Abstract: The relative importance of density, acoustic velocity, and microfibril angle (MFA) for the prediction of stiffness(MOE) and strength (MOR) has not been well established for Douglas-fir (Pseudotsuga meniiesii (Mirb.) Franco). MOEand MOR of small clear specimens of mature wood were better predicted by density and velocity than by either variablealone (183 trees >20 years old, six specimens per tree, 1087 specimens total). Specimens sampled around the stem circum-ference had similar density (intraclass con-elation coefficient t = 0.74) but not MOE (t = 0.40) or acoustic velocity (t =0.32), indicating benefits from sampling several circumferential positions. For MOE, the path coefficients (B) were moder-ate for density and velocity. For MOR, B as only high for density. End-matched samples of one specimen per tree wereanalyzed with SilviScan. Simple con-elations with MOE were highest for density (t = 0.67) and then acoustic velocity2

(0.53), MFA (-0.50), earlywood MFA (-0.45), and latewood proportion (0.40) .Most correlations were weaker for MOR.Density had a higher B than did MFA for either MOE or MOR. In more complex path models, latewood proportion and la-tewood density were the most important contributors to MOE and MOR, and MFA was relatively unimportant. The pathanalyses showed what simple correlation did not: that latewood proportion has strong predictive value for Douglas-fir ma-ture wood quality.

IntroductionDouglas-fir (Pseudotsuga menziesii (Mirb.) Franco) logs

and lumber are of economic importance for forest productsindustries in western North America, New Zealand, parts ofEurope, and other locations where acceptable growth ratescan be achieved. Its primary uses are building and construc-tion, and it is used in such products as dimension lumber,

piles, and plywood (Alden 1997; Forest Products Laboratory1999) as well as laminated veneer lumber and many otherproducts. Douglas-fir has maintained a strong market pres-ence because of its superior stiffness (modulus of elasticity,MOE) and strength (modulus of rupture, MOR) (Barbourand Kellogg 1990). This research was designed to betterunderstand the relationships between MOE, MOR, and twofactors often used for their estimation, wood density and

MFA are typically weaker than the corresponding relation-ships for MOE (Downes et al. 2002; Yang and Evans2003). Materials science shows that MOR should be deter-mined less by a simple relationship with density and (or)MFA than by the weakest component within the sample.Such weak components can result from stress risers at allscales, such as at the molecular level in the polymer struc-ture, and at pits, rays, resin canals, compression wood, inter-nal checks, rnicrofractures, slope of grain, and knots(Bohannan 1966; reviewed in Kollmann and Cote 1968,Chap. 7). Therefore, knowledge of the average materialproperties will not necessarily give good predictions ofMOR. Nonetheless, correlations and regressions are com-monly tabulated, studied, and refined for MOR and MFA(e.g., Dinwoodie 2000; Yang and Evans 2003) or MOR anddensity (e.g., Zhang 1995; Forest Products Laboratory 1999;Downes et al. 2002; Kumar 2004; Liu et al. 2007). Theserelationships are of value, even though theory does not sup-port that they are entirely causal.

The first objective of this study was to determine the pre-dictive power of density and acoustic velocity for estimatingMOE and MOR in small clear specimens of Douglas-fir.This information will be helpful in interpretation of acousticvelocity data for samples of various sizes for both researchand operations. The second objective was to better under-stand the relationships between Douglas-fir's mature woodMFA and its MOE and MOR with and without consideringwood density. This information will be useful for silvicultu-rists as they manipulate tree growth through practices affect-ing anatomy, for tree breeders as they make selections thatinclude wood quality, and for tree growers and log buyersas they estimate the value and quality of trees and logs.

Materials and methodsWood material

Wood samples came from 17 Douglas-fir stands. We se-lected stands that were >20 years old at breast height, in-cluded a range of site indices ensuring a range of radialgrowth rates, had no fertilization or precommercial thinningin the previous seven years, and had needle retention valuesindicating that Swiss needle cast disease was not impactinggrowth: all stands had more than three annual cohorts ofneedles at the fifth whorl from the top (see Maguire et al.2002). The stands were located in the Coast Range and thewestern Cascades in northwestern Oregon (between latitudes44.2 and 45.6° and between longitudes -122.0 and -123.7")at intermediate elevations (from 220 to 1012 m), Standsaveraged 22-41 years old at breast height and site indices(King 1966) ranged from 35.0 to 45.3 m at age 50. Ringwidth was estimated for each tree as the average of the sixspecimens per tree from the outer 10 mm of the breastheight bolts (see below). Of the 183 trees, radial growthrate varied by a factor of 6 from 1.1 to 6.2 mm/year (0.04-0.24 in./year).

