Transcription
PREDICTING INTERNAL YELLOW-POPLAR LOG DEFECT FEATURESUSING SURFACE INDICATORSR. Edward Thomas*USDA Forest ServiceNorthern Researh StationPrinceton, WV 21740(Received July 2007)ABSTRACTDetermining the defects that are located within the log is crucial to understanding the treeAog resourcefor efficient processing. However, existing means of doing this non-destructively requires the use ofexpensive X-raylCT, MRI, or microwave technology. These methods do not lend themselves to fast,efficient, and cost-effective analysis of logs and tree stems in the mill. This study quantified the relationship between external defect indicators and internal defect characteristics for yellow-poplar logs. Aseries of models were developed to predict internal features using visible external features, log diameter,indicator width, length, and rise. Good correlations and small prediction errors were observed with sound(sawn), overgrown, and unsound h o t defects. For less severe defects such as adventitious budslclustersand distortion type defects weaker correlations were observed, but the magnitude of prediction errors wassmall and acceptable.Keywords.Yellow-poplar, hardwood log, defect modeling, internal defect prediction, external indicator.INTRODUCTIONOne of the major emphasis areas today inhardwood research is the development of equipment and a methodology that can accuratelysense internal defect locations and structures.Determining the location and characteristics ofdefects located inside logs promises to dramatically improve log recovery in terms of bothquantity and quality (Steele et al. 1994). In addition, accurate internal defect informationwould permit researchers to analyze, refine, andexpand log grading rules, multi-product potential, stand differences, and the impact of silvicultural treatments on quality in ways previouslynot available or economically feasible. The goalof this research is to provide a mathematicalmethod of predicting internal defect size and location based on external surface indicators. Thisstudy was limited to the most common defecttypes that are those that have the greatest impacton hardwood quality (tree or lumber grade).Although there are many benefits to determin-":Col espondingauthor: [email protected] artd Fiber Science,40(1). 2008, pp. 14-220 2008 by the Society of Wood Science and Technologying internal log information, an inexpensive andefficient method of obtaining these data does notexist today. Researchers are currently examiningvarious approaches to this problem including theuse of X-ray/CT (computerized tomography),ultrasound, MRI (Magnetic Resonance Imaging), or radar technology (Chang 1992). Somehigh-volume softwood lumber mills in Europeand the Pacific Northwest have installed multiple-head X-ray scanners. This type of scanneruses 3 or 4 X-ray transmitters and detectors tocapture internal log defect data. Although thesetypes of scanners operate much faster, they obtain lower quality/resolution data than do CTscanners. Although X-ray/CT and MRI methodsshow promise, the technology is expensive,slow, and does not permit fast, efficient analysisof logs and tree stems. Both CT and three-headX-ray systems are still limited to being used onsmaller-diameter logs due to energy level issues.During the past 50 yr there has been a significant amount of research conducted examiningthe relationship of external hardwood log defectindicators to internal defect characteristics. Several guides and pictorial series have been pub-
Tlzonzas-PREDICTING.YELLOW-POPLAR LOG DEFECTS USING SURFACE INDICATORSlished illustrating various externallinternal defect characteristics and their relationship forvarious hardwood species (Marden and Stayton1970; Rast 1982; Rast et al. 1985, 1988, 1989,1990a, 1990b, 1990 ).While these guides areuseful references for providing insight on theexternallinternal relationship, only one or twoexainples of each defect type are provided. Thus,they do not fulfill the need of a definitive modelcapable of predicting internal defect featuresbased on observable external defect features.Researchers have studied the relationshipsamong surface indicators and internal derectmanifestation in depth for various hardwood andsoftwood species. Schultz (1961) exanlined German beech and found that for this species