Multi-species NIR calibration for estimating holocellulose in plantation timber

Feasibility of near-infrared (NIR) spectroscopy for developing multi-species model for plantation timber was explored for estimation of holocellulose in un-extracted milled wood samples. Six commonly planted species of Eucalyptus tereticornis, E. camaldulensis, E. grandis, Leucaena leucocephala, Dalbergia sissoo and Populus deltoides from a wide range of locations and varying age groups were taken for the present study. Few samples of E. hybrid between E. tereticornis and E. camaldulensis were also included in the study to make the model useful for practical application. NIR models were evaluated using partial least squares regression (PLSR-1—full cross-validation, PLSR-2—cross-validation which leaves more than one out) and by dividing the samples into calibration and prediction (test) sets and interchanging them from calibration to prediction sets. The predictive ability of the model was assessed by calculating four ratios of multivariate statistics for individual species model and combined species models. A final combined model for all the species having component range of 76.14–63.03 % and standard deviation of 2.586 % was developed in the spectral range of 7502–4246 cm^?1 wave number using 1st derivative plus multiplicative scatter correction using factor of nine by removing samples with outliers found in all the PLSR-2 evaluation steps and in most of the models. The model remained stable even when 30 % of the samples were left out with no outlier detected.