species l i k e l y to be found i n the species mix being studied. When these pl o t s were compared, some differences were noted between the average spectra between species, but the standard deviations were large enough to cause s i g n i f i c a n t overlap among species (Figure 7). A method was needed to discern which spectral areas were p o t e n t i a l l y useful i n species discrimination. P r i n c i p a l component analysis provides a technique for f i n d i n g which var i a b l e s have the most importance i n accounting for differences between samples, so i t was the next method of data analysis attempted. Using techniques suggested by Dr. Veltkamp, University of Washington, each spectrum was mean centered i n order to 34 f a c i l i t a t e quantitative comparisons, using the formula: A(mct)= A ( i ) - Aav where: A(mct)= the mean centered signal at point ( i ) ; A(i) = the ratioed signal at point ( i ) ; and Aav = the average of the ratioed signals across the entire spectrum. Figure 8 i l l u s t r a t e s a mean centered spectrum. Note that t h i s method w i l l make the mean for each spectrum equal to zero, but w i l l not change r e l a t i v e i n t e n s i t i e s of the absorbance values. This i s i n contrast with the normalizing of the spectra where the r e l a t i v e i n t e n s i t i e s are manipulated i n order to force the standard deviation of the entire spectrum to equal one. From the mean centered spectra, 20 samples were randomly chosen from each species to create a data set for p r i n c i p a l component analysis. This data set was thus 3 species with 2 0 spectra per species and 2116 measurements (variables) i n the spectra representing the mean centered signal at evenly spaced points from 2500 to 400 cm - 1. Note that a requirement of p r i n c i p a l component analysis i s that the system be overdetermined, that i s that the number of samples should exceed the number of variables. Since i n our case the opposite case i s c l e a r l y true, we are i n v i o l a t i o n of t h i s assumption. Given t h i s v i o l a t i o n , since i t was impractical to c o l l e c t enough spectra to meet t h i s condition, i t was decided to use p r i n c i p a l component analysis, but to be cautious i n in t e r p r e t a t i o n of the r e s u l t s . In p a r t i c u l a r , the f i n a l t e s t of the r e s u l t s was i n the success (or fa i l u r e ) of the f i n a l s o r t i n g procedure developed, where a l l samples (308) were c l a s s i f i e d 35 using the method developed from a small number of samples (60). I t should also be noted here that the complex character of i n f r a r e d spectra makes the use of multivariate s t a t i s t i c a l methods e s p e c i a l l y d i f f i c u l t . F i r s t , even for a spectrum of a si n g l e compound, the spectrum w i l l represent a s e r i e s of non-independent measurements, with each point i n any absorbance band being correlated. Second, bands i n one area of the spectrum may be strongly or weakly correlated with bands i n another area depending upon which functional groups are present. As well, for a spectrum of a complex mixture of compounds (such as wood) the f i n a l spectrum may not be a l i n e a r combination of i t s component parts, due to chemical interactions between the compounds. These factors make interpretation of the r e s u l t s of any s t a t i s t i c a l method d i f f i c u l t . A further complication to the IR spectrum of wood i s the presence of water i n the wood. B r i e f l y , water can e x i s t i n three d i f f e r e n t conditions within the wood. F i r s t , there i s \"free\" water which i s contained i n i n t r a c e l l u l a r spaces at moisture contents above the f i b e r saturation point. This water i s the f i r s t to be l o s t upon drying. Second, there i s \"bound\" water which i s chemically bound (through hydrogen bonding) to chemical constituents of the wood (such as c e l l u l o s e ) . F i n a l l y , there i s what has been termed \"water of c o n s t i t u t i o n \" which cannot be removed from the wood by drying at 105\u00b0C. This constitutes around 2% of the wood by weight. Each of these forms of water w i l l contribute to the IR spectrum i n a s l i g h t l y d i f f e r e n t way due to t h e i r d i f f e r e n t chemical environments. A Fisher weight was calculated for each of the 2116 36 v a r i a b l e s using the technique described by Sharaf et a l . (1986), where for each variable a difference factor i s calculated between i t and every other variable by: Wk(i,ii)= | X f i ) - X ( i i ) | Variance(i) + Variance(ii) where: Wk(i,ii)= difference factor for points (i) and ( i i ) ; X(i)= mean of variable (i) across the data set; and X ( i i ) = mean of variable ( i i ) across the data set. This factor i s calculated for each possible pair of variables involving point ( i ) , then the o v e r a l l Fisher weight for point (i) i s calculated by : NJ Wk= 1 V\" Wk(j) NJ 2-where: Wk= the o v e r a l l Fisher weight of variable k; NJ= the number of pairs of variables including k; and Wk(j)= the factors calculated i n the preceding equation. Note that where two variables have the same mean, or have large variances, the factor associated with them i s very small and that the Fisher weight can approach zero for variables with l i t t l e discriminating power. Next, the c o r r e l a t i o n matrix for the data set was computed. Using t h i s c o r r e l a t i o n any point highly correlated (r> 0.9995) with the preceding point was eliminated. It was hoped that by removing variables which were e s s e n t i a l l y the same, some degree of c o l i n e a r i t y between variables would be removed, and the number of variables reduced. At the same time, i t was recognized that the spectral differences being looked f o r were probably small so that removal of variables that were not 37 exactly the same could remove useful information. I t was f e l t that a c o r r e l a t i o n of r> 0.9995 would be a good compromise. This resulted i n an abbreviated spectrum f o r each sample, cons i s t i n g of 700 wavelengths. The next step i n the analysis was p r i n c i p a l component analysis on the selected subset (3 species, 20 samples per species) of abbreviated (700 variables) spectra. The SAS routine PRINCOMP (SAS users guide: s t a t i s t i c s , 1982) was used for a l l p r i n c i p a l component analyses. The r e s u l t s of the p r i n c i p a l component analysis are shown i n Table 4. The r e l a t i v e importance of the p r i n c i p a l components can be shown by a \"scree\" plot of t h e i r eigenvalues (Figure 9). Note that the f i r s t f i v e p r i n c i p a l components account for v i r t u a l l y a l l of the v a r i a t i o n found i n the data set. On t h i s basis, the p r i n c i p a l component scores associated with the f i r s t f i v e p r i n c i p a l components of each sample (Table 5) were plotted i n pairs to see i f they were s e n s i t i v e to species -differences (Figures 10- 19). As can be seen from Figure 12, the combination of the f i r s t and fourth p r i n c i p a l components roughly sort the three species into three d i f f e r e n t regions, i n d i c a t i n g that they may be useful for species i d e n t i f i c a t i o n . S i m i l a r l y , the combination of the fourth and f i f t h p r i n c i p a l components (Figure 19) i s seen to separate white spruce from subalpine f i r . The p r i n c i p a l component analysis can also be used to determine which variables exert the strongest influence on each p r i n c i p a l component. Table 6 i l l u s t r a t e s the eigenvectors associated with the f i r s t s i x p r i n c i p a l components and the 38 loadings associated with each variable within each eigenvector. The loadings within each eigenvector can be sorted i n decreasing order to see which variables exert the strongest influence on each p r i n c i p a l component. Note that the loadings range from p o s i t i v e through negative and that the e f f e c t of any variable i s d i f f e r e n t for d i f f e r e n t p r i n c i p a l components. The variables with the three highest and three lowest loadings were selected fo r the f i r s t f i v e p r i n c i p a l components (3x2x5= 30 variables) to perform a discriminant analysis on the SPF mixture. Figure 2 0 shows the location of these points on a t y p i c a l spectrum. A l l discriminant analyses were performed using the SAS routine DISCRIM (SAS users guide: s t a t i s t i c s , 1982) which cal c u l a t e s a l i n e a r discriminant function based upon a measure of generalized squared distance between groups. F i r s t a t r a i n i n g set (3 species, 2 0 samples per species and 30 variables for each sample) was used to c a l c u l a t e a discriminant function. This function was then used to c l a s s i f y the a l l 308 samples (60 for t r a i n i n g set, 248 for t e s t set) into species using the same 