From each stand, 7-12 dominant or codominant treeswere sampled to include a range of diameters in each standfor a total of 183 trees. We selected trees such that whenthey were removed, the remaining stand spacing was consis-tent with a thinning, not a patch cut. To avoid complications

acoustic velocity, and to learn how microfibril angle (MFA)and density affect strength properties independently and incombination. The research was undertaken using small clearspecimens of mature wood.

MOE is often estimated rather than measured directly tominimize complexity and cost and (or) to avoid destructivesampling. One common technique for such estimation is touse the velocity of acoustic waves through wood, with orwithout the additional factor of wood density. Theory showsthat MOE = wood density x acoustic velocity2 (Knowles etal. 2004; Bucur 2006). Experimentally, acoustic velocitymeasurements are easier to make than mechanical measure-ments and can be undertaken on sawn specimens, entirelogs, and standing trees. Moreover, the acoustic data aresimple to manipulate and interpret. Here, we model MOEand MOR using density, MFA, acoustic velocity, and acous-tic velocity2 to learn which factors best explain variation inour sample set.

We know in a general way that the microscopic structureof wood is very important to the stiffness and strength ofclear samples (defined as samples that have a minimum of"defects" such as knots, compression wood, slope of grain,and rnicrofractures). However, we have limited understand-ing of the microscopic properties that best account for thevariations in stiffness and strength in Douglas-fir maturewood. The stiffness of lumber can be estimated by analyzingits component materials at several different scales. On thelarger scale, one can ask whether MOE is better predictedby density or MFA. Bulk density is an isotropic measure ofthe amount of material (often assumed to have uniform spe-cific stiffness) available to accommodate stresses. MFA is amajor determinant of the anisotropic properties of wood anda measure of the effectiveness of that material to accommo-date stresses (Cave 1968; Cave and Walker 1994). The ef-fect of density on MOE has been studied extensivelybecause of its relative ease of measurement (e.g., Cown etal. 1999, 2004; Rozenberg et al. 1999) and because it is cor-related with many wood quality traits. MFA effects arebeing increasingly studied (Evans 1997; Yang and Evans2003; Barnett and Bonham 2004) as measurement methodsbecome more accessible (Evans 1997; Evans and Ilic 2001).The variation in MOE of clear straight-grained wood can bebetter explained by using both density and MFA in combi-nation rather than using either single measure (Megraw1986; Evans and Ilic 2001; Downes et al. 2002; Knowles etal. 2003; Vikram 2008).

On the smaller scale, one can ask whether MOE can bepredicted by the proportion of latewood (LW) in a samplebecause LW is denser and has a lower MFA than earlywood(EW) and therefore contributes a larger proportion to thesample's stiffness. Composite theory shows that the MOEof wood should be the simple volume-weighted average ofthe MOE of the EW and the MOE of the LW (Bodig andJayne 1982). Therefore, we could find that LW proportion,a factor that is quite simple to characterize in Douglas-fir,is a strong driver of MOE because it combines informationon both LW and EW MOEs. On an even smaller scale, onecan ask whether MOE can be predicted from detailed EW orLW characteristics alone and the extent to which each ofthese characteristics contributes to local MOE.

The relationships between MOR and either density or

Acoustic velocityAfter specimens had been tested mechanically, a 15 cm or

longer unbroken piece was sent to CSIRO (Clayton, Aus-tralia). Acoustic velocity of each segment was measured at12% moisture content (I1ic 2001).

Data analysisWe described the data by determining the mean and var-

iance values for the dependent variables MOEd and MORdand the independent variables densityd, acoustic velocityd,

densityss, MFAss, EW densityss, EW MFAss, LW densityss,LW MFAss, and LW proportionss. For the dependent varia-bles and the first two independent variables, there were datafor 1087 specimens. For the remaining variables, there wasone sample for each of the 183 trees.

The intraclass correlation coefficient (t) was used to quan-tify the relative repeatability of within-tree estimates. Thestronger the correlation, the more the individual estimateswithin a tree are alike compared with estimates from othertrees. The intraclass correlation coefficient was estimated as

of compression wood, trees were rejected if their lower stemwas leaning at breast height.

After trees were felled. a 30 cm tall bolt was cut from di-rectly above breast height (1.4 m) and then a thin disk wascut from above that bolt. While the wood was still green, sixvertical-grained specimens 1 x 1 x 30 cm (radial xtangential x longitudinal) were cut from the outer xylem ofeach bolt. Depending on the radial growth rate of the tree, aspecimen contained from 1.5 to 9 annual rings. The speci-mens were air-dried to constant mass at 12% moisture con-tent and weighed and then measured in all three dimensionsfor calculation of wood density (dry mass per volume at12% moisture content) defined as density, (directly meas-ured density). This value is distinct from the density valuefrom SilviScan, densityss described below.

