1-s2.0-s0378112707001211-main
TRANSCRIPT
-
8/19/2019 1-s2.0-S0378112707001211-main
1/8
Modeling lumber bending stiffness and strength in naturalblack spruce stands using stand and tree characteristics
Chuangmin Liu 1, S.Y. Zhang a,*, Alain Cloutier b,2, Tadeusz Rycabel b,3
a Forintek Canada Corp., 319 rue Franquet, Quebec City, Qué bec, Canada G1P 4R4b Centre de recherche sur le bois, Dé partement des sciences du bois et de la forê t, Université Laval, Qué bec, Qué bec, Canada G1K 7P4
Received 2 October 2006; received in revised form 25 January 2007; accepted 25 January 2007
Abstract
Static bending modulus of rupture (MOR) and modulus of elasticity (MOE) were measured on lumberfrom trees in the natural and mature black spruce stands grown in Eastern Canada. A sample of total 157 trees from the 90–100-year-old black spruce natural stands covering a range of sites
and growing conditions was used for the model development (n = 102) and validation (n = 55). A stepwise regression method was employed to
identify the best variables for predicting MOE and MOR using stand/tree characteristics and wood properties. Then, regression equations with
different explanatory variables were developed to predict lumber bending stiffness and strength. Based on the results of model validation from the
independent dataset, the regression models developed were able to predict the lumber bending MOE and MOR satisfactorily, especially for small-
and middle-sized trees. The results (equation parameter estimates and predictions) obtained in this study, along with those for plantation-grown
black spruce in Eastern Canada, will be highly useful in predicting lumber bending static stiffness and strength for both natural and managed black
spruce stands.
# 2007 Elsevier B.V. All rights reserved.
Keywords: Black spruce; Tree characteristics; Bending strength and stiffness; Regression model; Picea mariana; Stand characteristics
1. Introduction
Black spruce (Picea mariana (Mill.) B.S.P.) is the most
important reforestation and commercial species in eastern
Canada. In Canada, black spruce is mainly used for lumber and
high quality pulp and paper manufacturing. Black spruce is
highly valued for machine stress-rated (MSR) lumber produc-
tion thanks to its strong mechanical properties. Lumber bending
strength or modulus of rupture (MOR) and stiffness or modulus
of elasticity (MOE) are among the most important product
properties used to determine the end uses and MSR grade yield
of black spruce (Bruchert et al., 2000; Zhang et al., 2002).Lumber MOE and MOR can be estimated through various non-
destructive testing (NDT) methods (Ross et al., 1997; Wang
et al., 2002b).
It is well-known that both external tree size (e.g., DBH),
stem taper and internal wood characteristics (e.g., knottiness
and wood density) affect the lumber mechanical properties such
as static bending MOE and MOR (Haartveit and Flate, 2002;
Zhang and Chauret, 2001; Zhang et al., 2002). The relationship
of those lumber properties with stand and tree characteristics
(e.g., stand density, DBH, stem taper, total tree height, crown
size, branchiness) would therefore be useful for estimating the
lumber bending properties using inventory data. Use of easily
measured variables on standing trees is preferred if this does notunduly compromise equation accuracy.
Among tree characteristics, DBH is easily measured and is
the most common parameter determined in forest inventory. In
fact, many inventories record DBH only. Tree height is often
recorded for forest inventory. Tree DBH and height are the most
important variables in determining the yield and quality of
lumber due to their effects on the volume and grade of lumber
(Aubry et al., 1998; Fahey, 1980; Houllier et al., 1995; Zhang
and Chauret, 2001). For softwoods, knottiness is an important
internal wood characteristic affecting strength and appearance
www.elsevier.com/locate/forecoForest Ecology and Management 242 (2007) 648–655
* Corresponding author. Tel.: +1 604 222 5741; fax: +1 604 222 5690.
E-mail addresses: [email protected] (C. Liu),
[email protected] (S.Y. Zhang), [email protected]
(A. Cloutier), [email protected] (T. Rycabel).1 Tel.: +1 418 659 2647.2 Tel.: +1 418 656 5851; fax: +1 418 656 2091.3 Tel.: +1 418 656 2438; fax: +1 418 656 2091.
