3.1. Board Treatment Stages
Data followed normality and homogeneity of variance. The dynamic board modulus of elasticity measured at different treatment stages is represented in
Figure 1. There was a general increase in MOE
dyn from the green to the kiln-dry stage, after which the MOE
dyn recorded for the dressed stage decreased. Differences in MOE
dyn were due both to the position of the log in the tree and to the treatment; however, the interaction between log position and treatment was not significant (
Table 2).
Air-dry and kiln-dry MOE
dyn values were significantly higher than green MOE
dyn for all log positions (
Figure 1). Dressed MOE
dyn was not significantly different between green MOE
dyn for boards coming from the first, second, and top logs (log A t(106) = 2.53,
p = 0.07, log B t(83) = 2.31,
p = 0.14, log D t(14) = 0.09,
p = 1), but significantly different for boards coming from third logs (log C t(65) = 3.42,
p < 0.05). Dressed MOE
dyn was also significantly lower than kiln-dry MOE
dyn (log A t(106) = 17.7,
p < 0.001, log B t(83) = 13.9,
p < 0.001, log C t(65) = 20.5,
p < 0.001, log D t(14) = 6.72,
p < 0.001).
There were no significant differences between modulus of elasticity measured at the green stage or dressed stage through AWV and actual modulus of elasticity measured at the dressed stage via mechanical testing (F = 1.94,
p = 0.15); however, significant differences among log positions were present (F = 17.3,
p < 0.001). There was no interaction between log position and measurement type (F = 0.17,
p = 0.99).
Table 3 reports the values of modulus of elasticity at the green and dressed stages measured via AWV (MOE
dyn) and actual modulus of elasticity measured mechanically (MOE
stat).
Log position in the stem influenced the modulus of elasticity of boards (
Figure 1). Boards coming from bottom logs (log A) had consistently lower modulus of elasticity than boards from other positions in all treatment stages, but this relationship was not always significant. Although lower, the modulus of elasticity of boards from bottom logs (A) was not significantly different than that of boards from top logs (D), but significantly lower than that of boards from middle positions in the stems (log B and C).
The correlations between AWV MOE
dyn measured at different stages and actual MOE
stat were strong and significant at the 0.001 level (
Figure 2). Considering specifically the first and last panel, measurement of MOE
dyn on green boards through AWV can explain almost 60% of the variability in actual MOE
stat, and measurement of MOE
dyn on dressed boards through AWV can explain almost 70% of the variability in actual MOE
stat.
3.2. Grading Systems
The results of the allocation of boards to F-grades according to both visual grading (VSG) and AWV are presented in
Table 4 and
Table 5, respectively. These are reported alongside the F-grade of the boards determined using the actual modulus of elasticity of the boards measured through mechanical testing (MOE
stat). The VSG method had a large error, misclassifying the boards with an 82.5% rate of error. 70.2% of the board F-grades were underestimated, while 12.3% were overestimated, in respect to the actual MOE
stat of the boards.
The classification error was considerably lower when utilizing the non-destructive technique of AWV to classify the boards into F-grades. A total error of 45.2% was recorded, with 22% of the boards being underestimated in their grade and 23.5% being overestimated. The indirect measure of MOE
dyn via AWV is correlated with the actual MOE
stat of the boards (last panel of
Figure 2), which is likely the reason for a better classification of boards compared to the VSG method.
Table 6 presents the comparison of AWV MOE
dyn and actual MOE
stat of the F-grades determined through VSG method. No significant difference was detected between the MOE
stat of the grades, except for MOE
stat of F17 and F11 which were found to be significantly different (
p < 0.05). This further strengthens the lack of reliability of the VSG to be applied to plantation
E. nitens timber in detecting the actual stiffness of the boards, either measured as MOE
dyn or MOE
stat.
We found that the features of boards that are mostly correlated with actual MOE
stat were board density (r = 0.66,
p < 0.001), number of sound knots (r = −0.40,
p < 0.001), number of knots (r = −0.37,
p < 0.001), number of major knots (r = −0.27,
p < 0.001), presence of pith in the boards (r = −0.38,
p < 0.001), and presence of checks deeper than 3 mm (r = −0.18,
p < 0.01). Although significantly correlated with the modulus of elasticity of the boards, those features did not contribute to improving the correlation between MOE
dyn measured via AWV and actual MOE
stat. Therefore, we modelled MOE
stat only using MOE
dyn as tested with AWV, without accounting further for board features. The predictions of this model were validated against the observed MOE
stat values and it was found that the model could explain 69% of the variability in actual MOE
stat of the boards with a RMSE of 1.26 GPa (
Figure 3a). The residuals of this model were normally distributed and did not show apparent bias with the fitted values (
Figure 3b).
The regression equation showed that for an increase of 0.67 GPa in MOE
dyn there would be a corresponding increase in MOE
stat of 1 GPa. Using this model, the actual MOE
stat was predicted on the overall dataset, obtaining the classification of boards as presented in
Table 7. The overall error was less than both the VSG and the classification made directly with the MOE
dyn values tested through AWV, for a total of 43.3%. Using the regression equation to predict the MOE
stat of the boards, 18.3% of the boards would be over-estimated, while 25% would be underestimated. In respect to the actual values recorded through AWV, the overestimation is lower, and the underestimation is higher.