| Abstract: | Abstract Eighteen sapwood and 18 heartwood samples from three Scots pine (Pinus sylvestris L.) trees were used to identify correlations between the features of their near-infrared spectra (400–2500 nm) measured before and after impregnation with linseed oil (Linogard®) and (1) the amount of oil taken up by the samples at three different longitudinal positions; and (2) the tissue type of sample. Calibration models were cross-validated and oil uptake models were also validated on separate test sets. Partial least squares regression models for the prediction of uptake and tissue type in non-impregnated wood were promisingly accurate, having coefficients of multiple determination (Q 2) values of > 0.8, 0.8 and 0.75 for predictions of tissue type, oil taken up as a percentage of sample mass and total mass of oil taken up, respectively. The models developed for linseed oil-impregnated samples were better still, generating Q 2 values of > 0.9, > 0.8 and 0.8 for the three key properties. In general, models based on spectra acquired along two directions of measurement (longitudinal and tangential) were more accurate than models based on only one. Tandem models consisting of two submodels, one for sapwood and one for heartwood, were more successful than single models applicable to both tissue types. |