30 variables. As shown i n Table 7, of 308 samples 235 (76%) were c o r r e c t l y i d e n t i f i e d by species. To look at t h i s from a p r a c t i c a l standpoint, i f t h i s sorting algorithm was applied to a SPF mix i n a m i l l , the r e s u l t i n g c l a s s i f i c a t i o n would be a set of subalpine f i r 88% pure, a set of lodgepole pine 76% pure and a set of white spruce 66% pure. A s u r p r i s i n g r e s u l t i s the success of the method for sapwood samples (89% correct) i n addition to heartwood (74% correct). This was unexpected, as no sapwood samples were included i n the t r a i n i n g set, and i t was 39 anticipated that the sorting routine would thus be based upon heartwood extractives. The same variables and Fisher weights were then applied to the spectra of the green samples, to assess how the system might work under m i l l conditions. This introduces another complication into the sorting procedure, that of varying moisture content i n the freshly cut, green wood. The r e s u l t s of t h i s t r i a l were 240 of 310 samples c o r r e c t l y i d e n t i f i e d (77%), with 83% of the heartwood samples and 46% of the sapwood samples c o r r e c t l y i d e n t i f i e d (Table 8). Again, the c l a s s i f i e d samples would be 83% pure for subalpine f i r , 81% pure for lodgepole pine and 61% pure for white spruce. Because of the unexpected success with the sapwood samples, i t was decided to t e s t the theory that the sorting procedure was based upon differences i n wood extractive chemicals, and not differences i n the l i g n i n or hemicellulose components. To do t h i s , exactly the same sorting routine was applied to the 264 matching extractive free samples. As expected, the r e s u l t s were very poor (Table 9). V i r t u a l l y a l l (97%) of the samples were i d e n t i f i e d as lodgepole pine, r e s u l t i n g i n 44% (the pine fraction) of the samples being c o r r e c t l y i d e n t i f i e d . I t i s apparent that removing the wood extractives defeated the sorting procedure with the SPF set, indicating that c r i t e r i a for sorting these were based upon wood extractives. This i s a confirmation of previous work by Swan (1966) and Lawrence (1989) whose sorting techniques both r e l i e d upon the presence of s p e c i f i c extractive chemicals to i d e n t i f y species. I t i s also a contradiction of Chow's (1972) predictions that IR 40 species i d e n t i f i c a t i o n would be possible only for sapwood. Once a s u i t a b l e set of known samples are used for t r a i n i n g , the present method i s an improvement upon former wood i d e n t i f i c a t i o n techniques i n that no sample preparation or chemical reagents are required and that i t applies e f f e c t i v e l y to both heartwood and sapwood. The same sorting routine (same set of t r a i n i n g samples, wavelengths used and Fisher weights) was then applied to the 72 samples of eastern SPF (black spruce, white spruce, red pine, jack pine, white pine and balsam f i r ) to determine whether or not i t would be e f f e c t i v e without modifications. The r e s u l t s (Table 10) indicate that the method may be useful (a 68% success rate f o r heartwood and 28% for sapwood, 50% o v e r a l l ) , but would require refinements to be applied for t h i s species mix. This was expected because the eastern SPF mixture i s composed of d i f f e r e n t species from the western SPF mixture, and thus d i f f e r e n t extractive chemicals could be associated with each group. Refinements would involve determining a s p e c i f i c set of wavelengths to use and developing the discriminant function using only eastern SPF samples. The next step was an attempt to reduce the number of wavelengths (variables) used for sorting. The optimum s i t u a t i o n would be where the largest number of samples are i d e n t i f i e d c o r r e c t l y with the lowest possible number of var i a b l e s . F i r s t , the two variables with the smallest loadings were dropped from each p r i n c i p a l component, leaving 2 0 variables (5 p r i n c i p a l components with four variables each). The r e s u l t s (Table 11) show that t h i s set of variables i s s l i g h t l y more 41 e f f e c t i v e i n o v e r a l l sorting (80% correct), less e f f e c t i v e for sapwood (35% correct) and more e f f e c t i v e for heartwood (88% correct) than using 30 variables. Next, the two variables with the next smallest loadings were dropped from each p r i n c i p a l component, leaving 10 variables (5 p r i n c i p a l components with two variables each). These r e s u l t s (Table 12) again show a s l i g h t improvement o v e r a l l (83% c o r r e c t ) , with s l i g h t l y poorer sorting for heartwood (84% correct) and s l i g h t l y better for sapwood (76% c o r r e c t ) . This sorti n g technique was also applied to extractive free wood (Table 13), again with very poor r e s u l t s (29% c o r r e c t ) , with most (74%) of the samples c l a s s i f i e d as white spruce. Because of the improved r e s u l t s with the western SPF t h i s technique was also applied to the eastern SPF. The r e s u l t s (Table 14) were a s l i g h t improvement over the technique using 3 0 variables, with 58% correct o v e r a l l (70% of heartwood and 44% of sapwood). This again showed promise for the technique i f the sort i n g routine was optimized for t h i s species mix. F i n a l l y , the two p r i n c i p a l components showing the smallest apparent usefulness for separation, P2 and P5 (Figures 10, 13-16, 18, 19) were dropped, leaving only s i x variables (three p r i n c i p a l components with two variables each). The r e s u l t s of t h i s s o r t i n g (Table 15) were only s l i g h t l y poorer than using 10 variables, with 81% correct o v e r a l l and 80% correct for heartwood and a s l i g h t improvement for sapwood, with 85% correct. To summarize, the most e f f e c t i v e o v e r a l l c l a s s i f i c a t i o n model for the dry samples used 3 0 wavelengths and c o r r e c t l y 42 c l a s s i f i e d 76% of samples, including 74% of heartwood samples and 89% of sapwood samples (Table 7). For the green samples, the most e f f e c t i v e sort used ten wavelengths and c o r r e c t l y assigned species to 83% of green samples representing 84% of heartwood samples and 76% of sapwood samples (Table 12). Since the combination of Fisher weighting, PC analysis and discriminant analysis resulted i n a discriminant function that was successful for both the t r a i n i n g set and the t e s t set, i t would appear that although we have viol a t e d some assumptions of the techniques, the approach outlined i s nonetheless v a l i d . I t was thus decided to attempt the c l a s s i f i c a t i o n of other species groups using the same technique. F i n a l l y , as a check on the accuracy of the anatomical i d e n t i f i c a t i o n of the samples, 16 samples which had been m i s c l a s s i f i e d by discriminant analysis were submitted for r e i d e n t i f i c a t i o n . A l l 16 were i d e n t i f i e d as belonging to the same species as o r i g i n a l l y assigned. DOUGLAS-FIR\/WESTERN LARCH Spectra were f i r s t ratioed against the appropriate background spectra, then mean centered. Ten dry heartwood spectra were randomly chosen for each species to develop the sor t i n g method, following the same procedures as outlined for the western SPF mixture. The r e s u l t i n g data set was 2 0 samples (two species with ten samples each) with 2116 wavelengths each. The Fisher weights for each wavelength were calculated and the c o r r e l a t i o n matrix was computed. Again, any point highly correlated (r> 0.9995) with the preceding point was eliminated. This resulted i n an abbreviated spectrum for each sample, 43 c o n s i s t i n g of 931 wavelengths. The next step was the p r i n c i p a l component analysis on the selected subset (two species with ten samples each) of abbreviated (931 variables) spectra. Due to memory li m i t a t i o n s of the computer system, the p r i n c i p a l component analysis had to be completed i n stages. F i r s t , the 931 variables were divided into two parts, the f i r s t 460 variables, then the l a s t 471 variables. Each of these parts was analyzed (Table 16) and the r e l a t i v e importance of the p r i n c i p a l components in each determined by a \"scree\" p l o t of t h e i r eigenvalues (Figures 21, 22). Note that the f i r s t four p r i n c i p a l components accounted for v i r t u a l l y a l l of the v a r i a t i o n found i n each data set. The variables with the ten highest and ten lowest loadings were selected for the f i r s t four p r i n c i p a l components i n each data set, y i e l d i n g a set of 160 variables (four p r i n c i p a l components with 20 variables each i n each of 2 data s e t s ) . A p r i n c i p a l component analysis was then performed on t h i s data set (Table 17). Again