We randomly chose one specimen from each tree (n =183) and then marked its end-matched location on the diskwhile the disk was still fresh. A short strip 2 cm wide (tan-gential direction) and about 2 cm long (radial direction) wasremoved from that location, extracted in ethanol, and air-dried following SilviScan requirements (Downes et al.1997).

Bending testsThe specimens were subjected to static bending tests on

an Instron universal testing machine with a 4200 N loadcell that had a standard error of 4 N. We followed ASTMstandard D 143-94 (ASTM 1999) with the load applied tothe tangential-longitudinal face closest to the pith but withspecimen supports set 15 cm apart for a 15: 1 span to depthratio. The load was applied continuously at a rate of5 mm/min. MOEd (stiffness) and MORd (the maximumforce withstood before breaking) were then calculated ac-cording to the formulas specified in Markwardt and Wilson(1935); MOEd was the slope of the linear portion of thestress-strain curve and MORd was the stress applied atspecimen failure. Note that the MOEd and MORd data usedin all analyses came from direct measurements and not fromSilviScan. Values are designated with subscript "d" if theywere determined directly on entire small clear specimensand with subscript "ss" if they derived from the SilviScanestimates of the end-matched strip.

SilviScan analysesSilviScan was used to estimate microfibril angle (MFAss)

and density traits (Evans and Ilic 2001) on the xylem thatwas end-matched to the one specimen per tree (n = 183trees). Densityss and MFAss were estimated in the radial di-rection at 0.05 and 0.2 mm intervals, respectively. For eachsample, SilviScan provided ring-by-ring values for ringwidthss, EW widthss, LW widthss, LW proportionss, ling den-sityss, EW densityss, LW densityss, ring MFAss, EW MFAss,and LW MFAss. We counted one growth ring in from thecambium and then used the data from entire growth lingsthat were totally or partially in the next 1 cm of the strip.Note that the actual beam, while roughly coming from thissame location, did not necessarily include entire growthlings. We weighted values by the EW and LW ling widthsfor those selected growth rings to estimate the averagespecimen value for densityss, EW densityss, LW densityss,MFAss, EW MFAss, LW MFAss, and LW proportion.

Variance components were estimated with the SAS varcompprocedure with the REML option using the model Variable =Tree.

We used simple correlations to determine the associationof the dependent variables (MOEd and MORd) with the in-dependent variables (densityd, velocityd, velocityd

2, densityss,MFAss, 1/MFAss, MFAss

2, EW densityss, EW MFAss, LWdensityss, LW MFAss, LW proportionss, and densityd/MFAss).These correlations were run for three data sets: specimen-level data using the 183 samples for which there were asso-ciated data from SilviScan, specimen-level data using all1087 specimens, and tree-level data using the tree meansfrom the six specimens per tree. The specimen-level analy-ses treated each specimen as an independent sample.

We had examined the variable densityd/MFAss because inEucalyptus, it was more strongly correlated with MOE thanwas either MFA or density alone (Yang and Evans 2003). Inthe current study, the correlation of MOE with den-sityd/MFAss (r = 0.611) was intermediate between the corre-lation of MOEd with density, (r = 0.672) and MOEd withMFAss (r = -0.498). The results were similar for MORd: thecorrelation of MOR with densitYdlMFAss (r = 0.426) fell be-tween those with density; (r = 0.626) and with MFAss(-0.283) alone. Because it did not improve predictions overthose from the best variable alone (densityd), no furtheranalyses were done with densityd/MFAss.

Regressions were also performed with the REG procedurein SAS (SAS 1999) to estimate either MOEd or MORd usingall combinations of one, two, or three of the following vari-abies associated with the short clear specimens: densityd, ve-locityd, and velocity d

2.Lastly, we performed path analyses to better understand

the independent effects (the direct paths) in several associa-tions of dependent and independent variables. Figure 1 showsa simple path diagram for the relationships among density,MFA, and MOE. The direct path between density and MOE(labeled B1) indicates the correlation if the other independentvariables (here, just MFA) are held constant. The simple cor-

that were relatively similar. In contrast, MOEd and acousticvelocityd had relatively small t values (0.40 and 0.32, re-spectively) and MORd had a moderate t value (0.57). Thesevalues demonstrate considerable within-tree variation forMOEd, acoustic velocityd, and MORd (Table 1).