0378-1127/$ – see front matter # 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.foreco.2007.01.077
mailto:[email protected]:[email protected]:[email protected]:[email protected]://dx.doi.org/10.1016/j.foreco.2007.01.077http://dx.doi.org/10.1016/j.foreco.2007.01.077mailto:[email protected]:[email protected]:[email protected]:[email protected]
-
8/19/2019 1-s2.0-S0378112707001211-main
2/8
-
8/19/2019 1-s2.0-S0378112707001211-main
3/8
-
8/19/2019 1-s2.0-S0378112707001211-main
4/8
set to 0.15 and 0.05, respectively. Therefore, all parameters left
in the final regression models after stepwise selection were
significant at the 0.05 level. Because all sample trees were from
six stands with different site quality, the effect of site was
introduced as a class level into the final regression models and
tested for significance using analysis of covariance (ANCOVA).
The models were evaluated based on the multiple coefficient
of determination ( R2), the root mean square error (RMSE), the
mean absolute error (MAE) and error index (EI) of the
predictions, and bias. The R2, RMSE, MAE, RE, and bias were
computed as follows:
R2 ¼ 1
PðY j ˆ Y jÞ
2PðY j ¯ Y Þ
2 (3)
RMSE ¼
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPm j¼1 ðY j
ˆ Y jÞ2
m
s (4)
MAE ¼Pm
j¼1 jY j ˆ Y jj
m(5)
EI ¼ jY j ˆ Y jj (6)
bias ð%Þ ¼ð1=mÞ
Pm j¼1ðY j
ˆ Y jÞ
¯ Y 100 (7)
where Y j and ˆ Y j are the observed response variable value of tree
j and predicted response variable value of tree j for each
regression model, respectively; ¯ Y is the mean of observed
response variable value for the m trees; j = 1, 2, . . ., m; and
m is the number of trees. The RMSE, MAE, and EI provide an
indication of model fit due to no cancellation between positiveand negative values. Positive bias represents underprediction by
the model and negative bias represents overprediction by the
model. For model comparison, those statistical criteria were
evaluated based on the test data set. For each regression model,
the normality of distribution of residuals (observed–predicted),
heteroscedasticity, and multicollinearity were checked by using
the Shapiro–Wilk test (Shapiro and Wilk, 1965), visually by
plotting the residuals against the estimated lumber volume, and
using the maximum variance inflation factor (MVIF), respec-
tively. SAS (SAS Institute Inc., 1999) was used for all compu-
tations.
3. Results
3.1. Model development
Table 3 presents the results of the stepwise linear regression
equations using the fit data set. All parameters in the equations
after stepwise selection were statistically significant at the 0.05
probability level. In the two models, maximum VIFs in the
range of 1.432–3.068 indicated that multicollinearity was not
severe. MOE was best predicted by wood density, DBH, and
crown length, and this model explained 65% of the total
variance in MOE (model 1). In the MOR model (model 2), only
two variables (DBH and wood density) were selected as
predictors by stepwise regression.
Although wood density is an important factor for bending
strength and stiffness, the variable is much more difficult to
measure than tree characteristics. Since the objective of this
study was to predict the MOE and MOR using the input data
from forest inventory, we further conducted the stepwise
regression analysis without wood density variable. The resultsare listed in Table 3. From Table 3, when wood density was
excluded from model 1, stem taper and stand density were
added into model 1a at the 0.05 probability level. Based on
statistical criteria such as R2, RMSE and MAE, it was expected
that the model 1a seemed to have lower prediction precision
than the model 1. In the MOR model (model 2a), DBH became
the best single predictor of MOR. Approximately 58% of the
variance in MOR was explained by this model (Table 3).
The parameter estimates in the above four models were
obtained by ordinary least squares (OLS) method. Like the
study by Lei et al. (2005), the SUR method was used to estimate
the parameters in the system of two equations (i.e., models 1band 2b). The results are presented in Table 3. From Table 3, the
model 1b had comparable prediction precision to the model 1a
in terms of R2, RMSE, and MAE. In the MOR model, the
endogenous variable MOE was by far the best single predictor
of MOR based on the t -statistics and its p-value of less than
0.0001, followed by DBH.