the f i r s t four p r i n c i p a l components were found to be the most important (Figure 23). The variables with the f i v e highest and f i v e lowest loadings were selected for the f i r s t four p r i n c i p a l components (4x2x5= 40 variables) to perform a discriminant analysis on the mixture. Figure 2 4 shows the l o c a t i o n of these points on a t y p i c a l spectrum. Using these selected variables, a discriminant analysis was performed upon the entire sample set. F i r s t the t r a i n i n g set (two species with ten samples per species and 40 variables per sample) was used to calculate a discriminant function. This 44 function was then used to c l a s s i f y the remaining 172 samples into species using the same 40 variables. SAS (DISCRIM procedure) output for t h i s t r i a l reported a warning that the covariance matrix was not f u l l rank and that only the f i r s t 18 variables had been used for the discriminant function. The not f u l l rank warning indicates that some vari a b l e s are highly correlated, thus redundant and dropped from the analysis. As shown in Table 18, by using only the f i r s t 18 v a r i a b l e s 188 of 192 samples were c o r r e c t l y i d e n t i f i e d by species, or 98%. This f r a c t i o n applied both to heartwood and sapwood, western larch was 100% c o r r e c t l y c l a s s i f i e d and Douglas-fir 88%. In an attempt to include the dropped variables, 2 0 more larch samples were moved to the t r a i n i n g set, hoping to decrease the c o r r e l a t i o n between variables. In t h i s t r i a l (Table 19), again the rank warning occurred, but only two variables were dropped, leaving 38 variables. The c l a s s i f i c a t i o n r e s u l t s were 85% correct, which was not as good as with only 18 variables considered. Heartwood re s u l t s (91% correct) were better than sapwood (77%) and larch (88%) better than Douglas-fir (70%). Next, the f i r s t method (18 variables, 10 samples\/species for training) was applied to the wet samples (Table 20). Again the rank warning occurred, t h i s time dropping another variable, meaning only 17 variables were used. The r e s u l t s were good, with 81% c o r r e c t l y c l a s s i f i e d , representing 93% of the heartwood and 66% of the sapwood. However, by species the r e s u l t s were disappointing, with only 59% of the Douglas-fir c o r r e c t l y c l a s s i f i e d . 45 In an attempt to f i n d a better sorting method, the o r i g i n a l 40 variables were analyzed i n three groups. The f i r s t group included the f i r s t 14 variables, the second group the next 14 and the t h i r d group the l a s t 12. Results are presented i n Tables 21, 22 and 23, respectively. Using the f i r s t 14 variables, 90% of a l l samples were c o r r e c t l y c l a s s i f i e d (95% of heartwood and 82% of sapwood, 88% of Douglas-fir and 90% of la r c h ) . Using the next 14 variables, 84% of a l l samples were c o r r e c t l y c l a s s i f i e d (90% of heartwood and 77% of sapwood, 68% of Douglas-fir and 88% of larch). Using the l a s t 12 variables, 91% of a l l samples were corre c t l y c l a s s i f i e d (90% of heartwood and 91% of sapwood, 85% of Douglas-fir and 92% of l a r c h ) . To summarize, the most e f f e c t i v e o v e r a l l c l a s s i f i c a t i o n model f o r t h i s species group used 18 wavelengths and c l a s s i f i e d 98% of dry samples co r r e c t l y for both heartwood and sapwood (Table 18). For green samples, the best sort used 12 wavelengths and co r r e c t l y assigned species to 91% of green samples, representing 90% of heartwood samples and 91% of sapwood samples (Table 23). Once again the combination of Fisher weighting, PC analysis and discriminant analysis resulted i n a discriminant function that was successful for both the t r a i n i n g set and the t e s t set. It would appear that although we have v i o l a t e d some assumptions of the techniques, the approach outlined i s again v a l i d . WESTERN HEMLOCK\/ SITKA SPRUCE\/ AMABILIS FIR Spectra were f i r s t ratioed against the appropriate background spectra, then mean centered. Ten dry heartwood spectra were randomly chosen for each species to develop the 46 s o r t i n g method, following the same procedures as outlined for western SPF. The r e s u l t i n g data set was 3 0 samples (three species with ten samples per species) with 2116 wavelengths each. The Fisher weights for each wavelength were calculated and the c o r r e l a t i o n matrix was computed. Again, any point highly correlated (r> 0.9995) with the preceding point was eliminated. This resulted i n an abbreviated spectrum for each sample, cons i s t i n g of 1085 wavelengths. The next step was the p r i n c i p a l component analysis on the selected subset (three species with ten samples per species) of abbreviated (1085 variables) spectra. Due to memory li m i t a t i o n s of the computer system, the p r i n c i p a l component analysis had to be completed i n stages. F i r s t , the 1085 variables were divided into two parts, the f i r s t 542 variables, then the l a s t 543 variables. Each of these parts was analyzed (Table 24) and the r e l a t i v e importance of the p r i n c i p a l components i n each determined by a \"scree\" p l o t of t h e i r eigenvalues (Figures 25,26). Note that the f i r s t f i v e p r i n c i p a l components account for v i r t u a l l y a l l of the v a r i a t i o n found i n each data set. The r e s u l t s from the two p r i n c i p a l component analyses were pooled and the variables with the 20 highest and 20 lowest loadings were selected for the f i r s t f i v e p r i n c i p a l components y i e l d i n g a set of 200 variables (40 variables i n each of f i v e p r i n c i p a l components). When a variable occurred more than once i n t h i s set of 200, duplicates were dropped, r e s u l t i n g i n a f i n a l set of 171 unique variables. 47 A p r i n c i p a l component analysis was then performed on t h i s data set (Table 25). Again the f i r s t f i v e p r i n c i p a l components were found to be the most important (Figure 27). The variables with the two highest and two lowest loadings were selected for the f i r s t f i v e p r i n c i p a l components (4x5= 20 variables) to perform a discriminant analysis on the mixture. Figure 28 shows the l o c a t i o n of these points on a t y p i c a l spectrum. Using these selected variables, a discriminant analysis was performed upon the entire sample set. F i r s t the t r a i n i n g set (three species with ten samples per species and 2 0 variables i n each sample) was used to calculate a discriminant function. This function was then used to c l a s s i f y the remaining 136 samples into species using the same 20 variables. SAS (DISCRIM procedure) output for t h i s t r i a l reported a warning that the covariance matrix was not f u l l rank and that one variable had been dropped from the discriminant function. As shown i n Table 26, by using 19 variables 138 of 166 samples were c o r r e c t l y i d e n t i f i e d by species, or 83 %. In t h i s species group not enough sapwood pieces were avail a b l e to make any meaningful predictions, so sapwood samples w i l l be r e f e r r e d to only as part of the t o t a l sample set. The sorting routine was roughly as e f f e c t i v e for a l l three species with Sitka spruce, western hemlock and amabilis f i r being c l a s s i f i e d c o r r e c t l y i n 84%, 83%, and 81% of samples respectively. Next, the discriminant analysis was attempted using only the variables with the highest and lowest loadings for the f i r s t f i v e p r i n c i p a l components, r e s u l t i n g i n ten variables (fiv e p r i n c i p a l components with two variables each). The r e s u l t s 48 (Table 27) were somewhat poorer than for 20 variables, with only 70% of a l l samples being corr e c t l y c l a s s i f i e d , with a decrease i n effectiveness for each species (amabilis f i r to 75%, Sitka spruce to 78% and western hemlock to 65%). Next, the f i r s t method (20 variables 10 samples\/species for training) was applied to the wet samples (Table 28). Again the rank warning occurred, t h i s time dropping f i v e variables, meaning only 15 variables were used. The r e s u l t s were poor, with 67% c o r r e c t l y c l a s s i f i e d , representing 69% of the Sitka spruce, 63% of the amabilis f i r and 67% of the western hemlock. The discriminant analysis was then attempted using only the variables with the highest and lowest loadings for the f i r s t f i v e p r i n c i p a l components, r e s u l t i n g i n 10 variables (fiv e p r i n c i p a l components with two variables each). The r e s u l t s (Table 29) were somewhat poorer than for 20 variables, with only 62% of a l l samples being c o r r e c t l y c l a s s i f i e d , with a decrease i n effectiveness for each species (amabilis f i r to 56%, Sitka spruce to 60% and western hemlock to 65%). In