Here, we report on the relationships of MOEd and MORdwith variables that were measured directly on 1087 speci-

. mens. Among-tree correlations (Table 2c) were generallyhigher than were among-specimen correlations (Tables 2aand 2b) as a result of the more stable estimates with thetree means. The strongest correlation of MOEd with direct-measured values was with MORd followed by densityd, thenvelocity 2d, and then velocityd, (Table 2). MORd had a similarcorrelation with density as did MOEd but much lower corre-lations with velocityd, or velocity 2

d. The correlations of veloc-ity 2

d with MOEd

and MORd were only slightly stronger thanthose with velocityd. Among the ring MFA variables, MFA 2sshad slightly higher correlations with MOEd and MORd thandid MFAss or 1/MFAss, but all values were of very similarmagnitude.

Table 3 shows the regression models predicting MOEdand MORd with various combinations of densityd, acousticvelocityd, and acoustic velocity 2

d. All of the models had ad-justed R2 values >0.3 for MOEd,' Unlike MOEd, MORI waspredicted poorly by velocityd, velocity 2

d, or their combina-tion; adjusted R2 values ranged from 0.08 to 0.25. Whendensity, was included in the models for MORl, the adjustedR2 values increased, although they were substantially lowerthan the values for the same model for MOEd.

The single-variable regression models indicated that ve-locity 2

d was a slightly better predictorthan was velocity, ofMOEd or MOR

d(Table 3). Two-variable models using den-

sityd and velocityd, essentially explained the same amount ofvariation as the two-variable models using density, and ve-locity 2

d. The three-variable models offered no real improve-ment over the two-variable models: only one of the three-variable models was significantly better than its two-variablecounterpart (MOEd for small clear specimens), but it ex-plained <1 % more of the variation as R2 increased from0.547 to 0.551.

The regression models for the tree averages explainedmore variation than did the corresponding models using thespecimens (Table 3) because of the greater stability of thetree averages. Likewise, in the path model analyses, themodel with tree-level data (Model 1a, Table 4) explainedmore of the variation than did the models including onlyspecimen-level data (remaining models in Table 4).

For MOEd, the path coefficients (B) for both density, andvelocity 2d were relatively high (0.44-0.57) (Models 1a and1b, Table 4). For MORd, however, the B value for densityd(0.58-0.61) was much higher than that for velocity 2

d (0.17-0.20).

Most of the correlations involving MOEd (r, Table 2b) layon the direct paths (B, Model 1b, Table 4): for MOEd anddensityd, r = 0.59 and B = 0.49 and for MOEd and velocityd,r = 0.57 and B = 0.45. The same pattern held for MORd anddensity, (r = 0.62 and B = 0.58), but most of the correlationbetween MORd and velocity 2

d derived from the indirect path

relation between density and MOE (rdensity-MOE)is the sum ofthe direct path (B1) and the indirect path (rdensity-MFAB

2)'Amore thorough explanation of path analysis can be found inSokal and Rohlf (1995).

The path analyses were undertaken for two dependentvariables (MOEd and MORd) and four combinations of inde-pendent variables: densityd, and velocityd

2 (Models 1a and1b), densityd, and MFAss (Model 2), densityss, and MFAss(Model 3), and EW densityss, EW MFAss, LW densityss,LW MFAss, and LW proportionss, (Model 4). The last modelis more complex than the three-variable model in Fig. 1, butthe same principles apply. Instead of one indirect path foreach independent variable, four independent paths exist foreach correlation between each independent variable and thedependent variable.

Direct paths were determined by first standardizin all ofthe traits to have a mean of 0 and a standard deviation of 1and then by running a regression with MOEd or MORd asthe dependent variable and the combinations listed above asthe independent variables. The regression coefficients repre-sent the path coefficients (B). Even though there was a slightimprovement in the models using MFA 2

ss over MFAss, we

used MFAss in the path analyses because the improvementin fit was negligible and there is no physical justificationfor the use of MFA 2

ss. The variance explained by MFA ortransformations of it (i.e., MFA2, 1/MFA) was 28% or less,and with such high residual variance, there is insufficientstatistical support for using variations of MFA in multipleregression analysis of these data. Analyses were performedon both the individual specimen data and tree averages us-ing the SAS REG procedure (SAS 1999).

ResultsData description

The dependent variables MOEd and MORd had more var-iation associated with them than did the density and veloci-tyd measurements, as shown by their higher coefficients ofvariation (Table 1). The MFAss measurements had the larg-est coefficients of variation of all values reported.