theratio of the bark distortion width to length is thesame as the ratio of the stem when the branch iscompletely healed over to the current stem dianleter. However, for species with heavier irregular bark, like hard maple, he found that itwas difficult to judge the clear area above thedefect in this manner.Hyvannen (1976) used Stayton et al. (1970)maple defect data to explore the relationshipsamong the internal features of grain orientationand height of clear wood above an encapsulatedknot defect and the external features of surfacerise, width, and length. The sugar maple defectdata were collected from 44 trees obtained fromthree sites in upper Michigan. Hyvarinen usedsimple linear regression methods to find goodcorrelations among clear wood above defects,bark distortion width, length, and rise measurements, as well as age, tree diameter, and stemtaper. The best simple correlation was with diameter inside bark (DIB) (r 0.66) and an 18im standard error of estimate. Colrelation wasfurther improved by using a stepwise regressionmethod. The final model (r 0.74) employedbark distortion vertical size and DIE as the mostsignificant predictor variables.A similar study was conducted on a sample of21 black spruce trees collected from a naturalstand 75 km north of Quebec City (Lemieux etal. 2001). Three trees, each with three logs, wereselected from which a total of 249 knot defectswere dissected and their data recorded. The re-15searchers found better correlations between external indicator and internal characteristics in themiddle and bottom logs as compared to the upper logs. Strong coil-elations (r .89) among thelength and width of internal defect zones andexternal features such as branch stub diameteran3 Tenith were found to exist. The defects weremodeled as-having three distinct zones conesponding to the manner in which the penetrationangle changes over time in black spruce. Thepenetration angle is the angle at which a linetl roughthe center of the defect intersects the logsurface. This study exanlined only branches thathad not been dropped nor pruned and thus couldnot exailline encapsulation depth.Carpenter (1950) found that although the frequency and pccurrence of surface indicatorswithin a given species vary by region, in generalthe same indicator will be found with its defectin the underlying wood. Thus, although certaindefect types may be more prevalent in some regions, the underlying manifestation of the defectwould remain more or less consistent across regions. Further, growth rate will vary from regionto region (or site to site within the same region),thus the defect encapsulation rate will differalso. However, the rate at which the encapsulation occurs, and the degree to which the defect isoccluded or covered over by clear wood is indicated in the bark pattern. Shigo and Larson(1969) discovered that the ratio of defect heightto-width indicates the depth of the defect withrespect to the radius of the stem at the defect(Fig. 1). The faster the diameter growth, thefaster the defect is encapsulated, and thus thefaster the bark distortion pattern changes.MATERIALS AND METHODSSanzple collectioizYellow-poplar defect samples were collectedfrom two sites in West Virginia: West VirginiaUniversity Forest (WVUF) near Morgantown(elevation: 700 in) and Camp Creek State Forest(CCSF) near Princeton (elevation: 790 m). Thetwo sites are separated by approximately 350krn. From each site, 33 trees were randoinly se-
16WOOD AND FIBER SCIENCE, JANUARY 2008, V. 40(1)knot was selected. Of course, not all trees had 4defects of every type. In other cases, selectingone defect would prevent another from beingselected due to defect overlap. Defects are overRelationship of rtub porilapping when the process of buclung the defecttion to rhope of rtub scar.from the log and processing it would destroy allor part of another. In these cases, preferencewent to the defect type that was more numerouson that tree and a different occurrence of theother defect selected.1-2The number of defect sanlples obtained fromeach site by defect type is shown in Table 1. Inimost cases, approximately equal numbers ofeach type of defect were obtained from each site.1-3The exceptions are the light distortion (LD) andunsound knot (UK) defects. UK defects oc2 curred in much fewer numbers at CCSF than atWVUF. More LD defects were found on CCSFFIG. 