Evaluating basic density calibrations based on NIR spectra recorded on the three wood faces and subject to different mathematical treatments

Background: Near infrared (NIR) spectroscopy has been successfully applied to estimate the chemical, physical and mechanical properties of various biological materials, including wood. This study aimed evaluate basic density calibrations based on NIR spectra collected from three wood faces subject different mathematical treatments. Methods: Diffuse reflectance were recorded using an integrating sphere transverse, radial tangential surfaces 278 specimens Eucalyptus urophylla x grandis. Basic was determined in laboratory by immersion method correlated with Partial Least Squares regression. Different statistical treatments then data, Standard Normal Variate, Multiplicative Scatter Correction, First Second Derivatives, Normalization, Autoscale MeanCenter transformations. Results: The predictive model measured transverse surface performed best (R²cv = 0.85 RMSE 25.5 kg/m³) while developed had poorest performance 0.53 46.8 kg/m³). difference between models original (untreated) mathematically-treated minimal. Conclusions: Multivariate fitted found be efficient for predicting wood, especially when transversal surface. For this data set, mathematically treated similar performance. reported findings show that transformations are not always able extract more information NIR.