Results of the Shapiro–Wilk test indicated that the normality
assumption for each regression model was held with a p-value
greater than 0.05. The plots of residuals (observed–predicted)
confirmed that the residuals for all the developed regression
models were evenly distributed over the predicted product
Table 3
Results for the linear regression equations for MOE and MOR
Model number Fitted regression equation Fit data set Test data set
R2 RMSE MAE R2 RMSE MAE
1 MOE = 2705.36 176.34DBH + 175.31CL + 21808WD 0.65 1213.20 882.38 0.55 1109.07 1058.652 MOR = 21.18 1.6273DBH + 128.02WD 0.68 8.46 6.28 0.53 9.49 7.68
1a MOE = 17,065 222.38DBH + 261.78CL 3094.96ST + 0.41699SD 0.55 1366.31 1035.35 0.43 1117.36 1363.112a MOR = 100.78 2.4412DBH 0.58 9.60 7.33 0.50 10.24 9.48
1b MOE = 17031.74 222.49DBH + 262.38CL 3083.06ST 0.43011SD 0.55 1366.31 1035.07 0.43 1116.99 1362.362b MOR = 25.98 1.3242DBH + 0.00455MOE 0.79 6.82 5.30 0.72 7.23 5.84
Note. CL, crown length; DBH, diameter at breast height; SD, standdensity; ST, stem taper; WD, wood density; R2, multiple coefficient of determination; RMSE, root
mean square error; MAE, mean absolute error.
C. Liu et al. / Forest Ecology and Management 242 (2007) 648–655 651
-
8/19/2019 1-s2.0-S0378112707001211-main
5/8
recovery and had no obvious trend. Results of testing the effect
of site using ANCOVA analysis showed that the site effect in
each of six models was not significant at the probability level of
0.05. It indicated that the six equations developed here could be
used across all the sites in the study area. However, MOE and
MOR data from the 55 sample trees collected as the
independent data set were needed to further validate the
accuracy of all the developed models.
3.2. Model evaluation
With the test data (55 observations), MOE and MOR were
predicted using the six models under study. The root mean
square error (RMSE) and mean absolute prediction error
(MAE) for the test data set are also summarized in Table 3. For
each model, the difference in R2, RMSE, and MAE values
between the fit data set and test data set were relatively small.
This indicated that those models had an ability to estimate
MOE/MOR when using new inputs in the test data set. Model 1
had lower RMSE and MAE values than model 1a for the testdata set (Table 3). It appears that, on average, model 1 including
wood density predicted the MOE slightly more accurately than
did models 1a and 1b. For MOR models, it was evident that
model 2b in which MOE was included increased predictive
ability when compared to models 2 and 2a in terms of RMSE,
and MAE (Table 3).
To evaluate the model performance across tree sizes, the
computed prediction bias were averaged over 2-cm diameter
classes and listed in Table 4. It was evident that all MOE models
had relatively large biases for large-sized (24-cm DBH class)trees. It appears that all models estimated the MOE more
accurately for small- and medium-sized trees (10–22 cm DBHclasses). For MOE, the overall biases of models 1, 1a, and 1b
were 3.22%, 3.40%, and 3.41%, respectively. This indicated that
the three models produced underestimations for MOE. From
Table 4, we have noticed that the MOR models performed better
and had smaller biases for medium-sized trees (12–22 cm DBH
classes). On the other hand, they gave relatively larger MOE
values for the 10 cm DBH class and the 24 cm DBH class or
above in the test data set (Table 4), especially models 2 and 2a.
To assess the statistical significance of the prediction error
index (EI) for the different models, we performed paired t tests,
which can be used to compare the difference (pairing by tree)
between the error indices for any two models for each response
variable (Zhang et al., 2003). The normality assumption for the
prediction error index was assessed using the Shapiro–Wilk test
(Shapiro and Wilk, 1965) before we applied the paired t test.
Results indicated that the normality assumption held for the
prediction error index from each regression model. Table 5
presents the results of these paired t tests for each possible pair
of models. A negative paired t statistic implies that modelrowhas a comparatively better fit than modelcolumn, because the
difference between the error indices of two models is calculated
as modelrow modelcolumn, where row and column refer to theposition of the fitted models being compared in Table 5. In this
context, Table 5 shows that models 1a and 1b for MOE had
significantly higher error indices than model 1. For example,
the comparison between models 1 and 1a resulted in t = 3.02( p = 0.0040). Clearly, for the MOE, the model 1a had
insignificant difference in error index than the model 1b atthe probability level of 0.05 (Table 5). For the MOR, the
difference between the error indices for any two models was
significant at the probability level of 0.05.
