Comparative performance of bench and portable near infrared spectrometers for measuring wood samples of two Eucalyptus species (E. pellita and E. benthamii)

Abstract

The well-established and ever-growing applications of near infrared (NIR) spectroscopy for wood science and technology using bench spectrometers is unquestionable. However, the increasing interest in using portable spectrometers due to their perceived advantages raises the question of how good are they? In this work, we acquired spectra from wood sawdust samples of two Eucalyptus species, E. pellita and E. benthamii, totaling 1200 individual trees sampled in experimental breeding populations. Spectra were acquired in parallel using a bench spectrometer (FOSS NIRSystems 5000) and a portable instrument (Viavi MicroNIR1700). Spectra acquisition on the bench spectrometer had a longer wavelength (1100-2500 nm versus 908-1700 nm) and a shorter absorbance profile. Principal component analysis on the spectral data of both instruments was not able to clearly separate the two species suggesting some level of overlapping distributions of the chemical composition of their respective wood properties. Linear discriminant analysis, however, had excellent accuracies with the bench spectrometer (~99 %), showing significantly better discrimination than the one obtained with the portable one (~93 %).Similarly, partial least square discriminant analysis showed correlations (r) around 0.96 and standard errors of calibration (SEC) and of cross validation (SECV) lower than 0.155 on the bench instrument, while the portable spectrometer had r ~ 0.88 with SEC and SECV below 0.235. Although the bench instrument showed a better performance, the parameters estimated with the portable spectrometer were very satisfactory given its intrinsic limitations in robustness and handling needs. Chemical analyses for lignin content are in progress which, combined with data transformation and selection of spectra regions, could put the two instruments on similar performance grounds. These results are particularly relevant for rapid and simple wood phenotyping applications in advanced tree breeding operations.