Improving spatial synchronization between X-ray and near-infrared spectra information to predict wood density profiles

Wood density is one of the most important physical properties of the wood, used in improvement programs for wood quality of major timber species. Traditional core sampling of standing trees has been widely used to assess wood density profiles at high spatial resolution by X-ray microdensitometry methods, but alternative methods to predict wood properties quality are also needed. Near-infrared (NIR) spectroscopy, a non-destructive technique, is being increasingly used for wood property assessment and has already been demonstrated to be able to predict wood density. However, the estimation of wood density profiles by NIR has not yet been extensively studied, and improved models using spectra information (NIR) and X-ray data need to be developed. To this end, partial least square regression (PLS-R) models for predicting wood density were developed at a 1.4 mm spatial resolution on Pinus pinaster wood cores, with an improved spatial synchronization along the tangential and radial directions of the strip, between X-ray data and NIR spectra. The validation of the best model showed a high coefficient of determination (0.95), low error (0.026) and no outlier. Compression wood samples were not detected as outliers and were correctly predicted by the model. However, pith spectra were detected as outliers and its predicted values were overestimated by 33% due to unusual spectra suggesting a diverse chemical composition. The results suggest that NIR-PLS models obtained can be used for screening maritime pine wood density profiles along the radii at 1.4 mm spatial resolution.