Wood densityd had the highest intraclass correlation coef-ficient (t) of the four variables examined (Table 1): 74% ofthe variance in density, among and between trees was ex-plained by variance among trees. In other words, the sixspecimens taken from around the circumference had values

had SilviScan data (Fig. 2). We plotted the direct-measureddensityd rather than the SilviScan-estimated densityss be-cause density, is a better estimate of the sample from whichMOEd and MORd were measured. The correlation of densi-tydwith densityss was only 0.74 (Table 2), and mean densi-tydwas higher than mean densityss, (Table 1). The fact thatthe correlation was not stronger may have resulted fromthree separate issues. First, unlike the direct-measured speci-mens, the SilviScan samples had been in ethanol, removingsome of the mass. Second, densityd integrates the entirespecimen, whereas densityss was estimated from a transectacross an end-matched strip adjacent to the specimen (corre-sponding to only 0.5% of the volume of the short clear sam-ple). Third, growth rings may have been incomplete in thedirect-measured specimens, which were sawed to measure10 mm in the radial direction, but complete growth ringswere used for the SilviScan samples. This difference ingrowth rings is potentially of much importance in a species

(r = 0.31 and B = 0.17). Therefore, looking solely at the cor-relations could lead one to assume incorrectly an elevatedimportance of acoustic velocityd, for MORd.

Anatomical drivers of MOEd and MORdHere, we investigated the relationships of MOEd and

MORd with the SilviScan-estimated values from the 183end-matched samples (Table 2a). The strongest correlationof MOEd was with density., (r = 0.58) followed by theMFAss estimates (all around -0.50) and then EW MFAss(-0.45). The correlations between these characteristics andMORd were all weaker than for MOEd, with the exceptionof LW proportion, which was of similar strength. Of the Sil-viScan-estimated characteristics, density., (r = 0.41) and LWproportionss (r = 0.40) correlated most strongly with MORd.

Plots of MOEd and MORd versus three of the strongestdrivers (densityd, MFAss, and LW proportionss) are presentedto allow inspection of the data for the 183 specimens that

MOEd or MORd in Model 4, LW proportionss had the stron-gest association, but the corresponding B values were only0.31 and 0.36, respectively. To a large extent, the variationin LW proportion incorporates the LW versus EW differen-ces in densityss (0.86 versus 0.30 g/cm3) (Table 1) andMFAss (11.8 versus 16.3o) (Table 1). Other anatomical com-ponents that were statistically significant drivers of MOEdwere LW densityss, EW densityss, and EW MFAss (all ofwhich had B values <0.28). Besides LW proportionSS, theonly other significant anatomical driver of MORd was LWdensityss with a low B value of 0.21.

DiscussionPredicting MOE and MOR from density and acousticvelocity

The first objective of this study was to determine the pre-dictive power of density and acoustic velocity for estimatingMOE and MOR in small clear specimens of mature wood inDouglas-fir. Density, was only slightly better than acousticvelocity d as a means of predicting MOEd, but both variablestogether greatly improved the predictive power. For speci-mens, the adjusted R2 increased from 0.35 to 0.55 goingfrom the best single-variable to the two-variable model(Table 3), and for trees, the adjusted R2 increased from 0.55to 0.73 (Table 3). Because both densityd, and velocityd, hadfairly strong direct paths (B) for predicting MOEd the twovariables together gave better predictions than did any sin-gle-variable model. Of the two-variable models, it did notmake any meaningful difference in the R2 whether velocitydor velocity 2

d was used with densityd.In Pinus radiata D. Don, Chauhan and Walker (2006)

found that the acoustic velocity was tightly correlated withMOE (which had been estimated by SilviScan) for threeage classes of trees, despite the fact that the strength of thecorrelations between density and MOE depended on standage (either not correlated or correlated very weakly in the

such as Pseudotsuga menziesii that has such large differen-ces between EW and LW values of density and MFA.

As with the correlation analyses (Table 2), regressionanalyses showed that higher R2 values were obtained withmodels with densityd, than models with densityss (Table 5).This result may have been in part due to the fact that theSilviScan strips did not truly represent the adjacent shortclear samples. MFAss was not statistically significant in pre-dicting MORd when density was already included in themodel.

The path coefficients for the simple models examining theimpact of density (as either densityd or densityss) and MFAsson MOEd and MORd are shown as Models 2 and 3 in Ta-ble 4. Densityd was more strongly associated with MOEdand MORd than was densityss, and either value of densitywas more strongly associated with MOEd and MORd thanwas MFAss. For predicting MOEd, MFAss gave B valuesonly half the magnitude of those obtained using either densi-tyd or densityss, MFAss was not a significant predictor ofMORd, when density was in the model (Models 2 and 3, Ta-ble 4).

