Developing and evaluating a multisite and multispecies NIR calibration for the prediction of Kraft pulp yield in eucalypts

Abstract

Over recent years the application of near infra-red (NIR) spectroscopy to the prediction of wood properties has been demonstrated in many proof-of-concept studies. Previous work has demonstrated that NIR measurements can be used to predict basic density from woodmeal, chainsaw dust and solid wood, as well as microfibril angle and modulus of elasticity in solid samples. For over a decade, the prediction of Kraft pulp yield (KPY) has been a constant research focus, and numerous small studies have demonstrated this potential. However, because of the cost of obtaining calibration samples with known KPY, sample numbers are typically less than 100. While the potential for NIR  prediction of KPY is well recognised, the shift to routine commercial use  has not occurred. There still remains considerable scepticism in the research and industry communities about the use of NIR. Concern is typically expressed in two areas: (1) the consistency, accuracy and precision of predictions and (2) the need to prepare a separate calibration for each site and/or species group. To elevate NIR from proof-of-concept to a pilot scale, a large multisite, multispecies  calibration was developed over iterative cycles to: (1) determine whether KPY in eucalypts can be predicted from a single calibration independent of site and species, and (2) identify the potential limits of accuracy and precision. This paper reports the results of the first seven testing cycles. The NIR calibration was expanded from an initial sample  set of 104 mixed eucalypt samples to over 720 samples covering more than 40 species from predominantly temperate sites across Australia. The performance of the final calibration using two independent and contrasting data sets showed that a multisite and multispecies calibration is feasible. The expected potential accuracy and precision that can be expected from NIR predictions is discussed.

Southern Forests 2009, 71(2): 155–164