16 Wooden Material and Environmental Sciences
365
NIRS also could contribute to study bound water within wood, and it is a powerful
tool for evaluating molecular water dynamics based on wavelength shift characteristics [15]. Inagaki et al. compared the variation of water adsorption between modern
and archeological wood samples. The wavelength range of 1818–2128 nm due to
the O–H first overtone was selected. Curve-fitting method was used to separate the
baseline-corrected NIR difference spectra into three components that have different
vibrational energy. Ma. et al. used PCA to characterize the variance of NIR spectral
range of 1340–1610 nm due to the O–H second overtone after baseline correction.
The data analysis results showed PC1 loading mainly correlates with wood water
content by mass; however, the PC2 loading values contain the information about
water–wood hydrogen structure interactions [16].
16.2.3 Wood Density
Density is a crucial parameter for wood strength and stiffness, which are critical
considerations for a wooden structure. Most NIR calibration models were constructed
based on light absorption differences caused by the three main chemical wood components (i.e., cellulose, hemicellulose, and lignin). Alves et al. calibrated the maritime
pine and hybrid larch wood density measured by an X-ray densitometer with NIR
spectra using PLS-R analysis [17]. The best PLS-R model could fulfill the requirements for the NIR model development and maintenance guidelines provided by the
American Association of Cereal Chemists (AACC Method 39-00). Santos et al. also
estimated the wood density of Portuguese Blackwood using NIRS combined with
PLS-R analysis [18]. The RPD limit was 2.5, even though the number of spectra
collected from each wood disk was only three. Fujimoto et al. examined the effect
of MC on the accuracy of predicting wood density. They discussed the chemometric
background for the potential to predict the wood density (R
2
= 0.86 − 0.87, SEP =
22 kg m
−3 ) at various moisture conditions [19].
Some studies also focused on light scattering caused by physical wood structure
to predict density. Hans et al. measured seven softwood and hardwood species using
time-of-flight NIRS (TOF-NIRS) which provides additional light scattering information. Then, curve-fitting procedure was used to separate absorption and reduced
scattering coefficients. The square root of the adsorption/scattering ratio could correct
the scattering effects in absorbance NIR spectra [20]. Ma et al. used spatially resolved
spectroscopy (SRS) method, and a NIR imaging camera was utilized to catch the
light scattering patterns on Douglas fir wood surface which illuminated by a concentrated halogen light source (Ø 1 mm). A steady-state diffusion theory model was
applied to estimate light absorption and reduced scattering coefficients. The experimental results indicate that a few key wavelengths could achieve the prediction
of subsurface density and grain direction without relying on multivariate statistical
analysis [20]. Such an approach is worth pursuing further since it will contribute to
the design of a low-cost measurement system and with robust prediction accuracy.
Recently, Kitamura and Tsuchikawa [21] have shown the possibility of designing
365
NIRS also could contribute to study bound water within wood, and it is a powerful
tool for evaluating molecular water dynamics based on wavelength shift characteristics [15]. Inagaki et al. compared the variation of water adsorption between modern
and archeological wood samples. The wavelength range of 1818–2128 nm due to
the O–H first overtone was selected. Curve-fitting method was used to separate the
baseline-corrected NIR difference spectra into three components that have different
vibrational energy. Ma. et al. used PCA to characterize the variance of NIR spectral
range of 1340–1610 nm due to the O–H second overtone after baseline correction.
The data analysis results showed PC1 loading mainly correlates with wood water
content by mass; however, the PC2 loading values contain the information about
water–wood hydrogen structure interactions [16].
16.2.3 Wood Density
Density is a crucial parameter for wood strength and stiffness, which are critical
considerations for a wooden structure. Most NIR calibration models were constructed
based on light absorption differences caused by the three main chemical wood components (i.e., cellulose, hemicellulose, and lignin). Alves et al. calibrated the maritime
pine and hybrid larch wood density measured by an X-ray densitometer with NIR
spectra using PLS-R analysis [17]. The best PLS-R model could fulfill the requirements for the NIR model development and maintenance guidelines provided by the
American Association of Cereal Chemists (AACC Method 39-00). Santos et al. also
estimated the wood density of Portuguese Blackwood using NIRS combined with
PLS-R analysis [18]. The RPD limit was 2.5, even though the number of spectra
collected from each wood disk was only three. Fujimoto et al. examined the effect
of MC on the accuracy of predicting wood density. They discussed the chemometric
background for the potential to predict the wood density (R
2
= 0.86 − 0.87, SEP =
22 kg m
−3 ) at various moisture conditions [19].
Some studies also focused on light scattering caused by physical wood structure
to predict density. Hans et al. measured seven softwood and hardwood species using
time-of-flight NIRS (TOF-NIRS) which provides additional light scattering information. Then, curve-fitting procedure was used to separate absorption and reduced
scattering coefficients. The square root of the adsorption/scattering ratio could correct
the scattering effects in absorbance NIR spectra [20]. Ma et al. used spatially resolved
spectroscopy (SRS) method, and a NIR imaging camera was utilized to catch the
light scattering patterns on Douglas fir wood surface which illuminated by a concentrated halogen light source (Ø 1 mm). A steady-state diffusion theory model was
applied to estimate light absorption and reduced scattering coefficients. The experimental results indicate that a few key wavelengths could achieve the prediction
of subsurface density and grain direction without relying on multivariate statistical
analysis [20]. Such an approach is worth pursuing further since it will contribute to
the design of a low-cost measurement system and with robust prediction accuracy.
Recently, Kitamura and Tsuchikawa [21] have shown the possibility of designing