It is interesting to note that MFAss behaved similarly toacoustic velocity, (and acoustic velocity 2

d) in the previoussection in that both were less predictive of MOE than wasdensityd, Likewise, acoustic velocityd was correlated morestrongly with MFAss (r = -0.69) than with densityd (r =0.33) or density., (r = 0.36) (Table 2a). These results sup-port the theoretical expectation that velocity is much morestrongly determined by MFA than by density. In fact, whenpath analysis was used to estimate the dependence of veloc-ityd on density, and MFAss the direct path for MFAss wasstatistically significant (B = -0.667, p < 0.001), but the di-rect path for densityd, was not (B = -0.067, p = 0.255).

The more complex path diagram examining the EW andLW components are shown in Model 4 (Table 4). Of all ofthe associations tested between independent variables and

8- and 16-year-old stands and moderately con-elated in the25-year-old stand). Their work suggests that acoustic veloc-ity is a good measure of the nondensity components that de-termine MOE, presumably MFA. This interpretation isconsistent with results reported here, that acoustic velocitydwas more strongly associated with MFAss (r = -0.69) thanwith densityd (r = 0.33) or density., (r = 0.36) (Table 2a)and that the direct path for MFAss with velocity was statisti-cally significant (B = -0.667, p < 0.001) but that the directpath for densityd and velocity was not (B = -0.067, p =0.255).

Acoustic velocity alone was a very poor predictor ofMORd, but its inclusion in models with density improvedthe predictions over density alone, although to a lesser ex-tent than they did for MOEd (Table 3). This smaller im-provement was because of the relatively weak direct pathfor velocityd (Table 4), which was nevertheless a reasonablepredictor because of its correlation with density and the in-direct path with MOE; i.e., the con-elation of velocityd with

MOE was not much weaker than the correlation of densitywith MOE, despite the weak direct path. If one had tochoose between densityd and velocityd for predicting MORd,the choice would be densityd unless densityd had a signifi-cant cost disadvantage.

The circumferential variability of properties was quanti-fied using intraclass correlation analysis. Much higher uni-formity was found for wood densityd (t = 0.74) than forMORd (t = 0.57), MOEd (t = 0.40), or acoustic velocityd(t = 0.32). Given that the combination of density and MFAexplains much of the variability of these dependent varia-bles, this result suggests that MFA varies greatly around thecircumference. It would be interesting to study the circum-ferential variability of MFA and density in many trees tofurther elucidate their contributions to the variability ofMOE; variable growth stresses caused by factors such asprevailing wind directions or branch placement (had thespecimens not been clear) could contribute to these differen-ces in interclass correlation among variables. These data

suggest that sampling for MOE and testing for acoustic ve-locity should be done at more than one location around thecircumference at a given height.

Anatomical drivers of MOE and MORThe second objective was to better understand the rela-

tionship between mature wood MFA, and MOE and MOR.The current research was based on small clear specimens. Itwill be important to understand the magnitude of the effectson strength values of wood inhomogeneities that raise stresslocally (e.g., rays, checks, and knots) and to learn which ofthese stress risers have the largest effects as the size andvolume of the wood sample are increased (e.g., Bohannan1966). We would expect such information to show that mi-croscopic anatomy is of less importance than was reportedhere, as other features become increasingly important inlarger pieces of wood.

EW MFA was more influential than LW MFA based onthe direct paths (B) (Table 4), possibly because EW had awider range of MFAs than did LW (9.5-26.7o versus 8.7-21.2o, respectively). Path analyses showed that the variablethat most impacted MOEd and MORd was LW proportion, afact not revealed in simple correlations. On average, LWhad 2.9 times the densityd and 0.7 times the MFAss of EW(Table 1). Using the fifth equation in Table 5 and the valuesof densityss and MFAss from Table 1, LW is estimated tohave 2.3 times the MOE of EW (17400 versus 7700 MPa),showing the strong potential influence of LW proportion onMOE.

Density had a larger influence than MFA on both MOEand MOR in the current study on Pseudotsuga mentiesii.This result may have been different had we studied juvenilewood. For example, in studies of juvenile wood of hardpines, MFA had a larger intluence than did density onMOE of Pinus radiata (Baltunis et al. 2007) and on MOEand MOR of Pinus resinosa Ail. (Deresse et al. 2003). Instudies that looked at samples of both mature wood and ju-venile wood in hard pines, researchers have shown that den-sity is the most important predictor of MOE (and sometimesMOR) in mature wood, but in the juvenile wood, MFA is