1. Encapsulation depth and stub scar relationship Sanlples than the olles from WVUF.1IIIrn qI- "ratio.lected. For each tree, the number of defects bytype were counted. The types of defects identified, counted, and analyzed in this study arelisted in Table 1. These counts were used todevelop a random sampling plan. The goal wasto collect 4 defects of each type from each tree,whenever possible. For example, if there were 8sound knots on the tree, every%,second soundITABLE1. Types and 7zwzbel-s of defects collected by siteand overall.LocationDefect nameDefectabbreviationAdventitious knotAdventitious knot clusterBumpHeavy distortionLight distortionMedium distortionOvergrown knotOvergrown knot clusterSound knotSound knot clusterUnsound Uforest756659683174609869179877620. 9271004Sanzple-processingAll defects were identified according to thecharacteristics as defined in Defects irz Hardwood Tinzber (Carpenter et al. 1989). Once adefect was located and classified, the sectioncontaining the defect was cut from the log. Thedefect sections ranged from 305 to 310 mm inlength. If we discovered during slicing that partof the interior defect was not completely contained within the section, the sample was discarded. An alignment groove was cut into thetop of the section in a line between the indicatorand the pith. This provided a smooth area foreasier ring counts and a common point of reference for each slice for collection positionloffsetmeasurements. An identifying tag was stapled tothe section surface and the defect indicator wasdigitally photographed (Fig. 2). The tag identifies the source tree and log, defect type, defectnumber, and height up the stem.For each sample the following informationwas recorded: defect type, surface width (acrossgrain) and length (along grain), growth rate, barkthickness, and surface height rise, if any. Thesample was then sawn into 25-mm-thick slices at90" to the reference notch. This resulted in aseries showing the defect penetrating the log
Thonzas-PREDICTINGFIG.2.groove.17YELLOW-POPLAR L(3G DEFECTS USING SURFACE INDICATORSDefect section with tagged defect and reference(Fig. 3). For each slice the depth, defect width,length, and distance of defect center to notchbottom center were recorded. When a defect terminated between slices, it was assumed that itterminated at the halfway point through theslice.Modeling statisticsA series of chi-squared tests were used to testfor outliers in the internallexternal data set(Komsta 2006). Data identified by the tests asoutliers were examined and corrected if in error.The data were grouped by defect type. Using thestatistics program "R," stepwise muliple-linearregression analyses were used to test for correlations anlong surface indicators and internalfeatures (R Development Core Team 2006). Theindependent variables used were surface indicator width (SWID), length (SLEN), rise (SRISE),and log diameter inside bark (DIB). These variables ere selected because they are measurableduring log sul'face inspection. Area (SWID *SLEN), SLEN', SWID ,and SRISE') and allother possible combinations of interaction teilnswere also examined as potential predictor variables. The dependent variables selected were 1)rake (penetration angle), 2) clear wood abovedefect, 3) total depth, 4) halfway-point crosssection width (HWID), and 5 ) halfway-pointcross-section length (HLEN). These variablesperinit an internal model of a defect to be constructed and determine an approximate internallocation (Fig. 4).Within each defect type class, the data wererandomly partitioned into two groups using thecaTools package (Tuszynsh 2006) for the R statistical analysis program. The first group contained 66.7% of the records and was used formodel development and determining the internallexternal feature correlation statistic. The second set contained the remaining records and wasused for testing the prediction models (modelvalidation set). Table 2 shows the numbers ofobservations used in the model development andtesting steps. Adjusted multiple R' tests wereSurfaceIndicatorWidth and Lengthat Midpoint.-.-.-1ClearWood4Penetration DepthFIG.3. Series of internal defect sections for surface indicator shown in Fig. 2.FIG.4. Illustration of internal features predicted by themodel.