4. Discussion
Three models (models 1, 1a, and 1b) to predict MOE and
three models (models 2, 2a, and 2b) to predict MOR were
developed using stepwise regression analysis. Explanatory
variables were stand and tree characteristics, and wood
properties such as wood density. Wood density and DBH were
selected to predict MOR (model 2) and crown length along withthe two variables were selected to predict MOE (model 1) by
stepwise regression. The signs of the regression coefficients
were as expected as for all variables in the two models. Wood
density was positively correlated with MOE and MOR. Some
studies (Via et al., 2003; Aubry et al., 1998; Oja et al., 2000,
2001) reported that density was a positive contributor to the
MOE and MOR of softwood species. Log DBH contributed
negatively to both static bending and stiffness of natural black
Table 4
The bias (%) in each DBH class and overall bias (%) for model validation using the test data set
DBH class Number of trees MOE MOR
Model 1 Model 1a Model 1b Model 2 Model 2a Model 2b
10 5 5.41 7.02 7.02 16.12 15.01 11.7412 8 7.49 9.19 9.18 8.10 10.79 2.41
14 10 1.59 0.40 0.39 5.54 2.55 3.7516 6 0.08 3.81 3.81 7.35 0.96 6.5918 3 5.26 8.11 8.15 0.94 0.79 2.1820 6 1.79 4.17 4.15 3.65 0.35 1.1822 7 8.48 4.11 4.13 10.63 8.43 0.10
24 3 12.41 13.78 13.80 12.86 19.35 1.66
26 2 10.47 9.53 9.53 20.09 28.98 7.79
28 2 14.17 9.25 9.27 29.65 26.27 20.20
30 3 4.51 12.15 12.15 19.19 43.14 19.42
Total 55 3.22 3.40 3.41 1.49 4.10 1.41
C. Liu et al. / Forest Ecology and Management 242 (2007) 648–655652
-
8/19/2019 1-s2.0-S0378112707001211-main
6/8
spruce trees like the findings in other tree species such as
Norway spruce (Haartveit and Flate, 2002) and four softwood
species (Wang et al., 2002a). Crown length positively
influenced static bending strength and stiffness. From
Table 2, crown width seemed to be closely related to both
MOE and MOR. Becauseit had a strong linear relationship with
DBH ( R2 = 0.51), crown length was not included in the modelsby stepwise regression, through efforts to minimize multi-
collinearity effects by excluding sets of highly correlated
variables (Freund and Littell, 2000).
MOE and MOR are important indicators for the quality of
wood products. The use of MOE and MOR models allowed
industries to estimate thelumber quality of individual trees based
on stand and tree characteristics from forest inventory. However,
forest inventory very rarely collects the wood density data,
although wood density had impacts on bending strength and
stiffness. Models 1a and 2a were useful in a practical sense when
wood density was excluded from the two models. Based on the
model 1a, the impact of either stand density or stem taper wasstatistically significant at the probability level of 0.05. Large
values of stemtaper could affect MOE negatively (Haartveit and
Flate, 2002). Stand density had a positive effect on MOE due to
its effects on growth rate, wood density and branchiness.
The system of two equations (i.e., models 1b and 2b) were
obtained by SUR method. The selected variables in the two
equations and values of R2 were similar to those in the study by
Lei et al. (2005), although the estimated parameters wereobviously different due to different types of black spruce stands
(i.e., plantations in the study of Lei et al. (2005) and natural
stands in this study). MOE was found to be the most important
predictor for bending strength like other studies (e.g., Castera
et al., 1996; Divos and Tanaka, 1997; Green et al., 2001, 2004;
Lei et al., 2005).
