Spectrometric prediction of wood basic density by comparison of different grain angles and variable selection methods

Area of Science:

• Wood Science
• Spectroscopy
• Data Analysis

Background:

• Wood basic density (WBD) is critical for industrial applications but traditionally measured destructively.
• Developing non-destructive, rapid WBD measurement methods is essential for forestry and wood industries.
• Near-infrared (NIR) spectroscopy presents a promising alternative for WBD quantification.

Purpose of the Study:

• To evaluate the efficacy of NIR spectroscopy coupled with partial least squares regression (PLSR) for quantifying WBD in multiple wood species.
• To investigate the influence of grain angle on NIR spectral data acquisition and model accuracy.
• To compare variable selection methods for optimizing NIR spectral data and enhancing prediction models.

Main Methods:

• Collected solid wood NIR spectra at different grain angles (0°, 45°, 90°).
• Employed PLSR for WBD prediction.
• Utilized variable selection methods: significant Multivariate Correlation (sMC), Regularized elimination procedure (Rep), Iterative predictor weighting (Ipw), and Genetic algorithm (Ga).
• Validated models through 200 random calibration data selections.

Main Results:

• Models using 90° grain angle spectra showed the highest efficiency.
• A mixed-angle model effectively reduced grain angle influence and provided satisfactory WBD predictions.
• The Rep method outperformed other variable selection techniques in enhancing model performance.
• NIR spectroscopy combined with PLSR demonstrated robust WBD prediction capabilities.

Conclusions:

• NIR spectroscopy provides a rapid and efficient method for measuring WBD in solid wood across different grain angles and species.
• The developed methodology, utilizing NIR and PLSR, can be a valuable tool for wood property breeding programs.
• This approach supports advancements in silviculture studies by enabling efficient wood quality assessment.