18WOOD AND FIBER SCIENCE, JANUARY 2008, V. 40(1)TABLE2. Nuinbers of obseivntiorzs used iiz nzoclel development aizd testing by defect type.types: OK, OKC, sound knots (SK), and unsound knots (UK). The strength of the relationship (adjusted multiple R') between the interiorNumber of observatio lshalfwaypoint width (HWID) illeasureinent andDerectModelTest ngTotalcodedataseldatasetobservationsexternal features ranged from 0.47 to 0.73. SimiAK9447141lar results were found to exist among externalAKC8443127features and the halfway point length (HLEN)HD8945134measurement (adjusted multiple R2 from 0.46 toLD6232940.73).Most of the severe defect observationsMD11357170terminatedat the pith, approximately the center163OK10855OKC14721of the slab for most samples. This is demon72SK4824strated in the strong relationship among penetraUK261339tion depth and external features-specifically diameter with adjusted multiple R2 ranging from,used to determine the con-elation among external 0.63 to 0.76. The strongest con-elation for rakedefect measurements and internal features. Ad- angle was for the sound knot defects (adjustedjusted R' is a modification of R2 that adjusts for illultiple R2 0.70). However, in most cases,the number of explanatory terms in a model. the relationship between rake and external feaUnlike R2, the adjusted R' increases only if the tures was not very strong with adjusted multiplenew term improves the model inore than would R2 ranging from 0.22 to 0.36 for the other severeknot defects. For all knot defects the correlationsbe expected by chance.between external indicators and internal featureswere significant (p 0.01). Clear area above anRESULTS AND DISCUSSIONencapsulated defect is not normally present withknot defects; thus no model was pursued for thisCorrelation results for model develop nent feature.and significant predictor variables are presentedTable 3 lists the most significant independentin Table 3. The number of overgrown knot clus- or predictor variables for each defect type andter (OKC) defects was not sufficient for estab- internal feature. The independent variables arelishing a defect prediction model. Due to feature listed in the order of most to least significant.similarities between overgrown knots (OK) and Correlations with diameter inside bark (DIB)overgrown knot clusters (OKC), data from these and/or surface indicator length (SLEN) were thedefect types were combined. This data set was most common. DIB was a significant variable inthen analyzed to see if the combined data could 3 more instances (27 vs. 24) than SLEN, andbe used to predict internal features for OKC de- DIB had a stronger influence in almost all cases.fects. A separate analysis was performed for SRISE and SWID had significant influences onovergrown knot model development and testing the external internal relations in fewer instances;using only overgrown knot data. In addition, 15 and 12, respectively. Interaction terms suchthere was not a sufficient sample size of wound as surface area (SWIDSkSLEN), volumeand bump defects for model development. These (SWID*SLEN4:SRISE), and others had a sigdefect types have been excluded from further nificant correlation to internal features in 11 indiscussion here.stances.The mean-absolute error (MAE) is the meanofthe absolute value of the residual errors forKnots and severe defectsthe fitted equation. MAE indicates the /- errorIn most instances, strong coi elationswere range that can be expected using the fitted equafound to exist among external defect indicators tion to predict defect features. The associatedand internal characteristics for severe defect MAE with the model development samples were
Thoinas-PREDICTING19YELLOW-POPLAR LOG DEFECTS USING SURFACE INDICATORSTABLE3. Model developnie ztand testing correlation .esullsModel develo mentresultsDefect tedR squaredMeanabsoluteerrorCol esYesYesYesNoYesYesYesYesNoYesYesYesYesNoModel tesdnn resullsCol elationcoefficienlSignificant independent variables'R'MeanabsoluteerrorDIB, SriseSWid, S V i d 4 ' S e n ,SWid"'SRise, SWid"DIB,SWid'kSLen":SRiseSLen, SRise, SWid"SRiseDIB, SRiseDIB, SRiseDIB, SWid, DIB*SWidSwid, SWid"SLenSWid, SlenDIB, SWid, DIB'l'SWidSLen, SWid, SRise,SWid'l'SLen, SRise'l'SLenDIB, SLen, DIB:l:SLenDIB, SLen, DIBZkSLenDIB, SRiseDIB, SRiseDIB, SRiseSLenSLenSWid, SLenDIB, SWidDIB, SWid, D1B"SWidDIB, SLenDIB, SLenDIB, SWid, SLenDIB, SWid, SLenSLen, SLen'i'SWid,SWid*DIBSLen"D1BDIB, SLenDIB, 1.130.56.626.9YesNoYesYesNoYesNoYes-DIB, SLen, SRiseDIB, SLenSLen, SRiseDIB, SWid-DIB, SWid, SLenDIB, SWid, SRiseSLen, SRiseDIB, SLenDIB, SLenSWid, SLenSlenDIB, SLen-0.74Correlations NoNoNo
