Mapping within-stem variation of chemical composition by near infrared hyperspectral imaging

A near infrared ( NIR)-imaging spectrograph was used to generate maps of chemical composition distribution on the surface of transverse wood discs taken from tree stems. The measured chemical components were lignin, galactose, glucose and mannose as well as cellulose and hemicelluloses, which were calculated from monomeric sugars. These components were determined by NIR-based chemistry models, which had been developed specifically for the imaging spectrograph. Explained test-set variation for key constituents ranged between 60% (galactan) and 78% (lignin). Day-to-day variability was 1-2% (Stdev/ range) depending on the chemical property. Various operational parameters such as room temperature, sample temperature, sample surface preparation and sample thickness were found to have a non-negligible, but manageable, influence on predicted results. The influence of room and sample temperature could be reduced by incorporating temperature changes into the chemistry model. Extractives, transported to, and concentrated at, the disc surface during drying, needed to be physically removed from the surface to avoid an unpredictable influence on chemical results. Wood fibre angles at the disc surface needed to be aligned in a consistent manner to the camera. NIR information was found to derive from a sample depth of up to 10 mm. This distance was consequently chosen as the minimal sample thickness.