Prediction of Wood Mechanical and Chemical Properties in the Presence and Absence of Blue Stain Using Two near Infrared Instruments

The objective of this research was to (a) determine if blue stain in solid wood influenced calibration equations developed from a non-stained wood population, (b) assess the bias introduced when scanning was performed by the slave instrument without calibration transfer from the master instrument and (c) partition absorbance-based variation by instrument, stain and instrument × stain interaction. The results helped to determine the calibration transfer needed for this case. The dependent variables assessed from clear and stained wood were lignin, extractives, modulus of elasticity (MOE), modulus of rupture (MOR) and density When the master instrument was used for both calibration and prediction, it was found that stain-insensitive equations for the five traits could be built. However, when a slave near infrared instrument was introduced without calibration transfer, three out of five predicted traits were significantly biased by the presence of stain. Further analysis revealed an interaction between stain and instrument indicating that instrument bias was also introduced during scanning with a slave. For both multiple linear regression (MLR) and principal components regression (PCR), it was found that if a trait needed more wavelengths (or principal components) for prediction of the dependent variable, bias due to blue stain became increasingly prominent. PCR was found to perform better than MLR when stain was introduced with no calibration transfer. Such a finding alludes that PCR works better than MLR under extrapolation conditions but is not intended to support a lack of calibration transfer. Finally, the Mallows C_p diagnostic proved valuable in model selection although the well-known requirement of (C_p – p ≤ 0) appeared conservative. For MLR and PCR, a C_p – p ≤ 5 often yielded applicable models while C_p – p > 7 was about the threshold where model performance dropped.