20WOOD AND FIBER SCIENCE, JANUARY 2008, V. 40(1)small in most cases. The MAE values in Table 3are reported in millimeters for all measurementsexcept rake angle, which is in degrees. In 10 outof 12 cases for the knot defects, the MAE is 13mm or less. In the remaining two cases the MAEvalues are 15 and 17 rnm for the halfway-inlength measurements for sound and unsoundknots. The MAE values for rake angle variedbetween 8 and 11".The model testing samples were used to analyze the predictive capabilities. The regressionequations generated with the model development samples were used to predict internal feature measurements. The con-elation coefficient,r, the mean absolute error (MAE), and the significance level of the correlation were determined for each defect type and feature combination (Table 3). For most knot defect types(overgrown, overgrown/overgrown knot clusters, and sound knots), all correlations were significant (p 0.01). For the unsound knot defects,only the correlation of halfway point crosssection width was significant with a MAE of 19rnm. The best results were with sound knots withcorrelation coefficients (R) ranging from 0.71 to0.87. The correlation coefficients with predicteddepth for the overgrown knots and knot clusterswere 0.85 and 0.86. Total depth had the smallestMAE, ranging from 11 to 14 mm. The rakeMAE ranged between 8 and 14". For a 400-mrndiameter log with a defect terminating near thecenter, a 10" error would change the defect center position by 43 (10" under-estimate) to 54 mrn(10" over-estimate) at the pith, depending onangle (Fig. 5). The difference in defect positionat the halfway point would have a maximumpositional variance of 22 (10" under-estimate) to27 mm (10" over-estimate) depending on degree.Overall, with the exception of rake angle, theovergrown knot correlation results were betterthan those from the grouped overgrown knot andcluster data (Table 3). Further, the differences inpredicted depth between the overgrown knot andcluster data were minimal. However, the differences between correlation results for predictedhalfway-in width and length suggest that using aseparate model for only overgrown knots isPith47400 mrnPFIG.5. Impact of rake error on internal defect position.better than using the grouped knot and clustermodel.Bark distortions and adventitious defectsIn general, the correlations between externalindicator measurements and internal features forbark distortion and adventitious defects were notas strong as those measured for the severe knotdefects. Distortion defects have been encapsulated longer than the more recent knot defects.Thus, less surface information is available forthe distortion defect types. The model development correlation results are listed in Table 3.The strongest correlations between externaland internal features for the minor defects werewith the adventitious defect types. Multiple adjusted R for AK defects ranged from 0.20 to0.48 for the correlations that were significant (p 0.01). In addition, the MAE for the HWID andHLEN internal variables was 4 mm, in bothcases. However, for AK defects the correlationbetween the clear area above the defect and external indicators was not significant (p 0.01).For AKC defects, all correlations among external indicators and internal features were significant (p 0.01), with the exception of rakeangle. Multiple adjusted R' for AKC defects