relatively more important, sometimes more important thandensity (Cown et al. 2004; Chauhan and Walker 2006; Viaet al. 2009). Therefore, further research is needed to learnwhether the relationships of anatomy to MOE and MOR dif-fer in juvenile wood from those in mature wood in Douglas-fir. Also, in contrast with the hard pines, Pseudotsuga men-ziesii shows a very high contrast between its EW and LWvalues of MFA and density, which may greatly magnify theimportance of LW proportion on stiffness and strength.Thus, it will be important to look at LW proportion as anexplicit factor in research on strength and stiffness of Doug-las-fir in juvenile and mature wood, in wood with acceler-ated growth rate, and in wood resulting from disease orsilvicultural practices such as pruning or fertilization thatcan alter LW proportion. Such proposed research togetherwith the results of this project may be of use to treebreeders, tree growers, and log and lumber processors asthey strive to obtain maximum value from their logs andproducts.

AcknowledgmentsWe thank Amy Grotta and Matt Peterson for field and

laboratory help, John Nairn for discussion, and anonymousreviewers for helpful suggestions. This research was fundedby the PNW Agenda 2020 program and a special USDAgrant to Oregon State University for wood utilization re-search.

ReferencesAlden, H.A. 1997. Softwoods of North America. Gen. Tech. Rep.

FPL-GTR-102. U.S. Department of Agriculture, Forest Service,Forest Products Laboratory, Madison, Wis.

ASTM. 1999. Annual book of ASTM standards. Vol. 04.10. Wood.American Society for Testing and Materials, West Consohocken,Pa.

Baltunis, B.S., Wu, H.X., and Powell, M.B. 2007. Inheritance ofdensity, microfibril angle, and modulus of elasticity in juvenilewood of Pinus radiata at two locations in Australia. Can. J.For. Res. 37(11): 2164-2174. doi:10.1139/X07-061.

Barbour, R.J., and Kellogg, R.M. 1990. Forest management and

end-product quality: a Canadian perspective. Can. J. For. Res.20(4): 405-414. doi: 10.1J39/x90-058.

Barnett, J.R., and Bonham, V.A. 2004. Cellulose microfibril angle inthe cell wall of wood fibres. Biol. Rev. Camb. Philos. Soc. 79(2):461-472. doi: 10.1017/S 1464793103006377. PMID: 15191232.

Bodig, J., and Jayne, B.A. 1982. Mechanics of wood and woodcomposites. Krieger Publishing Company, Malabar, Fla.

Bohannan, B. 1966. Effect of size on bending strength of woodmembers. For. Servo Res. Pap. FPL 56, May 1966. USDA ForestService, Forest Products Laboratory, Madison. Wis.

Bucur, V. 2006. Acoustics of wood. 2nd ed. Springer, New York.Cave, I.D. 1968. The anisotropic elasticity of the plant cell wall.

Wood Sci. Technol. 2(4): 268-278. doi:IO.1007/BF00350273.Cave, I.D., and Walker. J.e.F. 1994. Stiffness of wood in fast-

grown plantation softwoods: the influence of microfibril angle.For. Prod. J. 44(5): 43-48.

Chauhan. S.S., and Walker, J.C.F. 2006. Variations in acoustic ve-locity and density with age, and their interrelationships in ra-diata pine. For. Ecol Manag. 229(1-3): 388-394. doi:10.1016/j.foreco.2006.04.019.

Cown, D.J., Hebert, J., and Ball. R.D. 1999. Modelling Pinus ra-diata lumber characteristics. Part I: mechanical properties ofsmall clears. N.Z.J. For. Sci. 29(2): 203-213.

Cown, D.J., Ball. R.D., and Riddell, M.J.C. 2004. Wood densityand microfibril angle in 10 Pinus radiata clones: distributionand intluence on product performance. N.Z. J. For. Sci. 34(3):293-315.

Deresse, T., Shepard, R.K., and Shaler. S.M. 2003. Microfibril an-gIe variation in red pine (Pinus resinosa Air.) and its relation tothe strength and stiffness of early juvenile wood. For. Prod. J.53(718): 34-40.

Dinwoodie, J.M. 2000. Timber: its nature and behaviour. 2nd ed. E& FN Spon, London and New York.

Downes, G.M., Hudson. J.L., Raymond, e.A .. Dean, G,H., Michell,A.J., Schimleck, L.S., Evans. R., and Muneri, A. 1997. Sam-pling plantation eucalypts for wood and fibre properties. CSIROPublishing. Melbourne, Australia.

Downes. G.M., Nyakuengama, J.G., Evans, R., Northway, R.,Blakemore. P., Dickson, R.L.. and Lausberg, M. 2002. Relation-ship between wood density. microfibril angle and stiffness inthinned and fertilized Pinus radiata. IAWA J. 23: 235-266.