Watt et al. (2006) reported that site had a highly significant
effect on MOE across 21 plots covering a wide environmental
range in New Zealand. Results from ANCOVA in this study
showed that the site effect on MOE and MOR was not
significant at the probability level of 0.05. This probably was
because the range of site quality (i.e., from 5.5 to 12.5 m at baseage of 50 years) was still not large enough in our data set. Onthe
other hand, the models developed here were able to account for
variance in MOE or MOR across sites in the study area due to
the insignificant site effect. In this study, all the trees were
within the narrow age range (90–100 years). Therefore, the tree
age was not included in the models.
Since we were more interested in estimating MOE and MOR
using easily measured variables from forest inventory
information, the scatter plots of observed and predicted
MOE and MOR for models 1a and 2a are illustrated in
Fig. 1. It can be observed that the regression line for the two
regression models was close to a straight line. Hence, these
results show that model 1a and 2a could be applied to estimateMOE and MOR from measurements of DBH, crown length,
stem taper, and stand density for black spruce natural stands in
the study area. However, when wood density is available, it is
preferable to use models 1 and 2 in order to improve the
prediction accuracy. And when observed MOE from various
non-destructive testing methods are available, model 2b should
be used to predict MOR more accurately.
5. Conclusion
Lumber static bending modulus of elasticity (MOE) and
modulus of rupture (MOR) are important quality parameters for
Table 5
Comparison of prediction error indices in the two response variables for the models by paired t test based on the test data set
MOE MOR
Model 1 Model 1a Model 1b Model 2 Model 2a Model 2b
Model 1 – 3.02 (0.0040) 3.02 (0.0040)Model 1a – 1.54 (0.1275)
Model 1b –Model 2 – 2.56 (0.0135) 2.74 (0.0085)Model 2a – 4.22 (0.0001)
Model 2b –
Note. Numbers in parentheses are p values for the tests.
Fig. 1. Observed MOE and MOR against predicted MOE and MOR for (a)
model 1a and (b) model 2a, respectively.
C. Liu et al. / Forest Ecology and Management 242 (2007) 648–655 653
-
8/19/2019 1-s2.0-S0378112707001211-main
7/8
the lumber industry. In this study, a stepwise regression method
was employed to identify the best variables for estimating the
MOE and MOR based on stand and tree characteristics for
black spruce stands grown in Eastern Canada. When modeling
MOE using wood properties and stand and tree characteristics
(model 1), the best explanatory variables were wood density,
DBH, and crown length. They explained approximately 65% of
the total variance in MOE. When only stand and tree
characteristics (models 1a and 1b) were used, the best
explanatory variables were DBH, crown length, stem taper,
and stand density. The four variables explained approximately
55% of the total variance in MOE. In terms of the error index,
the difference in prediction accuracy between the models 1 and
1a (or 1b) for the test data set was statistically significant at the
probability level of 0.05. MOE as an exogenous variable in the
system (i.e., models 2a and 2b) was successfully used to predict
MOR. However, MOR could also be predicted using DBH
alone or using both DBH and wood density although the
prediction accuracy was significantly different based on the
independent test data set. Two models (models 1a and 1b)involving only stand and tree characteristics can be used for
estimating MOE and MOR. In conclusion, the developed MOE
and MOR models in this study were based on stand and tree
characteristics and/or an MOE model that can be obtained by
static, dynamic bending vibration and dynamic longitudinal
vibration as well as statistical approach (Divos and Tanaka,
1997). This study provides an approach to modeling wood
properties from tree and stand characteristics in natural black
spruce stands.
In this paper, we have focused mainly on the development of
models for lumber static bending stiffness and strength models
of natural black spruce stands. The results of this study arelimited to naturally grown black spruce trees. The density of
wood (lumber) in plantation trees may be different from that in
natural stands. Zhang et al. (2002) reported that lumber bending
stiffness and strength from plantation-grown black spruce are
significantly lower than from natural black spruce stands.
Therefore, the results (equation parameter estimates and
predictions) obtained in this study, along with those for
plantation-grown black spruce (Lei et al., 2005), would be very
useful for estimating lumber static bending stiffness and
strength for both natural and plantation-grown black spruce
stands.
References
Aubry, C.A., Adams, W.T., Fahey, T.D., 1998. Determination of relative
economic weights for multitrait selection in coastal Douglas–fir. Can. J.
Forest Res. 28, 1164–1170.