Thomas-PREDICTINGYELLOW-POPLAR LOG DEFECTS USING SURFACE INDICATORSranged from 0.15, for the clear area, to between0.27 and 0.39 for the remaining internal features.The MAE for the internal features also wassmall with HWID and HLEN MAE values of 7and 6 min, respectively.Good correlations between external and inlernal features were also found for the heavy andmedium distortion defects. With the exceptionof rake, all correlations for these defects weresignificant (p 0.01). The strongest relationshipwas associated with the prediction of total depthwith adjusted R' values of 0.76 and 0.81 forheavy and medium distortions, respectively. Themean absolute errors for the fitted models were5 mln for halfway-in width, 9 inm for halfway-in length, 5 11 mnl for predicted depth, and19 to 22 mm for encapsulation depth. For lightdistortion defects, depth was the only featuresignificantly related to external features with amultiple adjusted R' of 0.56 (p 0.01).All minor defect types had at least one featurethat did not have a significant correlation withexternal features. Rake angle was the most difficult to predict internal feature with a significant correlation to external features found onlywith adventitious knots. In this case, the adjustedR' was low, but the mean absolute error wasonly 4.09". The correlation among clear areaabove an encapsulated defect and external features was also weak with adjusted R' valuesranging from 0.10 to 0.25.Hyvarinen (1976) reported that the most common significant predictor variable for sugarmaple clear area was DIB. Thus, these resultsare in agreement. Diameter was the most frequent significant predictor for the distortion andadventitious defect types (17 of 25 cases).SWID, SRISE, and SLEN were the next mostfrequently significant variables, identified in 8,7, and 6 of the stepwise analyses, respectively.Defect surface area was a significant predictorvariable of internal features in only three cases.Analyzing the predictor equations using themodel testing samples showed that the modelsperform well. The col-relations for halfway-inlength and total depth were significant for alldefect types (p 0.01). The correlation coefficient R ranged from 0.37 to 0.79, and 0.45 and210.90 for halfway-in length and depth, respectively. The correlation between halfway-inwidth and the predictor equations was significant for all defect types with the exception ofadventitious knot clusters. For halfway-in widththe qo elationcoefficient R ranged from 0.29 to0.87. fr or relations with clear area were significant only for medium and heavy distortions withcorrelation coefficientsof 0.52 and 0.38, respectively. The MAE in these two cases was 15 and20 mnl for heavy and medium distortions, respectively. Although the clear area MAE forthese defect types could be regarded as low, theyindicate that the equations are accurate to approximate 30-mrn precision (i.e.: 15 n m).It isnot clear if this level of precison is adequate foruse in optimizing log-breakdown. For the distortion and adventitious defects with the modeltesting sample, none of the rake angle modelsprovided a significant correlation (p 0.01).CONCLUSIONSOverall, the strongest correlations, both modeldevelopment and testing, occurred with the mostsevere defect types (OK, OKC, SK, UK). Theseare also the most recently occurring defects onthe tree and have had the least time to be encapsulated or grown over. Thus, more detail aboutthe surface indicator exists. Conversely, some ofthe weakest correlations that were discoveredinvolved the least severe defect types (AK,AKC, LD, MD, HD). The adventitious knots/buds and the light and medium distortions werethe oldest defects examined and had the longesttime to encapsulate. Thus, less surface indicatorexisted for these defect types. A heavy distortiondefect is at the point in the encapsulation processwhere an overgrown knot has made the transition to a distortion defect. More surface indicator detail exists for this type of distortion thanthe others. This was evident in the predictivepower of the HD feature models when comparedto the MD and LD defect models.The goal of this research was to develop models capable of predicting internal defect featuresbased on external defect characteristics. The results from this study indicate that most internal