an attempt to develop a sorting method applicable to western hemlock\/Sitka spruce only, a l l amabilis f i r samples were dropped from the data and c l a s s i f i c a t i o n attempted f o r western hemlock and Sitka spruce only. Using the same 2 0 variables as selected for the sort of a l l three species (with the same f i v e dropped by SAS), the r e s u l t s were promising, with 82% of a l l samples being c l a s s i f i e d correctly, 88% of Sitka spruce and 78% of western hemlock (Table 30) . When only 10 variables were used (fiv e p r i n c i p a l components with two variables each), the r e s u l t s were almost the same 49 (Table 31), with 82% of samples c o r r e c t l y c l a s s i f i e d , although Sitka spruce was s l i g h t l y poorer (83%) and western hemlock s l i g h t l y better (82%). To summarize for t h i s species mixture, the most e f f e c t i v e sort f o r the dry samples used 19 wavelengths and c o r r e c t l y c l a s s i f i e d 83% of the samples, 85% of heartwood samples and 56% of sapwood samples (Table 26). C l a s s i f i c a t i o n of the green samples proved d i f f i c u l t , with the best sort only 67% correct and using 15 wavelengths (Table 28). However, i f only western hemlock and Sitka spruce were sorted, the effectiveness rose to 82%, 82% for heartwood and 78% for sapwood (Table 30). Once again the combination of Fisher weighting, PC analysis and discriminant analysis resulted i n a discriminant function that was successful for both the t r a i n i n g set and the t e s t set. I t would appear that although we have v i o l a t e d some assumptions of the techniques, the approach outlined i s again v a l i d . To summarize t h i s section, we have demonstrated that the combination of Fisher weighting, p r i n c i p a l component analysis, and discriminant analysis with a l i n e a r discriminant function enable us to sort the three species groups with a good degree of success. This demonstrates the v a l i d i t y of the o r i g i n a l hypothesis that the reflectance FTIR spectra of the wood samples contain the information needed to c l a s s i f y the samples by species. Further, we have shown that the c l a s s i f i c a t i o n c r i t e r i a are based upon the extractive chemicals, not upon other chemical differences. This technique has proven successful for both heartwood and sapwood, and for green and freeze-dried samples. 50 This c l a s s i f i c a t i o n scheme i s by no means presented as the most e f f e c t i v e achievable, nor as the only method which w i l l work. Evaluation of alternative multivariate techniques would be required before any such conclusions could be supported. In p a r t i c u l a r , a sorting scheme which allows a separate c l a s s i f i c a t i o n f or samples which do not f i t well into a s p e c i f i e d c lass would be advantageous. SIMCA i s a possible technique for t h i s purpose. The techniques chosen are biased towards using the fewest possible number of wavelengths, mainly for the purpose of designing a simple i d e n t i f i c a t i o n scheme which would be of use to the wood industry. Further investigation of other techniques which make use of more of the variables could y i e l d a more accurate sorting algorithm, although at the possible cost of more complicated systems. 51 CONCLUSIONS The reflectance Fourier transform i n f r a r e d technique presented here i s shown to be an e f f e c t i v e and quick method for d i f f e r e n t i a t i n g between woods of c e r t a i n species i n the groupings which were studied. I t was shown to be f a i r l y accurate f o r separating green and freeze-dried samples, as well as heartwood and sapwood for the species studied. When the extractive chemicals were removed from the wood c l a s s i f i c a t i o n was unsuccessful using the same c l a s s i f i c a t i o n parameters. This indicates that the sorting c r i t e r i a are dependant upon the presence of extractive chemicals, both i n heartwood and sapwood. Although the method as outlined measured the e n t i r e mid-IR spectrum of the samples studied, in a l l cases only a very small number of wavelengths (between 6 and 30) were employed to e f f e c t i v e l y d i f f e r e n t i a t e between species. While i t i s beyond the scope of t h i s thesis, i t i s l i k e l y that species i d e n t i f i c a t i o n as i l l u s t r a t e d could be performed by measuring the IR r e f l e c t i o n of samples only at the wavelengths a c t u a l l y employed. This would eliminate the time consuming measurement of the e n t i r e spectrum at high resolution. Such a technique would involve only very simple equipment, capable of measuring at most 3 0 s p e c i f i c wavelengths, which would be much more fe a s i b l e i n an i n d u s t r i a l setting than the complex and s e n s i t i v e FTIR instrument\u2022used here. The method as described i s the f i r s t use of reflectance IR to d i f f e r e n t i a t e wood species by t h e i r unique extractive chemicals and demonstrates the potential of t h i s technique for t h i s a p p l i c a t i o n . The e a r l i e r attempt to i d e n t i f y wood species 52 by IR spectroscopy (Chow, 1972) u t i l i z e d transmission spectra which would be impractical i n a m i l l , and claimed that the ex t r a c t i v e chemicals interfered with i d e n t i f i c a t i o n . Other work i n t h i s area has been r e s t r i c t e d to d i f f e r e n t i a t i o n of hardwoods from softwoods (Owen and Thomas, 1989), and d i f f e r e n t i a t i o n of charcoals (Wood, 1988). The p o s s i b i l i t y also e x i s t s for a p p l i c a t i o n of t h i s technique i n other areas where automated sample i d e n t i f i c a t i o n or sorting i s required. The method also d i f f e r s from previous work i n that no use of absorbance ranges attributable to s p e c i f i c chemical groups i s attempted. Instead, raw reflectance signal i n t e n s i t y data i s a l l that i s used. Based upon the r e s u l t s from the extractive free samples, the method uses presence or absence of absorptions caused by p a r t i c u l a r wood extractive chemicals to d i f f e r e n t i a t e between species. The sort as practiced here was not 100% correct for any species group, i t was more r e l i a b l e than t r a d i t i o n a l chemical t e s t s described and the ion mobility spectrometry technique ( a l l of which are e f f e c t i v e only i n d i f f e r e n t i a t i n g heartwoods) and much fa s t e r than the anatomical i d e n t i f i c a t i o n of wood samples, which takes a trained technologist three to f i v e minutes per sample for sample preparation and i d e n t i f i c a t i o n . I t i s also a non-invasive technique, requiring no physical contact with the sample, i n contrast with the other species i d e n t i f i c a t i o n procedures. The procedure could be e a s i l y automated using computers and f i b e r optics could be used to i s o l a t e the s e n s i t i v e equipment from the m i l l environment, making i t e s p e c i a l l y a t t r a c t i v e from an applications standpoint. LITERATURE CITED SAS user's guide: s t a t i s t i c s . 1982. SAS I n s t i t u t e Inc., Cary, North Carolina. 584pp. Abbot, T. P., Palmer, D. M., Gordon, S. H. and M. 0. Bagby. 19 S o l i d state analysis of plant polymers by FTIR. J . Wood Chem. and Tech. 8(1): 1-28. Barton, G. M. 1973. Chemical color tests for Canadian woods. Canadian Forest Industries 93(2): 57-60,62. Berben, S.A., Rademacher, J.P., Lowell, O.S. and D. B. Easty. 1987. Estimation of l i g n i n i n wood pulp by d i f f u s e reflectance Fourier transform infrared spectrometry. Tappi 70(11): 129-133. Chow, S. Z. 1972. Infrared spectral study of woody ti s s u e s from four conifers. Wood Science 5(1): 27-33. Devaux, M., Bertrand, D., Robert, P. and J. Morat. 1987. Extraction of near infra-red spectral information by fa s t Fourier transform and p r i n c i p a l component analysis Application to the discrimination of baking q u a l i t y of wheat f l o u r s . J. Chemometrics. 1(2): 103-110. Durig, D. T., Esterle, T. J., Dickson, T. J. and J. R. Durig. 1988. An investigation of the chemical v a r i a b i l i t y of woody peat by FTIR spectroscopy. Applied Spectrosc. 42(7): 1239-1244. Faix, O. 1986. Investigation of l i g n i n polymer models (DHP's) by FTIR spectroscopy. Holzforsch. 40: 273-280. Faix, O. 1988. P r a c t i c a l uses of FTIR spectroscopy i n wood science and technology. Mikrochim. Acta 1988, I: 21-2 5. Fengel, D. and M. Przylenk. 1983. Comparative extract determination for replacing benzene by cyclohexane. Holz a l s Roh und Werkstoff 41: 193-194. Fengel, D. and G. Wegener. 1984. Wood: chemistry, u l t r a s t r u c t u r e , reactions. Walter de Gruyter, New York. 613pp. Goring, D. A. I. 1970. Microscopic patterns of l i g n i n removal during chemical pulping. Tappi Special Technical Association Publication STAP No. 8.: 107-124. Grandmaison, J. L., Thibault, J. and S. Kaliaguine. 