A multi-site, multi-species near infrared calibration for the prediction of cellulose content in eucalypt woodmeal

Calibrations between spectra derived from near infrared (NIR) spectroscopy and chemical analyses can be used to predict cellulose content in samples of ground eucalypt woodmeal. Past calibrations have used small sample sets (<100 samples) often representing a single stand or region and only one species, making their application to wider populations of samples problematic. A robust, non-destructive prediction capability for eucalypt wood cellulose that works across stands, regions and species would find many applications in tree breeding and resource assessment. Here, we describe the development and test the performance of a large (>1000 samples) multi-site and species NIR calibration for predicting cellulose content of eucalypt woodmeal obtained from increment cores, wood chips and stem cross-sections. Most of the samples came from Eucalyptus globulus and E. nitens. The calibration was tested against four independent sample sets and explained between 65% and 84% of the variance in each set. A higher proportion of variance was explained in those sample sets that had a wider range of cellulose content. Standard errors of prediction were between 0.5% and 1.5% cellulose for the four independent sample sets. The test samples were added to the large calibration and a new calibration with 1260 samples was constructed. Principal components analysis suggests additional wood samples with more diverse chemistries are required to enable the calibration to capture more fully the chemical variation present in the genus.