Evans, R. 1997. Rapid scanning of microfibril angle in incrementcores by X-ray diffractometry. In Microfibril angle in wood.Edited by B.G. Butterfield. University of Canterbury, Canter-bury. N.Z. pp. 116-139.

Evans, R.. and Ilic. J. 2001. Rapid prediction of wood stiffnessfrom microfibril angle and density. For. Prod. J. 51: 53-57.

Forest Products Laboratory. 1999. Wood handbook - wood as anengineering material. Gen. Tech. Rep. FPL-GTR-II3. U.S. De-partment of Agriculture, Forest Service, Forest Products Labora-tory, Madison, Wis.

llic, J. 2001. Relationship among the dynamic and static elasticproperties of air-dry Eucalyptus delegatensis R. Baker. Holz alsRoh- und Werkstoff, 59: 169-175. doi: 10.1007/s001 0701001 98.

King, J.E. 1966. Site index curves for Douglas fir in the Pacific

Northwest. Weyerhaeuser For. Rep. NO.8. Weyerhaeuser For-estry Research Center, Centralia, Wash.

Knowles, L., Hansen, L., Downes, G., Kimberley. M., Gaunt, D.,Lee, J., and Roper. J. 2003. Modelling within-tree and between-tree variation in Douglas-fir wood and lumber properties. Paperpresented at IUFRO Division 5 Conference. Rotorua, New Zeal-and, 11-15 March 2003.IUFRO. Rotorua, New Zealand.

Knowles. R.L.. Hansen, L.W., Wedding. A., and Downes, G. 2004.Evaluation of non-destructive methods for assessing stiffness ofDouglas fir trees. N.Z.J. For. Sci. 34: 87-101.

Kollmann, F.F.P., and Cote, W.A .• Jr. 1968. Principles of woodscience and technology. l. Solid wood. Springer-Verlag, Berlin.

Kumar. S. 2004. Genetic parameter estimates for wood stiffness.strength, internal checking, and resin bleeding for radiata pine.Can. 1. For. Res. 34(12): 2601-2610. doi:10.1139/x04-128.

Liu,C.M .• Zhang. S.Y .• Cloutier, A.• and Rycabel, T. 2007. Model-ing lumber bending stiffness and strength in natural black sprucestands using stand and tree characteristics. For. Ecol. Manag.242(2-3): 648--655. doi: 10.1016/j.foreco.2007.01.077.

Maguire. D.A .. Kanaskie, A., Voelker, W.. Johnson, R., andJohnson. G. 2002. Growth of young Douglas-fir plantationsacross a gradient in Swiss needle cast severity. West.J. Appl.For. 17: 86-95.

Markwardt, L.J., and Wilson, T.R.C. 1935. Strength and relatedproperties of woods grown in the United States. U.S. Dep. Ag-ric. Tech. Bull. 479.

Megraw, R.A. 1986. Effect of silvicultural practices on wood qual-ity. 1986 Research and Development Conference. TAPPI Press.Atlanta, Ga. pp. 27-34.

Rozenberg. P., Franc, A., Mamdy, C., Launay, L Schermann, N.,and Bastien, J.C. 1999. Genetic control of stiffness of standingDouglas fir; from the standing stem to the standardised woodsample, relationships between modulus of elasticity and wooddensity parameters. Part II. Ann. Sci. For. 56(2): 145-154.doi: 10.1051 Iforest: 19990206.

SAS. 1999. SAS / STAT@ user's guide, version 8. SAS InstituteInc., Cary. N.C.

Sokal, R.R., and Rohlf, F.J. 1995. Biometry: the principles andpractices of statistics in biological research. 3rd ed. W.H.Freeman, San Francisco. Calif.

Via, B.K .. So, C.L., Shupe. T.F., Groom. L.H., and Wikaira, J.2009. Mechanical response of longleaf pine to variation in mi-crofibril angle, chemistry associated wavelengths, density, andradial position. Composites: Part A, 40(1): 60-66. d0i:10.1016/j.compositesa.2008.10.007.

Vikram, V. 2008. Stiffness of Douglas-fir lumber: effects of woodproperties and genetics. M.S. thesis, Department of ForestScience, Oregon State University. Corvallis, Ore.

Yang. J.L., and Evans, R. 2003. Prediction of MOE of eucalyptwood from microfibril angle and density. Holz als Roh- undWerkstoff, 61: 449-452. doi:l0.1007/s00107-003-0424-3.

Zhang, S.Y. 1995. Effect of growth rate on wood specific gravityand selected mechanical properties in individual species fromdistinct wood categories. Wood Sci. Technol. 29(6): 451-465.doi: 10.1007/BFOOI94204.