1987. Fourier transform infrared spectrometry and thermograv-imetry of p a r t i a l l y converted l i g n o c e l l u l o s i c materials. Anal. Chem. 59: 2153-2157. 54 Gurnagul, N., St-Germain, F. G. T. and D. G. Gray. 1986. Photoacoustic Fourier transform infrared measurements on paper. J. Pulp Paper S c i . 12(5): J156-J159. H i l l i s , W. E. 1962. Wood extractives and t h e i r s i g n i f i c a n c e to the pulp and paper industries. Academic Press, New York. 513pp. . 1987. Heartwood and tree exudates. Springer-Verlag, New York. 2 68pp. Hope, K. 1968. Methods of multivariate analysis. University of London Press Ltd. 165pp. Johnson, R. A. and D. W. Wichern. 1982. Applied M u l t i v a r i a t e s t a t i s t i c a l analysis. Prentice-Hall, Englewood C l i f f s , N.J. 594pp. Klecka, W. R. 1980. Discriminant analysis. Sage Publications, New York. 71pp. Kolboe, S. and 0. E l l e f s e n . 1962. Infrared investigations of l i g n i n : A discusssion of some recent r e s u l t s . Tappi 45(2): 163-166. Kuo, M., McClelland, J. F., Luo, S., Chien, P. and R. D. Walker. 1988. Applications of infrared photoacoustic spectroscopy for wood samples. Wood Fiber S c i . 20(1): 132-145. Lavine, B. K., Carlson, D. A., Henry, D. and P. C. Jurs. 1988. Taxonomy based on chemical const i t u t i o n : d i f f e r e n t i a t i o n of A f r i c a n i z e d honey-bees from european honey-bees. J. Chemometrics. 2(1): 29-37. Lawrence, A.H. 1989. Rapid characterization of wood species by ion mobility spectrometry. J. Pulp and Paper Science 15(5): J196-J199 Malinowski, E. R. and D. G. Howery. 1980. Factor analysis i n chemistry. John Wiley & Sons, New York. 251pp. Manville, J . F. and A. S. Tracey. 1989. Chemical differences between alpine f i r s of B r i t i s h Columbia. Phytochemistry. 28(10): 2681-2686. Marton, J . and H. E. Sparks. 1967. Determination of l i g n i n i n pulp and paper by infrared multiple i n t e r n a l reflectance. Tappi 50(7): 363-368. Massart, D. L. , Vandeginste, B. G. M., Deming, S. N., Michotte, Y. and L. Kaufman. 1988. Chemometrics: a textbook. E l s e v i e r Science Publishers, New York. 488pp. 55 M i l l e r , R. B. , Quirk, J. T. and D. J. Christensen. 1985. Identifying white oak logs with sodium n i t r i t e . Forest Products J. 35(2): 33-38. Mi c h e l l , A. J. 1988a. Note on a technique for obtaining i n f r a r e d spectra of treated wood surfaces. Wood Fiber S c i . 20(2): 272-276. . 1988b. Usefulness of Fourier transform i n f r a r e d difference spectroscopy for studying the reactions of wood during pulping. Cellulose Chem. Technol. 22: 105-113. ., Watson, A. J. and H. G. Higgins. 1965. An inf r a r e d spectroscopic study of d e l i g n i f i c a t i o n of Eucalyptus regnans. Tappi 48(9): 520-532. Obst, J. R. 1982. Guaiacyl and syringyl l i g n i n composition i n hardwood c e l l components. Holzforsch. 36: 143-152. Owen, N. L. and D. W. Thomas. 1989. Infrared studies of \"Hard\" and \"Soft\" woods. Applied Spectrosc. 43(3): 451-455. Saad, S. M., Issa, R. M. and S. Fahmy. 1980. Infrared spectroscopic study of bagasse and unbleached high - y i e l d soda bagasse pulps. Holzforsch. 34: 218-222. Sarkanen, K. V., Chang, H. and B. Ericsson. 1967a. Species v a r i a t i o n i n l i g n i n s . I. Infrared spectra of guaiacyl and s y r i n g y l models. Tappi 50(11): 572-575. and G. G. Al l a n . 1967b. Species v a r i a t i o n i n l i g n i n s . I I. Conifer l i g n i n s . Tappi 50(12): 583-587. . 1967c. Species v a r i a t i o n i n l i g n i n s . I I I . Hardwood l i g n i n s . Tappi 50(12): 587-590. Schultz, T. P. and W. G. Glasser. 1986. Quantitative s t r u c t u r a l analysis of l i g n i n by d i f f u s e reflectance Fourier transform infrared spectrometry. Holzforsch. 40(SUPP): 37-44. Schultz, T. P., Templeton, M. C. and G. D. McGinnis. 1985. Rapid determination of li g n o c e l l u l o s e by d i f f u s e reflectance Fourier transform infrared spectrometry. Anal. Chem. 57: 2867-2869. Sharaf, M. A., Illman, D. L. and B. R. Kowalski. 1986. In Chemometrics. John Wiley and sons, Inc. 33 2pp. S t r e l i s , L. and R. W. Kennedy. 1967. I d e n t i f i c a t i o n of North American commercial pulpwoods and pulp f i b e r s . University of Toronto Press. 117pp. 56 Swan, E. P. 1966. Chemical methods of d i f f e r e n t i a t i n g the wood of several western conifers. Forest Products J . 16(1): 51-54. Wood, D. J. 1988. Characterisation of charcoals by DRIFT. Mikrochim. Acta 1988, I I : 167-169. TABLE 1. INFRARED ABSORPTION BAND ASSIGNMENTS FOR LIGNIN WAVENUMBERS(cm x) FUNCTIONAL GROUP ASSIGNMENT REFERENCES 3435 Bonded O-H stretching v i b r a t i o n 2,5 3000 C-H stretching v i b r a t i o n 2,5 2940 CH2 antisymmetric stretching v i b r a t i o n 2,5 2880 C-H stretching v i b r a t i o n i n t e r t i a r y CH groups 2,5 2845 C-H stretching v i b r a t i o n i n methoxyl and CH2 symmetrical stretching v i b r a t i o n 2 1735 Acetyl groups i n softwood l i g n i n s 4,5 1730 C=0 stretching v i b r a t i o n of B keto groups 2,5 1715 Unconjugated carbonyl i n guaiacyl or guaicyl\/syringyl l i g n i n 8 1675 Conjugated carbonyl i n guaiacyl or guaicyl\/syringyl l i g n i n 8 167 0 C=0 i n A position on benzene nucleus 1 1660 C=0 stretching v i b r a t i o n of A keto groups 2 1630 In plane deformation of water molecules 5 1600 benzene 1\/5,8 Syringyl units 4 1595 C=C stretching v i b r a t i o n of benzene r i n g 2,3,5,8 1515 benzene rin g vi b r a t i o n i n guaicyl group 1\/5,8 1505 C=C stretching v i b r a t i o n of benzene r i n g 2,3,8 1495 benzene rin g vi b r a t i o n i n coumyrl group 1 1470 Syringyl units 4 Methyl and methylene i n guaiacyl or guaicyl\/syringyl l i g n i n 8 14 60 possibly aromatic mixed C=C stretching and C-H i n plane bending or C-H bending vibrat i o n s in methoxyl and methylene groups 2,8 1430 Aromatic ri n g i n guaicyl l i g n i n 8 1425 C-H bending vibrati o n i n methoxyl groups 2 Aromatic ri n g i n guaicyl\/syringyl l i g n i n 8 1375 Softwood l i g n i n s 4 1370-1365 Methyl CH, OH, phenols in guaiacyl or guaicyl\/syringyl l i g n i n 8 1360 O-H i n plane bending v i b r a t i o n 2 1350 In plane bending of CH3 5 1335 Syringyl units 4 1330 Syringyl l i g n i n 7 1330-1325 CO of syringyl ring, OH i n alcohols 8 1280 Deformation of COOH 5 1275 CO of guaiacyl ring, a r y l CO of a r y l - a l k y l ethers 8 1270 Guaiacyl l i g n i n 7 1265 C-O-C asymmetric stretching v i b r a t i o n of a r y l ether linkages 2 TABLE 1 (CONT). INFRARED ABSORPTION BAND ASSIGNMENTS FOR LIGNIN WAVENUMBERS(cm A ) - FUNCTIONAL GROUP ASSIGNMENT REFERENCES 1260-1230 Softwood l i g n i n s 4 1240 Deformation of phenolic OH 5 1235-1230 CO of syri n g y l r i n g 8 12 3 0 C-O-C asymmetric stretching v i b r a t i o n of a r y l - a r y l ether linkages 2,8 CO of guaiacyl ring, CO of phenols 8 1180 possibly less common C-0 linkage or mixed vi b r a t i o n 2 1150 possibly less common C-0 linkage or mixed vi b r a t i o n 2 1140 OH 5 Guaiacyl or guaicyl\/syringyl CH 8 1130 Syringyl units 4 OH 5 Syringyl CH 8 112 0 C-O-C asymmetric stretching v i b r a t i o n of d i a l k y l ether linkages 2 1090 symmetric stretching analog of band at 1265 or 1230 2 1085 CO of secondary alcohols i n guaiacyl or guaicyl\/syringyl l i g n i n , a l k y l CO i n a r y l - a l k y l ethers i n guaiacyl or guaic y l \/ s y r i n g y l l i g n i n 8 1035 CO i n primary alcohols i n guaiacyl l i g n i n , guaiacyl CH 8 1030 CO i n primary alcohols i n syri n g y l l i g n i n 8 922 unassigned 2 92 0 Syringyl units 4 915 Syringyl l i g n i n 8 875 C-H out of plane bending v i b r a t i o n 2 860 Syringyl l i g n i n 8 855 Guaiacyl l i g n i n 7,8 835 C-H out of plane bending v i b r a t i o n 2 Syringyl l i g n i n 7 815 Guaiacyl l i g n i n 7,8 800 unassigned 2 620 unassigned 2 53 0 C-O-C bending vibrat i o n 2 470 unassigned, possibly aromatic r i n g deformation 2 REFERENCES 1 Kolboe and E l l e f s e n (1962) 2 Michel l et a l ^ (1965) 3 Marton and Sparks (1967) 4 Sarkanen et a l . (1967a,b,c) 5 Saad (1980) 6 Chow (1972) 7 Obst (1982) 8 Schultz and Glasser (1986) 9 Grandmaison et a l . (1987) TABLE 2. INFRARED ABSORPTION BAND ASSIGNMENTS FOR WOOD WAVENUMBERS (cin ) FUNCTIONAL GROUP ASSIGNMENT REFERENCES 3600-3200 H-bonded OH stretching 3 3 300 Bonded O-H stretching v i b r a t i o n 2 297 0 CH stretching 3 2945 CH2 antisymmetrical stretching 3 2914-2870 CH stretching 3 2850 CH2 stretching 3 2900 C-H stretching v i b r a t i o n 2 174 0 Uronic acid and acetyl groups i n hemicellulose 9 1735 Xylan (B-D-xylose, 4-O-methyl-A-D glucuronic acid, B-L-arabinose) 3 1730 C=0 stretching v i b r a t i o n i n glucuronoxylan 2 C=0 stretching of acetyl or carboxyl groups i n hemicellulose, keto groups i n aromatic l i g n i n r i n g , and extractives 6 1730-1725 C=0 stretching of acetyl or carboxylic acid 3 1670 Lignin 3 1660 Lignin 2 1650 O-H bending i n absorbed water 2 Polyphenol, carbonyl and water 6 163 5 