22WOOD AND FIBER SCIENCE, JANUARY 2008, V. 40(1)vironment for statistical computing. R Foundation forStatistical Computing, Vienna, Austria. ISBN 3-90005107-0, URL http://www.R-project.org. (10 June 2006).RAST,E. D. 1982. Photographic guide of selected externaldefect indicators and associated internal defects in northern red oalc. Research Paper NE-511. Northeastern ForestExperiment Station, USDA Forest Service. 20 pp.AND J. A. BEATON.1985. Photographic guide of selected external defect indicators and associated internaldefects in black cherry. Research Paper NE-560. Northeastern Forest Experinlent Station, USDA Forest Service.22 pp.-, AND D. L. SONDERMAN.1988. Photographicguide of selected external defect indicators and associatedinternal defects in black walnut. Research Paper NE-617.ACKNOWLEDGMENTSNortheastern Forest Experime ltStation, USDA ForestThe author would like to thank A1 WaldronService. 24 pp., AND ---, 1989. Photographic guide ofand the West Virginia Division of Forestry for -selected external defect indicators and associated internaldonating the sample trees from the Camp Creekdefects in white oak. Research Paper NE-628. NortheastState Forest for this research and helping withern Forest Experiment Station, USDA Forest Service. 24sample collection. We also would like to thankPP.Robert Driscole and the staff at the WVU Ex- -, AND . 1990a. Photographic guide ofselected external defect indicators and associated internalperimental Forest for their assistance in sampledefects in yellow-poplar. Research Paper NE-646. Northcollection.eastern Forest Experiment Station, USDA Forest Service.29 pp.-, AND . 1990b. Photographic guide ofREFERENCESselected external defect indicators and associated internalCARPENTER,R. D. 1950. The identification and appraisal ofdefects in sugar maple. Research Paper NE-647. Northlog defects in southern hardwoods from surface indicaeastern Forest Experiment Stabon, USDA Forest Service.tors. Talk delivered at the Deep South Section-Forest35 pp., AND . 1990c. Photographic guide ofProducts Research Society Annual Meeting, Memphis, -selected external defect indicators and associated internalTN. Oct. 26, 1950. (unpublished).AM) E. D. RAST.1989. Defects inD. L. SONDERMAN:defects in yellow birch. Research Paper NE-648. Northhardwood timber. Ag. Handbook 678. USDA, Washingeastern Forest Experiment Station, USDA Forest Service.ton, DC. 88 p.25 PP.CHANG,S. J. 1992. External and intenlal defect detection to SCHULTZ,H. 1961. Die beureilung der Qualitatsentwicklungoptimize the cutting of hardwood logs and lumber. Transjunger Baum, Forstarchiv, XXXII (May 15, 1961), p. 97.ferring technologies to the hardwood industry: Handbook SHIGO,A. L. AND E. VH. LARSON.1969. A photo guide to theNo: 3. U.S. Dept. of Commerce, Beltsville, MD. ISSNpatterns of discoloration and decay in living northern1064-3451.hardwood trees. Research Paper NE-127. NortheasternHYVARINEN,M. J. 1976. Measuring quality in standingForest Experiment Station, USDA Forest Service. 100 pp.trees-Depth of knot-free wood and grain orientation un- STAYTON,C. L., R. M. MARDEN,R. G. BUCKMAN.1970. Exder sugar maple bark distortions with underlying knots.terior defect indicators and their associated interior defectPhD Dissertation. University of Michigan. 142 pp.in sugar maple. For. Prod. J. Vol 2 0 2 (55-58).L. 2006. Processing data for outliers. R News, Vol. STEELE,P. H., T. E. G. HARLESS,F. G. WAGNER,KOMSTA,L. KUMAR,612: 10-1 3. http:l/cran.r-project.orgldoc/Rnewsfiews AND F. W. TAYLOR.1994. Increased lumber value from2006-2.pdf (May 2006).optimum orientation of internal defects with respect toLEMIEUX,H., M. BEAUDON,AND S. Y. ZHANG.2001. Charsawing pattern in hardwood logs. For. Prod. J. 44(3):acterization and modeling of knots in black spruce (picea69 -72. 2006. PhD Dissertation. Virginia Polytechnic Inmariana) logs. Wood Fiber Sci. 33(3):465-475.stitution, Blacksburg, VA. 160 pp.MARDEN,R. M. AND C. L. STAYTON.1970. Defect indicatorsJ. 2006. caTools: Miscellaneous tools: UO, movin
curred in much fewer numbers at CCSF than at WVUF. More LD defects were found on CCSF FIG. 1. Encapsulation depth and stub scar relationship Sanlples than the olles from WVUF. ratio. Sanzple processing lected. For each tree, the number of defects by type were counted. The types of defects identi- fied, counted, and analyzed in this study are