Beauregard, R.L., Gazo, R., Ball, R.D.,2002. Grade recovery, value, and return-
to-log for the production of NZ visual grades (cuttings and framing) and
Australian machine stress grades. Wood Fiber Sci. 34 (4), 485–502.
Briggs, D.G., 1989. Tree value system: description and assumptions. General
Technical Report Pacific Northwest Research Station, USDA Forest Ser-
vice. No. PNW-GTR-239, 24 pp.
Bruchert, F., Becker, G., Speck, T., 2000. The mechanics of Norway spruce
[Picea abies (lL.) Karst ]: mechanical properties of standing trees from
different thinning regimes. Forest Ecol. Manage. 135, 45–62.
Castera, P., Faye, C., Ouadrani, A.E., 1996. Prevision of the bending strength of
timber with a multivariate statistical approach. Ann. Sci. Forest 53, 885–
898.
DeBell, J.D., Tappeiner, J.C., Krahmer, R.L., 1994. Branch diameter of western
hemlock: effects of precommercial thinningand implications for log grades.
Western J. Appl. Forest. 9 (3), 88–90.
Denig, J., Wengert, E.M., Brisbin, R., Schroeder, J., 1984. Structural lumber
grade and yield estimates for yellow-poplar sawlog. Forest Prod. J. 35, 26–
35.Divos, F., Tanaka, T., 1997. Lumber strength estimation by multiple regression.
Holzforschung 51, 467–471.
Fahey, T.D., 1980. Grading second-growth Douglas–fir by basic tree measure-
ments. J. Forest.
Freund, R.J., Littell, R., 2000. SAS System for Regression, 3rd ed. Wiley and
Sons Publisher, New York, USA, 236 pp.
Green, D.W., Falk, R.H., Lantz, S.F., 2001. Effect of heart checks on flexural
properties of reclaimed 6 by 8 Douglas–fir timbers. Forest Prod. J. 51, 82–
88.
Green, D.W., Gorman, T.M., Evans, J.W., Murphy, J.F., 2004. Improved
grading system for structural logs for log homes. Forest Prod. J. 54,
52–62.
Haartveit, E.Y., Flate, P.O., 2002. Mechanical properties of Norway spruce
lumber from monocultures and mixed stand—modeling bending stiffness
and strength using stand and tree characteristics. In: Springs, H.H. (Ed.),Connection Between Silviculture and Wood Quality: Modeling Approach
and Simulation Software, IUFRO WP S5.01-04 Workshop, British Colum-
bia, Canada, September.
Houllier, F., Leban, J.M., Colin, F., 1995. Linking growth modelling to timber
quality assessment for Norway spruce. Forest Ecol. Manage. 74, 91–102.
Kellogg, L., Kennedy, R.W., 1986. Implication of Douglas–fir wood quality
relative to practical end use. In: Oliver, C.D., Hanley, D.P., Johnson, J.A.
(Eds.), Douglas–fir: stand management for the future. University of
Washington, College of Forest Resources, Seattle, WA, Institute of Forest
Resources Contribution 55, pp. 73–78.
Kellogg, R.M., Warren, W.G., 1984. Evaluating western hemlock stem char-
acteristics in terms of lumber value. Wood Fiber Sci. 16 (4), 583–597.
Lei, Y.C., Zhang, S.Y., Jiang, Z., 2005. Models for predicting lumber bending
MOR and MOE based on tree and stand characteristics in black spruce.
Wood Sci. Technol. 39, 37–47.Liu, C., Zhang, S.Y., 2005. Models for predicting product recovery using
selected tree characteristics of black spruce. Can. J. Forest Res. 35, 930–
937.
Moguedec, G.L., Dhote, J.F., Nepveu, G., 2002. Choosing simplified mixed
models for simulations whendata havea complex hierarchical organization:
an example with some basic properties in Sessile oak wood (Quercus
petraea Liebl). Ann. Forest Sci. 59, 847–855.
NLGA, 1996. Standard Grading Rules for Canadian Lumber. National Lumber
Grades Authority, Vancouver, BC.
Oberg, J.C., 1989. Impacts on lumber and panel products. In: Proceedings of
the Southern Plantation Wood Quality Workshop, Athen, GA, June 6–7.
Oja, J., Grundberg, S., Gronlund, A., 2000. Predicting the strength of sawn
products by X-ray scanning of logs: a preliminary study. Wood Fiber Sci.