Adsorbed water 3 1600 COO- ion i n glucuronoxylan (after s a l t formation) 2,3 Lignin 4 1595 Lignin 2,3 1580 Aromatic s k e l e t a l vibration i n l i g n i n and benzenoid extractives 6 1505 Lignin 2,3,4,9 1500 Aromatic s k e l e t a l v i b r a t i o n i n l i g n i n and benzenoid extractives 6 1470 Lignin 4 1460 Lignin and CH2 bending v i b r a t i o n i n xylan 2,3 1455-1400 OH i n plane bending 3 143 0 CH2 symmetrical bending mode of hydroxymethyl 3 1430 Lignin 4 1425 CH2 scis s o r v i b r a t i o n i n c e l l u l o s e 2 B-D-glucose i n ce l l u l o s e 3 Carboxylic acid and COO- vi b r a t i o n 3 1420 CH bending of methoxyl groups i n l i g n i n , CH bending of methylol groups i n carbohydrates 6 1380 CH bending 3 1370 CH bending vibrati o n i n c e l l u l o s e and hemicellulose 2 1335-1315 CH2 wagging 3 1333 O-H i n plane bending v i b r a t i o n i n c e l l u l o s e and hemicellulose 2 1320 CH2 wagging vibrati o n i n c e l l u l o s e 2 1270 Lignin 3,4 1260 Lignin 2 1240 C-0 stretching v i b r a t i o n i n glucuronoxylan (mainly acetyl; some carboxyl) 2 CO of acetyl 3 TABLE 2(CONT). INFRARED ABSORPTION BAND ASSIGNMENTS FOR WOOD WAVENUMBERS(cm-1) FUNCTIONAL GROUP ASSIGNMENT REFERENCES 1230 Lignin 2,4 1205 Band i n c e l l u l o s e and hemicellulose (unassigned) 2 1160 C-O-C antisymmetric bridge stretching v i b r a t i o n i n c e l l u l o s e and hemicellulose 2,3 1125- 895 CO stretching and rin g v i b r a t i o n modes 3 1110 O-H association band i n c e l l u l o s e and hemicellulose (minor l i g n i n contribution) 2 1050 C-0 stretching v i b r a t i o n i n c e l l u l o s e and hemicellulose 2 1030 C-0 stretching v i b r a t i o n i n c e l l u l o s e and hemicellulose (minor l i g n i n contribution) 2 990 C-0 stretching v i b r a t i o n i n c e l l u l o s e and hemicellulose 2 895 Cha r a c t e r i s t i c of B-link 3 893 Anomeric carbon group frequency i n c e l l u l o s e and xylan 2,9 890 B-D-xylose in xylan 3 870 glucomannan (B-D-glucose, B-D-mannose) 3 835 Lignin 2 805 glucomannan (B-D-glucose, B-D-mannose) 3 768 Water-soluble polysaccharides (B-D-galactose, B-L-arabinose) 3 700-650 OH out of plane bending 3 663 B-D-glucose i n ce l l u l o s e 3 REFERENCES 1 Kolboe and E l l e f s e n (1962) 2 Michel l et a l . (1965) 3 Marton and Sparks (1967) 4 Sarkanen et a l . (1967a,b,c) 5 Saad (1980) 6 Chow (1972) 7 Obst (1982) 8 Schultz and Glasser (1986) 9 Grandmaison et a l . (1987) 61 TABLE 3. ORIGIN AND SPECIES OF WOOD SAMPLES SAMPLE SAMPLE NUMBER OF SOURCE OF SAMPLES CODE SPECIES SAMPLES 88A SITKA SPRUCE 54 POWELL RIVER, B.C. WESTERN HEMLOCK 45 AMABILIS FIR 16 88B LODGEPOLE PINE 108 CRESTON, B.C. WHITE SPRUCE 13 SUBALPINE FIR 13 88C DOUGLAS-FIR 20 CHEMAINUS, B.C. 88D JACK PINE 7 FORINTEK CANADA RED PINE 6 EASTERN LABORATORY WHITE PINE 6 OTTAWA, ONT. BLACK SPRUCE 9 WHITE SPRUCE 6 BALSAM FIR 6 89A LODGEPOLE PINE 7 CRESTON, B.C. WHITE SPRUCE- 40 SUBALPINE FIR 28 89B LODGEPOLE PINE 2 LILLOOET, B.C. WHITE SPRUCE 0 SUBALPINE FIR 53 89C WESTERN LARCH 89 CRANBROOK, B.C. DOUGLAS-FIR 1 89D WESTERN HEMLOCK 42 FORINTEK CANADA WESTERN LABORATORY VANCOUVER, B.C. TABLE 4. PRINCIPAL COMPONENT ANALYSIS OF MEAN CENTERED DRY SPF DATA FOR 700 SELECTED WAVELENGTHS FROM 60 SAMPLES (S= white spruce, P= lodgepole pine, F= subalpine f i r ) PRINCIPAL COMPONENT NUMBER EIGEN-VALUE PROPORTION OF VARIATION ACCOUNTED FOR CUMULATIVE PROPORTION 1 477. 6 0. 682 0. 682 2 108.8 0.155 0. 838 3 64.8 0.093 0.930 4 26.5 0.038. 0.968 5 7.1 0.010 0.978 6 4.8 0. 007 0.985 7 3 . 3 0. 005 0.990 8 1.4 0.002 0.992 9 1.2 0.002 0.994 10 0.8 0.001 0. 995 11 0.6 0. 001 0.996 12 0.5 0.001 0.997 Note: see also Figure 9. 63 TABLE 5. PRINCIPAL COMPONENT VALUES FOR INDIVIDUAL SAMPLES OF SPF (S= white spruce, P= lodgepole pine, F= subalpine f i r ) SAMPLE SPECIES PRINCIPAL COMPONENT ' NUMBER NAME ID 1 2 3 4 5 88A8 F -1.64 -23.25 13 . 01 6 . 37 -2.29 88A61 F 14.57 -4.35 -12 .26 -0 . 63 0.55 88A66 F 11.24 -10.74 -6 .22 -3 . 37 1.22 89A3 F 33.71 3 .18 -2 . 62 4 .27 -2 .70 89A4 F 8.79 -2.78 -13 .45 0 . 82 2 . 42 89A5 F 35.18 10.62 3 . 08 4 .74 -3 . 39 89A8 F 40.93 -9.57 -7 . 16 3 .94 5.20 89A16 F 7.61 6.45 -9 .58 -2 .50 1.12 89A22 F 45.54 5. 45 -2 .75 14 . 66 6.42 89A25 F 19.14 -5.23 2 .97 3 .94 7.02 89A34 F 2.33 5.74 -7 . 61 1 .83 -1.32 89A40 F 13 .10 7.69 7 . 58 1 . 02 5. 72 89A58 F -3.82 -4.27 -3 .79 -4 . 26 2 . 67 89A60 F -2.03 0.78 1 . 09 -1 . 38 4 . 54 89A61 F 17.61 -2 .32 -3 .24 4 .74 -1.76 89A68 F 7.72 -15.86 31 .45 4 .81 2 .90 89A71 F -5.09 6. 00 -2 . 05 7 . 34 5. 58 89A72 F 9.97 -2 . 59 -6 .58 -1 . 04 0.47 89A73 F 23.76 -10.88 -5 .84 10 . 12 2.22 89A75 F 19.22 1.30 4 .89 4 .75 -2 . 99 88A2 P 19.36 -10.14 0 .58 2 . 03 -1.70 88A7 P 25.89 -1.82 2 .58 3 . 22 -0.94 88A16 P 31.27 6.97 -2 .85 0 .93 -0. 07 88A18 P 28 .18 2 . 87 -3 .46 -2 . 52 0.98 88A24 P 25.52 -2 . 09 0 .20 -2 . 09 1. 00 88A26 P 20.22 -8.57 4 . 37 6 . 11 -4 . 09 88A31 P 22.56 -5.39 -2 .72 2 .23 -0. 17 88A32 P -8.87 -17.19 1 .43 5 .85 -3.67 88A35 P 23.55 -4.82 -2 .97 -0 . 62 -0. 66 88A40 P 19.99 3 .30 9 .81 4 . 19 -2.76 88A47 P 19.21 -11.05 -0 . 16 0 .43 -1.32 88A50 P 31.28 7.27 -0 .01 1 .41 0.25 88A81 P 15.29 -5.95 2 .91 -0 .25 -0.97 88A107 P 24.08 3.55 -7 .32 -3 .45 0.87 88A109 P 18.68 3.43 -2 . 63 0 .86 -1. 85 88A128 P 29.71 -0. 04 -0 .90 -0 . 64 1.49 88A133 P - 17.07 -4.23 -2 .02 -3 .36 0.52 89A9 P 49.21 -0.27 -6 .23 -3 . 13 -3.78 89A24 P 48.22 -2.08 -3 .43 -5 . 24 -2 . 55 89A70 P 8.76 7.58 -4 . 34 -0 . 11 -2.79 88A38 S 25.55 -2.31 -6 .62 -0 . 03 -0. 11 88A64 S 27.95 3.06 0 .64 0 .54 0.82 88A82 S 20.56 -6. 62 8 . 05 -0 .81 -0. 99 89A12 S -8.02 37.58 13 .52 -7 .31 2 . 82 89A15 s 25.19 6. 63 -4 .95 -5 .41 -0.79 89A18 s 17.09 33.42 15 .59 7 .22 -4.45 89A20 s 7.38 -0.87 23 . 59 -8 .70 1.49 89A27 s 7.87 9.30 -0 . 38 -8 . 44 2.53 89A35 s 32 . 51 -9.97 8 .26 17 . 60 0.84 89A36 s - 12.11 13.70 -14 .45 1 .33 -0. 77 64 TABLE 5. (cont.) PRINCIPAL COMPONENT VALUES FOR INDIVIDUAL SAMPLES OF SPF SAMPLE SPECIES PRINCIPAL COMPONENT NUMBER NAME ID 1 2 3 4 5 89A38 S 30. 07 -19.74 -0.46 -6.77 -2 . 60 89A42 S 7. 63 -3 . 65 -1. 10 -5.53 -0. 12 89A45 S 8.06 9. 00 -1.74 1.81 -2 . 64 89A47 S 7.49 6.56 0. 09 -2.21 -2. 08 89A48 s -3.75 -2.73 1.23 -6.77 0. 67 89A49 s -12.51 2 . 53 -7 . 82 2 . 73 -2 . 44 89A51 s 14 .10 6.21 -2.34 -3 . 55 -1. 49 89A54 s 3.22 2.58 -6.05 -4.11 -0. 29 89A56 s -1.41 13 .16 7. 18 2.01 -1. 24 89A67 s 5.74 -14.52 4 . 01 -4.40 -0. 56 TABLE 6. EIGENVECTORS ASSOCIATED WITH INDIVIDUAL WAVELENGTHS FOR SPF (S= white spruce, P= lodgepole pine, F= subalpine f i r ) VARIABLE EIGENVECTORS NUMBER 1 2 3 4 5 6 1 -0. 04505 -0. 01379 -0. 00199 -0. 01095 -0. 01909 -0. 00152 2 -0. 04508 -0. 01305 -0. 00270 -0. 01234 -0. 01746 -0. 00271 3 -0. 04512 -0. 01245 -0. 00316 -0. 01255 -0. 01542 -0. 00365 4 -0. 04516 -0. 01203 -0. 00348 -0. 01230 -0. 01343 -0. 00422 5 -0. 04523 -0. 01129 -0. 00242 -0. 01083 -0. 01965 0. 00379 6 -0. 04522 -0. 01106 -0. 00103 -0. 01059 -0. 02480 0. 00539 7 -0. 04524 -0. 00995 0. 00047 -0. 01123 -0. 02567 0. 00218 8 -0. 04515 -0. 01091 0. 00110 -0. 01123 -0. 03190 0. 00578 9 -0. 04500 -0. 01200 0. 00227 -0. 01085 -0. 03994 0. 00888 10 -0. \u2022 04479 -0. \u2022 01224 0. \u2022 00525 -0. 01086 -0. 04963 0. 00868 690 0. 04185 \u00ab \u2022 0. 03044 -0. 02111 0. 01776 -0. 01394 \u2022 -0. 01624 691 0. 04217 0. 02923 -0. 02094 0. 01629 -0. 01292 -0. 00872 692 0. 04230 0. 02794 -0. 02085 0. 01582 -0. 01019 -0. 01777 693 0. 04226 0. 02819 -0. 02109 0. 01605 -0. 01236 -0. 01817 694 0. 04243 0. 02703 -0. 02117 0. 01525 -0. 01195 -0. 02051 695 0. 04240 0. 02743 -0. 02066 0. 01486 -0. 01298 -0. 01812 696 0. 04262 0. 02667 -0. 02086 0. 01385 -0. 00723 -0. 02097 697 0. 04269 0. 02628 -0. 02063 0. 01349 -0. 00917 -0. 01554 698 0. 04261 0. 02639 -0. 02084 0. 01243 -0. 01499 -0. 01456 699 0. 04264 0. 02575 -0. 02107 0. 01358 -0. 01213 -0. 01647 700 0. 04270 0. 02536 -0. 02085 0. 01402 -0. 01208 -0. 