Prediction of wood and charcoal apparent density using benchtop and portable near-infrared instruments based on X-ray densitometry data

The increasing demand for high-quality wood and charcoal underscores the importance of monitoring key properties such as density. Near-infrared (NIR) spectroscopy combined with X-ray densitometry offers a rapid, non-destructive approach for estimating apparent density. This study aimed to predict the apparent density of wood and charcoal from 15 clones of Eucalyptus and Corymbia (84 months old) using NIR spectroscopy with a benchtop (MPA) and two portable (MicroNIR and Trinamix) instruments integrated with X-ray densitometry. A 10 cm thick wood disc was collected from each tree at breast height and sectioned radially from pith to bark into 10 subsamples (2 mm thick). Five subsamples were analyzed in their natural state, and the remaining five were carbonized in a muffle furnace. X-ray images and NIR spectra were acquired before and after carbonization. Five equidistant points were defined per sample for densitometric and spectral analyses, resulting in 375 variables per instrument for both samples. Principal component analysis explained more than 99% of the cumulative variance across instruments and materials. For wood, partial least squares regression models achieved prediction coefficients of 0.87 (MPA), 0.75 (MicroNIR), and 0.67 (Trinamix). For charcoal, the best models had prediction coefficients of 0.71 (MPA), 0.62 (MicroNIR), and 0.60 (Trinamix). Overall, NIR spectroscopy showed potential for estimating apparent density. The benchtop instrument outperformed portable devices, providing reliable wood models for clone selection and quality control and suitable charcoal models for preliminary screening. These findings reinforce the potential of integrating NIR spectroscopy and X-ray densitometry in wood and charcoal characterization. • Combination of NIR spectroscopy and X-ray densitometry enabled prediction of wood and charcoal apparent density. • PCA explained more than 99% of spectral variability but did not clearly discriminate clones within genera. • PLS-R models showed high predictive performance for wood, reaching R 2 p = 0.87. • For charcoal, satisfactory prediction was achieved with benchtop NIR (R 2 p = 0.71). • Portable NIR instruments offer lower cost and operational flexibility, despite reduced predictive accuracy.