32, 203–208.
Oja, J., Grundberg, S., Gronlund, A., 2001. Predicting the stiffness of sawnproducts by X-ray scanning of Norway spruce saw logs. Scand. J. Forest
Res. 16, 88–96.
Prestemon, J.P., Buongiorno, J., 2000. Determinants of tree quality and lumber
value in natural uneven-aged southern pine stands. Can. J. Forest Res. 30,
211–219.
Ross, R.J., McDonald, K.A., Green, D.W., Schad, K.C., 1997. Relationship
between log and lumber modulus of elasticity. Forest Prod. J. 42, 89–92.
SAS InstituteInc., 1999. SAS/STAT User’s Guide. Version 8. SAS Institute Inc.,
Cary, NC.
Shapiro, S.S., Wilk, M.B., 1965. An analysis of variance test for normality
(complete samples). Biometrika 52, 591–611.
Tong, Q.J., Zhang, S.Y., Thompson, M., 2005. Evaluation of growth response,
stand value and financial return for pre-commercially thinned jack pine
stands in northwestern Ontario. Forest Ecol. Manage. 209, 225–235.
C. Liu et al. / Forest Ecology and Management 242 (2007) 648–655654
-
8/19/2019 1-s2.0-S0378112707001211-main
8/8
Via, B.K., Shup, T.F., Groom, L.H., Stine, M., So, C.L., 2003. Multivariate
modelling of density, strength and stiffness from near infrared spectra for
mature, juvenile and pith wood of longleaf pine (Pinus palustris). J. Near
Infrared Spectrosc. 11, 365–378.
Wang, X., Ross, R.J.,Brashaw, B.K.,Erickson, J.R.,Forsman, J.W., Pellerin, R.,
2002a. Diameter effect on stress-wave evaluation of modulusof elasticity of
logs. In: Beall, F.C. (Ed.), Proceedings of the 13th International Symposium
on Non-destructive Testing of Wood. Forest ProductsSociety, Madison, WI,
USA.Wang, X., Ross, R.J., Mattson, J.A., Erickson, J.R., Forsman, J.W., Geske, E.A.,
Wehr, M.A., 2002b. Nondestructive evaluation techniques for assessing
modulus of elasticity and stiffnessof small-diameter logs. Forest Prod. J. 52,
79–85.
Watt, M.S., Moore, J.R., Façon, J.P., Downes, G.M., Clinton, P.W., Coker, G.,
Davis, M.R., Simcock, R., Parfitt, R.L., Dando, J., Mason, E.G., Bown,
H.E.,2006. Modellingthe influence of standstructural, edaphic and climatic
influences on juvenile Pinus radiata dynamic modulus of elasticity. Forest
Ecol. Manage. 229, 136–144.
Zellner, A., 1962. An efficient method of estimating seemingly unrelated
regressions and tests for aggregation bias. J. Am. Statist. Assoc. 57,
348–368.
Zhang, L., Packard, K.C., Liu, C., 2003. A comparison of estimation methods
for fitting Weibull and Johnson’s S B distributions to mixed spruce–fir stands
in northeastern North America. Can. J. Forest Res. 33, 1340–1347.
Zhang, S.Y., Chauret, G., 2001. Impact of initial spacing on tree and wood
characteristics, product quality and value recovery in black spruce (Picea
mariana). CFS Rep. No.35, Forintek Canada Corp., Sainte-Foy, Quebec.Zhang, S.Y., Chauret, G., Swift, E., 2001. Maximizing the value of jack pine
through intensive forest management. CFS Rep. No. 3171, Forintek Canada
Corp., Sainte-Foy, Quebec.
Zhang, S.Y., Chauret, G., Ren, H.Q., Desjardins, R., 2002. Impact of plantation
black spruce initial spacing on lumber grade yield, bending properties and
MSR yield. Wood Fiber Sci. 34 (3), 460–475.
Zhang, S.Y., Tong, Q.J., 2005. Modellingsimulated product recovery in relation
to tree characteristics in jack pine using sawing simulator Optitek. Ann.
Forest Sci. 62, 219–228.
C. Liu et al. / Forest Ecology and Management 242 (2007) 648–655 655