01338 TABLE 7. RESULTS FROM SORTING DRY SPF USING 3 0 WAVELENGTHS (S= white spruce, P= lodgepole pine, F= subalpine f i r ) ANATOMICAL CLASSIFIED AS: ID FIR PINE SPRUCE TOTAL ALL FIR 57 27 15 99 SAMPLES PINE 6 13 3 8 147 SPRUCE 2 15 45 62 TOTAL 65 175 68 308 HEARTWOOD FIR 53 26 13 92 SAMPLES PINE 6 103 8 117 SPRUCE 2 13 38 53 TOTAL 61 142 59 262 SAPWOOD FIR 4 1 2 7 SAMPLES PINE 0 30 0 30 SPRUCE 0 2 7 9 TOTAL 4 33 9 46 PERCENT CORRECTLY CLASSIFIED BY CATEGORY HEART- SAP- TOTAL WOOD WOOD FIR 58 57 58 PINE 88 100 90 SPRUCE 72 78 73 TOTAL 74 89 76 TABLE 8. RESULTS FROM SORTING WET SPF USING 30 WAVELENGTHS (S= white spruce, P= lodgepole pine, F= subalpine f i r ) ANATOMICAL CLASSIFIED AS: ID FIR PINE SPRUCE TOTAL ALL FIR 84 13 4 101 SAMPLES PINE 10 115 22 147 SPRUCE 7 14 41 62 TOTAL 101 142 67 310 HEARTWOOD FIR 79 11 4 94 SAMPLES PINE 3 103 11 117 SPRUCE 6 10 37 53 TOTAL 88 124 52 264 SAPWOOD FIR 5 2 0 7 SAMPLES PINE 7 12 11 30 SPRUCE 1 4 4 9 TOTAL 13 18 15 46 PERCENT CORRECTLY CLASSIFIED BY CATEGORY HEART- SAP- TOTAL WOOD WOOD FIR 84 71 83 PINE 88 40 78 SPRUCE 70 44 66 TOTAL 83 46 77 TABLE 9. RESULTS FROM SORTING WET EXTRACTIVE FREE SPF USING 3 0 WAVELENGTHS (S= white spruce, P= lodgepole pine, F= subalpine f i r ) ANATOMICAL CLASSIFIED AS: ID FIR PINE SPRUCE TOTAL ALL FIR 3 89 4 96 SAMPLES PINE 1 114 0 115 SPRUCE 0 53 0 53 TOTAL 4 256 4 264 % CORRECT 3 99 0 44 67 TABLE 10. RESULTS FROM SORTING WET EASTERN SPF USING 30 WAVELENGTHS (S= black or eastern white spruce, P= red, white or jack pine, F= balsam f i r ) ANATOMICAL CLASSIFIED AS: ID FIR PINE SPRUCE TOTAL ALL FIR 6 1 2 9 SAMPLES PINE 6 19 6 31 SPRUCE 9 12 11 32 TOTAL 21 32 19 72 HEARTWOOD FIR 5 0 1 6 SAMPLES PINE 3 13 1 17 SPRUCE 6 2 9 17 TOTAL 14 15 11 40 SAPWOOD FIR 1 1 1 3 SAMPLES PINE 3 6 5 14 SPRUCE 3 10 2 15 TOTAL 7 17 8 32 PERCENT CORRECTLY CLASSIFIED BY CATEGORY HEART- SAP- TOTAL WOOD WOOD FIR 83 33 67 PINE 76 43 61 SPRUCE 53 13 34 TOTAL 68 28 50 TABLE 11. RESULTS FROM SORTING WET SPF USING 20 WAVELENGTHS (S= white spruce, P= lodgepole pine, F= subalpine f i r ) ANATOMICAL CLASSIFIED AS: ID FIR PINE SPRUCE TOTAL ALL FIR 90 5 6 101 SAMPLES PINE 6 112 29 147 SPRUCE 6 10 46 62 TOTAL 102 127 81 310 HEARTWOOD FIR 87 4 3 94 SAMPLES PINE 1 107 9 117 SPRUCE 5 10 38 53 TOTAL 93 121 50 264 SAPWOOD FIR 3 1 3 7 SAMPLES PINE 5 5 20 30 SPRUCE 1 0 8 9 TOTAL 9 6 31 46 PERCENT CORRECTLY CLASSIFIED BY CATEGORY HEART- SAP- TOTAL WOOD WOOD FIR 93 43 89 PINE 91 17 76 SPRUCE 72 89 74 TOTAL 88 35 80 TABLE 12. RESULTS FROM SORTING WET SPF USING TEN WAVELENGTHS (S= white spruce, P= lodgepole pine, F= subalpine f i r ) ANATOMICAL CLASSIFIED AS: ID FIR PINE SPRUCE TOTAL ALL FIR 79 8 14 101 SAMPLES PINE 3 137 7 147 SPRUCE 3 17 42 62 TOTAL 85 162 63 310 HEARTWOOD FIR 76 7 11 94 SAMPLES PINE 3 110 4 117 SPRUCE 2 14 37 53 TOTAL 81 131 52 264 SAPWOOD FIR 3 1 3 7 SAMPLES PINE 0 27 3 30 SPRUCE 1 3 5 9 TOTAL 4 31 11 46 PERCENT CORRECTLY CLASSIFIED BY CATEGORY HEART- SAP- TOTAL WOOD WOOD FIR 81 43 78 PINE 94 90 93 SPRUCE 70 56 68 TOTAL 84 76 83 TABLE 13. RESULTS FROM SORTING WET EXTRACTIVE FREE SPF USING TEN WAVELENGTHS (S= white spruce, P= lodgepole pine, F= subalpine f i r ) ANATOMICAL CLASSIFIED AS: ID FIR PINE SPRUCE TOTAL ALL FIR 31 0 65 96 SAMPLES PINE 24 3 88 115 SPRUCE 8 2 43 53 TOTAL 63 5 196 264 % CORRECT 32 3 81 29 70 TABLE 14. RESULTS FROM SORTING WET EASTERN SPF USING TEN WAVELENGTHS (S= black or eastern white spruce, P= red, white or jack pine, F= balsam f i r ) ANATOMICAL CLASSIFIED AS: ID FIR PINE SPRUCE TOTAL ALL FIR 7 1 1 9 SAMPLES PINE 1 17 13 31 SPRUCE 1 13 18 32 TOTAL 9 31 32 72 HEARTWOOD FIR 6 0 0 6 SAMPLES PINE 1 9 7 17 SPRUCE 1 3 13 17 TOTAL 8 12 20 40 SAPWOOD FIR 1 1 1 3 SAMPLES PINE 0 8 6 14 SPRUCE 0 10 5 15 TOTAL 1 19 12 32 PERCENT CORRECTLY CLASSIFIED BY CATEGORY HEART- SAP- TOTAL WOOD WOOD FIR 100 33 78 PINE 53 57 55 SPRUCE 76 33 56 TOTAL 70 44 58 TABLE 15. RESULTS FROM SORTING WET SPF USING SIX WAVELENGTHS (S= white spruce, P= lodgepole pine, F= subalpine f i r ) ANATOMICAL CLASSIFIED AS: ID FIR PINE SPRUCE TOTAL ALL FIR 69 13 19 101 SAMPLES PINE 2 139 6 147 SPRUCE 3 16 43 62 TOTAL 74 168 68 310 HEARTWOOD FIR 66 12 16 94 SAMPLES PINE 2 110 5 117 SPRUCE 3 14 36 53 TOTAL 71 136 57 264 SAPWOOD FIR 3 1 3 7 SAMPLES PINE 0 29 1 30 SPRUCE 0 2 7 9 TOTAL 3 32 11 46 PERCENT CORRECTLY CLASSIFIED BY CATEGORY HEART- SAP- TOTAL WOOD WOOD FIR 70 43 68 PINE 94 97 95 SPRUCE 68 78 69 TOTAL 80 85 81 72 TABLE 16. PRINCIPAL COMPONENT ANALYSYIS OF MEAN CENTERED DRY WESTERN LARCH\/DOUGLAS-FIR DATA: A) FIRST 460 OF 931 SELECTED WAVELENGTHS FROM 20 SAMPLES PRINCIPAL COMPONENT NUMBER EIGEN-VALUE PROPORTION OF VARIATION ACCOUNTED FOR CUMULATIVE PROPORTION 1 2 3 4 5 6 7 8 9 10 1210, 228, 39, 10, 6, 4 , 2 , 1, 1. 0.803 0.151 026 007 004 003 002 001 001 001 803 954 980 987 991 0.994 0.996 997 998 999 0, 0. 0, Note: see also Figure 21. B) LAST 471 OF 931 SELECTED WAVELENGTHS FROM 20 SAMPLES PRINCIPAL EIGEN- PROPORTION CUMULATIVE COMPONENT VALUE OF VARIATION PROPORTK NUMBER ACCOUNTED FOR 1 547.9 0.762 0.762 2 124.7 0.174 0.936 3 24.5 0. 034 0. 970 4 11. 0 0. 015 0.985 5 2 . 6 0. 004 0.989 6 2 .1 0. 003 0.992 7 1.4 0.002 0.994 8 0.9 0.001 0.995 9 0.8 0. 001 0. 996 10 0. 001 0.997 Note: see also Figure 22. 73 TABLE 17. PRINCIPAL COMPONENT ANALYSIS OF MEAN CENTERED DRY WESTERN LARCH\/DOUGLAS-FIR DATA USING 160 WAVELENGTHS SELECTED ON THE BASIS OF EARLIER PRINCIPAL COMPONENT ANALYSES PRINCIPAL COMPONENT NUMBER EIGEN-VALUE PROPORTION OF VARIATION ACCOUNTED FOR CUMULATIVE PROPORTION 1 2 3 4 5 6 7 8 9 10 304 , 65, 37 , 5, 4 , 3. 2, 0, 0.5 0. 718 0.154 0.088 013 O i l 007 0. 005 0. 001 001 001 0, 0, 0, 0, 0, 0. 718 0. 872 0.960 ,972 983 990 0. 995 0.996 0.997 0.998 0, 0. 0, Note: see also Figure 23. TABLE 18. RESULTS FROM SORTING DRY WESTERN LARCH\/DOUGLAS-FIR USING THE FIRST 18 OF 40 WAVELENGTHS ANATOMICAL CLASSIFIED AS: ID D-FIR LARCH TOTAL ALL D-FIR 30 4 34 SAMPLES LARCH 0 158 158 TOTAL 30 162 192 HEARTWOOD D-FIR 19 2 21 SAMPLES LARCH 0 89 89 TOTAL 19 91 110 SAPWOOD D-FIR 11 2 13 SAMPLES LARCH 0 69 69 TOTAL 11 71 82 PERCENT CORRECTLY CLASSIFIED BY CATEGORY HEART- SAP- TOTAL WOOD WOOD D-FIR 90 85 88 LARCH 100 100 100 TOTAL 98 98 98 74 TABLE 19. RESULTS FROM SORTING DRY WESTERN LARCH\/DOUGLAS-FIR USING THE FIRST 3 8 OF 40 WAVELENGTHS ANATOMICAL CLASSIFIED AS: ID D-FIR LARCH TOTAL ALL D-FIR 24 10 34 SAMPLES LARCH 19 139 158 TOTAL 43 149 192 HEARTWOOD D-FIR 17 4 21 SAMPLES LARCH 6 83 89 TOTAL 23 87 110 SAPWOOD D-FIR 7 6 13 SAMPLES LARCH 13 56 69 TOTAL 20 62 82 PERCENT CORRECTLY CLASSIFIED BY CATEGORY HEART- SAP- TOTAL WOOD WOOD D-FIR 81 54 71 LARCH 93 81 88 TOTAL 91 77 85 TABLE 20. RESULTS FROM SORTING WET WESTERN LARCH\/DOUGLAS-FIR USING THE FIRST 18 OF 40 WAVELENGTHS ANATOMICAL CLASSIFIED AS: ID D-FIR LARCH TOTAL ALL D-FIR 20 14 34 SAMPLES LARCH 22 136 158 TOTAL 42 150 192 HEARTWOOD D-FIR 14 7 21 SAMPLES LARCH 1 88 89 TOTAL 15 95 110 SAPWOOD D-FIR 6 7 13 SAMPLES LARCH 21 48 69 TOTAL 27 55 82 PERCENT CORRECTLY CLASSIFIED BY CATEGORY HEART- SAP- TOTAL WOOD WOOD D-FIR 67 46 59 LARCH 99 70 86 TOTAL 93 66 81 75 TABLE 21. RESULTS FROM SORTING WET WESTERN LARCH\/DOUGLAS-FIR USING THE FIRST 14 OF 40 WAVELENGTHS ANATOMICAL CLASSIFIED AS: ID D-FIR LARCH TOTAL ALL D-FIR 30 4 34 SAMPLES LARCH 16 142 158 TOTAL 46 146 192 HEARTWOOD D-FIR 19 2 21 SAMPLES LARCH 3 86 89 TOTAL 22 88 110 SAPWOOD D-FIR 11 2 13 SAMPLES LARCH 13 56 69 TOTAL 24 58 82 PERCENT CORRECTLY CLASSIFIED BY CATEGORY HEART- SAP- TOTAL WOOD WOOD D-FIR 90 85 88 LARCH 97 81 90 TOTAL 95 82 90 TABLE 22. RESULTS FROM SORTING WET WESTERN LARCH\/DOUGLAS-FIR USING THE SECOND 14 OF 40 WAVELENGTHS ANATOMICAL CLASSIFIED AS: ID D-FIR LARCH TOTAL ALL D-FIR 23 11 34 SAMPLES LARCH 19 139 158 TOTAL 42 150 192 HEARTWOOD D-FIR 16 5 21 SAMPLES LARCH 6 83 89 TOTAL 22 88 110 SAPWOOD D-FIR 7 6 13 SAMPLES LARCH 13 56 69 TOTAL 20 62 82 PERCENT CORRECTLY CLASSIFIED BY CATEGORY HEART- SAP- TOTAL WOOD WOOD D-FIR 76 54 68 LARCH 93 81 88 TOTAL 90 77 84 76 TABLE 23. RESULTS FROM SORTING WET WESTERN LARCH\/DOUGLAS-FIR USING THE LAST 12 OF 4 0 WAVELENGTHS ANATOMICAL CLASSIFIED AS: ID D-FIR LARCH .TOTAL ALL D-FIR 29 5 34 SAMPLES LARCH 13 145 158 TOTAL 42 150 192 HEARTWOOD D-FIR 20 1 21 SAMPLES LARCH 10 79 89 TOTAL 30 80 110 SAPWOOD D-FIR 19 5 24 SAMPLES LARCH 13 135 148 TOTAL 32 140 172 PERCENT CORRECTLY CLASSIFIED BY CATEGORY HEART- SAP- TOTAL WOOD WOOD D-FIR 95 79 85 LARCH 89 91 92 TOTAL 90 90 91 77 TABLE 24. PRINCIPAL COMPONENT ANALYSIS OF MEAN CENTERED DRY WESTERN HEMLOCK\/SITKA SPRUCE\/AMABILIS FIR DATA: A) FIRST 542 OF 1085 SELECTED WAVELENGTHS FROM 30 SAMPLES PRINCIPAL COMPONENT NUMBER EIGEN-VALUE PROPORTION OF VARIATION CUMULATIVE PROPORTION ACCOUNTED FOR 1 1595. 5 0. 840 0. 840 2 175. 7 0. 093 0. 933 3 46. 6 0. 025 0. 957 4 30. 0 0. 016 0. 973 5 12. 4 0. 007 0. 980 6 11. 5 0. 006 0. 986 7 8. 0 0. 004 0. 990 8 6. 7 0. 004 0. 994 9 5. 1 0. 003 0. 996 10 1. 7 0. 001 0. 997 Note: see also Figure 25. B) LAST 543 OF 1085 SELECTED WAVELENGTHS FROM 30 SAMPLES PRINCIPAL COMPONENT NUMBER EIGEN-VALUE PROPORTION OF VARIATION CUMULATIVE PROPORTION ACCOUNTED FOR 1 499. 4 0. 640 0. 640 2 190. 4 0. 244 0. 884 3 47. 8 0. 061 0. 945 4 27. 5 0. 035 0. 980 5 \u2022 6. 2 0. 008 0. 988 6 2. 1 0. 003 0. 991 7 1. 4 0. 002 0. 993 8 1. 1 0. 001 0. 994 9 1. 0 0. 001 0. 995 10 0. 6 0. 001 0. 996 Note: see also Figure 26. 78 TABLE 25. PRINCIPAL COMPONENT ANALYSIS OF MEAN CENTERED DRY WESTERN HEMLOCK\/SITKA SPRUCE\/ AMABILIS FIR DATA USING 160 WAVELENGTHS SELECTED ON THE BASIS OF EARLIER PRINCIPAL COMPONENT ANALYSES PRINCIPAL EIGEN- PROPORTION CUMULATIVE COMPONENT VALUE OF VARIATION PROPORTION NUMBER ACCOUNTED FOR 1 434 . 3 0.736 0. 736 2 84. 8 0.144 0. 879 3 29. 6 0. 050 0. 929 4 10. 8 0. 018 0. 948 5 10. 3 0. 017 0. 965 6 7. 4 0. 012 0. 977 7 4. 7 0. 008 0. 985 8 3 . 4 0.006 0. 991 9 1. 9 0. 003 0. 994 10 1. 1 0.002 0. 