Impact of moisture and NIR sensors on calibration transfer between predictive models of Eucalyptus grandis wood density
Targino De Medeiros, Dayane | Gomes Batista, Felipe | Reis Prazeres Mascarenhas, Adriano | Chaix, Gilles | Gherardi Hein, Paulo Ricardo
The aim was to evaluate the potential for model transfer between NIR equipment for predicting the basic density of wood at different moisture. Samples of Eucalyptus grandis wood were produced. Basic density was determined by the ratio between anhydrous mass and saturated volume. Benchtop and portable near-infrared spectra were collected on freshly cut wood, at the fiber saturation point and at equilibrium moisture. A script was developed in the R software to make the NIR equipment matrices compatible, totaling four matrices: two with raw dimensions and two with standardized dimensions. PCA and PLS-R analyses were carried out to evaluate the data and obtain the models, which were validated using the leave-one-out and test set methods. Calibration transfer was applied to analyze the models under different instrumental variations. The models developed in each matrix showed R2 ranging from 0.74 to 0.92 and RMSE from 0.03 to 0.06 g/cm3, showing potential for predicting density from raw and compatible data. In the calibration transfer, the R2p reached 0.93 and the RMSEp 2.83 g/cm3. The FSP moisture models showed the best statistical parameters, and the calibrations of the benchtop equipment are more promising for transfer to other equipment.
Показать больше [+] Меньше [-]