Validation of models using near-infrared spectroscopy to estimate basic density and chemical composition of Eucalyptus wood

Determining the wood properties is fundamental because these properties are directly related to wood quality. The near-infrared (NIR) spectroscopy technique has been used to determine various properties of wood. However, even with promising results, NIR spectroscopy needs to be further investigated to evaluate the robustness of its estimates. The objective of this study was to develop regression models from NIR spectra to estimate the basic density and the extractive and lignin contents of wood as well as to verify their robustness through independent and cross-validation. NIR spectra were initially obtained through an integration sphere and optical fiber for the transverse and radial faces of solid wood and through an integration sphere for powdered wood. The wood basic density and the extractive and lignin contents were determined by conventional methods in 180 and 143 specimens, respectively. The samples were collected from Eucalyptus urophylla × Eucalyptus grandis clones aged 5 years. The basic density and extractive and lignin content values were correlated with the NIR spectra by a partial least squares regression. The best models for estimating the basic density of the wood were generated from the spectra obtained on the transverse surface, both with the integration sphere pathway and in the optical fiber pathway. For estimating the chemical properties of wood, the best models were developed from the powdered wood via the integration sphere for assessing the extractive content, Klason lignin, acid-soluble lignin content and total lignin. In all the models, the mathematical treatment of the data by the first derivative was essential for better fitting the models and reducing the error. We concluded that NIR spectroscopy was effective for the estimation of basic density and extractive and lignin contents of wood.