996 Note: see also Figure 27. TABLE 26. RESULTS FROM SORTING DRY WESTERN HEMLOCK\/SITKA SPRUCE\/ AMABILIS FIR USING 19 WAVELENGTHS ANATOMICAL CLASSIFIED AS: ID FIR SPRUCE HEMLOCK TOTAL ALL FIR 13 2 1 16 SAMPLES SPRUCE 2 49 7 58 HEMLOCK 13 3 76 92 TOTAL 28 54 84 166 SAPWOOD FIR 0 0 0 0 SAMPLES SPRUCE 0 0 4 4 HEMLOCK 0 0 5 5 TOTAL 0 0 9 9 HEARTWOOD FIR 13 2 1 16 SAMPLES SPRUCE 2 49 3 54 HEMLOCK 13 3 71 87 TOTAL 28 54 75 157 PERCENT CORRECTLY CLASSIFIED BY CATEGORY HEART- SAP- TOTAL WOOD WOOD FIR 81 no sample 81 SPRUCE 91 0 84 HEMLOCK 82 100 83 TOTAL 85 56 83 TABLE 27. RESULTS FROM SORTING DRY WESTERN HEMLOCK\/SITKA SPRUCE\/ AMABILIS FIR USING TEN WAVELENGTHS ANATOMICAL CLASSIFIED AS: ID FIR SPRUCE HEMLOCK TOTAL ALL FIR 12 3 1 16 SAMPLES SPRUCE 9 45 4 58 HEMLOCK 25 7 60 92 TOTAL 46 55 65 166 SAPWOOD FIR 0 0 0 0 SAMPLES SPRUCE 1 0 3 4 HEMLOCK 0 0 5 5 TOTAL 1 0 8 9 HEARTWOOD FIR 12 3 1 16 SAMPLES SPRUCE 8 45 1 54 HEMLOCK 25 7 55 87 TOTAL 45 55 57 157 PERCENT CORRECTLY CLASSIFIED BY CATEGORY HEART- SAP- TOTAL WOOD WOOD FIR 75 no sample 75 SPRUCE 83 0 78 HEMLOCK 63 100 65 TOTAL 71 56 70 TABLE 28. RESULTS FROM SORTING WET WESTERN HEMLOCK\/SITKA SPRUCE\/ AMABILIS FIR USING 15 WAVELENGTHS ANATOMICAL CLASSIFIED AS: ID FIR SPRUCE HEMLOCK TOTAL ALL FIR 10 4 2 16 SAMPLES SPRUCE 11 40 7 58 HEMLOCK 25 5 62 92 TOTAL 46 49 71 166 SAPWOOD FIR 0 0 0 0 SAMPLES SPRUCE 0 0 4 4 HEMLOCK 1 0 4 5 TOTAL 1 0 8 9 HEARTWOOD FIR 10 4 2 16 SAMPLES SPRUCE 11 40 3 54 HEMLOCK 24 5 58 87 TOTAL 45 49 63 157 PERCENT CORRECTLY CLASSIFIED BY CATEGORY HEART- SAP- TOTAL WOOD WOOD FIR 63 no sample 63 SPRUCE 74 0 69 HEMLOCK 67 80 67 TOTAL 69 44 67 TABLE 29. RESULTS FROM SORTING WET WESTERN HEMLOCK\/SITKA SPRUCE\/ AMABILIS FIR USING TEN WAVELENGTHS ANATOMICAL CLASSIFIED AS: ID FIR SPRUCE HEMLOCK TOTAL ALL FIR 9 3 4 16 SAMPLES SPRUCE 22 35 1 58 HEMLOCK 19 13 60 92 TOTAL 50 51 65 166 SAPWOOD FIR 0 0 0 0 SAMPLES SPRUCE 3 1 0 4 HEMLOCK 2 0 3 5 TOTAL 5 1 3 9 HEARTWOOD FIR 9 3 4 16 SAMPLES SPRUCE 19 34 1 54 HEMLOCK 17 13 57 87 TOTAL 45 50 62 157 PERCENT CORRECTLY CLASSIFIED BY CATEGORY HEART- SAP- TOTAL WOOD WOOD FIR 56 no sample 56 SPRUCE 63 25 60 HEMLOCK 66 60 65 TOTAL 64 44 63 TABLE 30. RESULTS FROM SORTING WET WESTERN HEMLOCK\/SITKA SPRUCE USING 15 WAVELENGTHS ANATOMICAL CLASSIFIED AS: ID SPRUCE HEMLOCK TOTAL ALL SPRUCE 51 7 58 SAMPLES HEMLOCK 20 72 92 TOTAL 71 79 150 SAPWOOD SPRUCE 2 2 4 SAMPLES HEMLOCK 0 5 5 TOTAL 2 7 9 HEARTWOOD SPRUCE 49 5 54 SAMPLES HEMLOCK 20 67 87 TOTAL 69 72 141 PERCENT CORRECTLY CLASSIFIED BY CATEGORY HEART- SAP- TOTAL WOOD WOOD SPRUCE 91 50 88 HEMLOCK 77 100 78 TOTAL 82 78 82 Table 31. RESULTS FROM SORTING WET WESTERN HEMLOCK\/SITKA SPRUCE USING TEN WAVELENGTHS ANATOMICAL CLASSIFIED AS: ID SPRUCE HEMLOCK TOTAL ALL SPRUCE 48 10 58 SAMPLES HEMLOCK 17 75 92 TOTAL 65 85 150 SAPWOOD SPRUCE 2 2 4 SAMPLES HEMLOCK 2 3 5 TOTAL 4 5 9 HEARTWOOD SPRUCE 46 8 54 SAMPLES HEMLOCK 15 72 87 TOTAL 61 80 141 PERCENT CORRECTLY CLASSIFIED BY CATEGORY HEART- SAP- TOTAL WOOD WOOD SPRUCE 85 50 83 HEMLOCK 83 60 82 TOTAL 84 56 82 FIGURE 1. TYPICAL SINGLE-BEAM REFLECTANCE SPECTRUM FOR WET LODGEPOLE PINE WOOD. 83 3.0 2.0 + 4850 3850 2850 WAVE NUMBER (cm 1850 1 ) 850 o c 70 m rO o 7\\ CD 70 TJ m F 2 r-U) z o I\u2014 m I CD 70 m m o o m tn u m o oo FIGURE 4. REGION OF SPECTRUM USED FOR DIFFERENTIATION OF SPECIES (TYPICAL LODGEPOLE PINE SHOWN). 86 CM 30NV1031J3d \/ 10 \u2022 \u2022 \u2022 SPRUCE - -P INE FIR 0 2500 \u2014 I \u2014 2000 \u2014 I \u2014 1500 \u2014 I \u2014 1000 o c m cn 500 WAVE NUMBER ( c m - 1 ) 00 WAVE NUMBER ( c m \" 1 ) 00 00 7. AVERAGE SPECTRA \u00b1 1 STANDARD DEVIATION (n=10) FOR WHITE SPRUCE, LODGEPOLE PINE AND SUBALPINE FIR. S 00 n 00 \u2022 \u2022 \u2022 <- o 33NV1031J3cJ \/ 1 FIGURE 8. TYPICAL MEAN CENTERED REFLECTANCE S P E C T R U M FOR WET LODGEPOLE PINE WOOD. m ^ - r O C N j T - O t - c M ro CN I I I 33NV1031J3d \/1 FIGURE 9. EIGENVALUES FOR FIRST 12 PRINCIPAL COMPONENTS OF WHITE SPRUCE\/ LODGEPOLE PINE\/ SUBALPINE FIR MIXTURE USING 700 SELECTED WAVELENGTHS. O O O O O O O O O O O m ^ - ro CM 3IY1VAN30I3 FIGURE 10. PRINCIPAL COMPONENT SCORES FOR FIRST AND SECOND PRINCIPAL COMPONENTS OF SAMPLES IN WHITE SPRUCE\/ LODGEPOLE PINE\/ SUBALPINE FIR MIX. LxJ C J Li_ Q_ CO O O \u2022 \u2022 O l \u2022 <>\u2022 < \u2022 o n 8 o _ a o \u2022 LxJ O CL O o _ J < CL o on Q_ (fi !N3N0dW00 IVdIONIdd p u Z 93 FIGURE 11 . PRINCIPAL COMPONENT SCORES FOR FIRST AND THIRD PRINCIPAL COMPONENTS OF SAMPLES IN WHITE SPRUCE\/ LODGEPOLE PINE\/ SUBALPINE FIR MIX. LxJ LxJ O U_ Q_ CO O O \u2022 OO \u2022 9 \u00b0 o \u2022 JpO < > o o LxJ Z o Q_ O O I < \u2022 L O CL \u2022 Ol o o CD O \u00a3<>o (1& < b 0 D @ 0 CO lN3N0dW00 IVdIONIdd pj\u00a3 94 FIGURE 12. PRINCIPAL COMPONENT SCORES FOR FIRST AND FOURTH PRINCIPAL COMPONENTS OF SAMPLES IN WHITE SPRUCE\/ LODGEPOLE PINE\/ SUBALPINE FIR MIX. !N3N0drM00 IVdIONIdd qi* FIGURE 1 3. PRINCIPAL COMPONENT SCORES FOR FIRST AND FIFTH PRINCIPAL COMPONENTS OF SAMPLES IN WHITE S P R U C E \/ LODGEPOLE P I N E \/ SUBALPINE FIR MIX. LU CJ ZD U_ Q_ CO O O \u2022 o o o \u2022 o o \u2022 \u2022 o o \u2022 \u2022 o f\u2014 z LU Z O 0_ o o _ J < D_ O CO o o o \u2022 \u2022 o 8 rr\u00a9~ rJ^o o !N3N0dlAJ00 IVdIONIdd M iS 96 FIGURE 14. PRINCIPAL COMPONENT SCORES FOR SECOND AND THIRD PRINCIPAL COMPONENTS OF SAMPLES IN WHITE SPRUCE\/ LODGEPOLE PINE\/ SUBALPINE FIR MIX. LU O Lu CL GO O O \u2022 \u2022 \u2022 \u2022 O O p o \u2022 \u2022 ft \u2022 o o \u2022 oo -L LU o O O _ J < Q_ O (Z CL \"O c CM !N3N0dW00 IVdIONIdd D J \u00a3 FIGURE 15. PRINCIPAL COMPONENT SCORES FOR SECOND AND FOURTH PRINCIPAL COMPONENTS OF SAMPLES IN WHITE SPRUCE\/ LODGEPOLE PINE\/ SUBALPINE FIR MIX. 1 LU LU O Z ) Li_ D_ CO O O \u2022 O \u2022 o CP 9d> o o o o o \u2022 \u2022 0LJ \u2022 ef t T J o \u2022 \u2022 \u2022 + TD C CN _LN3N0dn03 IVdIONIdd u i > FIGURE 16 PRINCIPAL COMPONENT SCORES FOR SECOND AND FIFTH PRINCIPAL COMPONENTS OF SAMPLES IN WHITE SPRUCE\/ LODGEPOLE PINE\/ SUBALPINE FIR MIX. LU O Li_ D_ CO O O \u2022 \u2022 o o o \u2022 \u2022 b q $ o
6b 0[3> \u2022 \u2022 LU O O C J _ J < O E \"D INGNOdWOO IVdIONIdd q i t FIGURE 1 8. PRINCIPAL COMPONENT SCORES FOR THIRD AND FIFTH PRINCIPAL COMPONENTS OF SAMPLES IN WHITE SPRUCE\/ LODGEPOLE PINE\/ SUBALPINE FIR MIX. LU CJ Lu CL CO O O \u2022 \u2022 \u2022 \u2022 oo o o o u ^ \u2022 o \u2022 \u2022 lN3N0dW00 IVdlONIdd qiS 1 0 1 F I G U R E 1 9 . P R I N C I P A L C O M P O N E N T S C O R E S F O R F O U R T H A N D F I F T H P R I N C I P A L C O M P O N E N T S O F S A M P L E S I N W H I T E S P R U C E \/ L O D G E P O L E P I N E \/ S U B A L P I N E F I R M I X . I\u2014 L U ! Z O D _ ^ O o _ J < O L O lN3N0dH00 IVdIONIdd M i 5 102 FIGURE 20. TYPICAL MEAN CENTERED REFLECTANCE SPECTRUM FOR WET LODGEPOLE PINE WOOD SHOWING LOCATION OF 30 WAVELENGTHS USED FOR DISCRIMINANT ANALYSIS O O LO < o o CN \/ O O OO CO ^d- CM O CM ^ \" CN 30NV1031J3cJ A 103 FIGURE 2 1 . EIGENVALUES FOR FIRST NINE PRINCIPAL COMPONENTS OF WESTERN LARCH\/ DOUGLAS-FIR MIXTURE USING FIRST 460 OF 931 SELECTED WAVELENGTHS. 3 n i V A N 3 9 l 3 104 FIGURE 22. EIGENVALUES FOR FIRST NINE PRINCIPAL COMPONENTS OF WESTERN LARCH\/ DOUGLAS-FIR MIXTURE USING LAST 471 OF 931 SELECTED WAVELENGTHS. O 3ITIVAN30I3 105 FIGURE 23. EIGENVALUES FOR FIRST NINE PRINCIPAL COMPONENTS OF WESTERN LARCH\/ DOUGLAS-FIR MIXTURE USING 1 60 WAVELENGTHS SELECTED ON BASIS OF PREVIOUS ANALYSES. 3 m V A N 3 0 G 106 FIGURE 24. TYPICAL MEAN CENTERED REFLECTANCE SPECTRUM FOR WET DOUGLAS-FIR WOOD SHOWING LOCATION OF 40 WAVELENGTHS USED FOR DISCRIMINANT ANALYSIS. 30NV1031J3cJ \/ t 107 FIGURE 25. EIGENVALUES FOR FIRST TEN PRINCIPAL COMPONENTS OF WESTERN HEMLOCK\/ SITKA SPRUCE\/ AMABILIS FIR MIXTURE USING FIRST 542 OF 1085 SELECTED WAVELENGTHS. _ 0 =tfc LU o Q_ O o _ l < Q_ O D_ 3IT1VAN30I3 108 FIGURE 26. EIGENVALUES FOR FIRST TEN PRINCIPAL COMPONENTS OF WESTERN HEMLOCK\/ SITKA SPRUCE\/ AMABILIS FIR MIXTURE USING LAST 543 OF 1085 SELECTED WAVELENGTHS. ^ O cn CO CO LO CM LU o CL o o _ l < O Q_ 3 n i V A N 3 0 G 109 FIGURE 27. EIGENVALUES FOR FIRST TEN PRINCIPAL COMPONENTS OF WESTERN HEMLOCK\/ SITKA SPRUCE\/ AMABILIS FIR MIXTURE USING 171 WAVELENGTHS SELECTED ON BASIS OF PREVIOUS ANALYSES. , O 3 n i V A N 3 0 l 3 110 FIGURE 28. TYPICAL MEAN CENTERED REFLECTANCE SPECTRUM FOR DRY SITKA SPRUCE WOOD SHOWING LOCATION OF 20 WAVELENGTHS USED FOR DISCRIMINANT ANALYSIS. u i ^ - n CN ^ O r - CN CN 3 0 N V 1 3 3 ~ L G c d \/ l ","@language":"en"}],"Genre":[{"@value":"Thesis\/Dissertation","@language":"en"}],"IsShownAt":[{"@value":"10.14288\/1.0100448","@language":"en"}],"Language":[{"@value":"eng","@language":"en"}],"Program":[{"@value":"Forestry","@language":"en"}],"Provider":[{"@value":"Vancouver : University of British Columbia Library","@language":"en"}],"Publisher":[{"@value":"University of British Columbia","@language":"en"}],"Rights":[{"@value":"For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https:\/\/open.library.ubc.ca\/terms_of_use.","@language":"en"}],"ScholarlyLevel":[{"@value":"Graduate","@language":"en"}],"Title":[{"@value":"Differentiation of some Canadian coniferous woods by combined diffuse and specular reflectance Fourier transform infrared spectrometry","@language":"en"}],"Type":[{"@value":"Text","@language":"en"}],"URI":[{"@value":"http:\/\/hdl.handle.net\/2429\/30788","@language":"en"}],"SortDate":[{"@value":"1989-12-31 AD","@language":"en"}],"@id":"doi:10.14288\/1.0100448"}