Measurement of cellulose content, Kraft pulp yield and basic density in eucalypt woodmeal using multisite and multispecies near infra-red spectroscopic calibrations
Downes, GM and Meder, R and Bond, H and Ebdon, N and Hicks, C and Harwood, CE, Measurement of cellulose content, Kraft pulp yield and basic density in eucalypt woodmeal using multisite and multispecies near infra-red spectroscopic calibrations, Southern Forests, 73, (3-4) pp. 181-186. ISSN 2070-2620 (2011) [Refereed Article]
Previous descriptions of multisite and multispecies near infra-red (NIR) spectroscopic calibrations for predicting cellulose content (CC) and Kraft pulp yield (KPY) in eucalypt woodmeal demonstrated that large, single calibrations provide precise predictions for a wide range of sites and species. These have since been used in a range of published studies. Their value lies in low-cost, fast analytical throughput and small required sample size. In particular, they allow precise values of KPY to be obtained from small wood samples obtained non-destructively. This study reports the ongoing performance of these calibrations with respect to laboratory chemical values, as well as describing a wood density calibration. Cellulose content was measured in an additional 550 samples taken from over 14 different data sets. The standard error of prediction (SEP) of CC was 0.35% with an average bias of only 0.28%. Adding these samples to the existing calibration gave a sample set of 1 800 samples, producing a calibration with a crossvalidation r2 of 0.87 and standard error of 0.78%. The existing KPY calibration was assessed against an additional 527 samples generating a standard error of prediction of 1.9% and a bias of -1.8%. Adding the samples to the calibration gave a sample set of 1 360 samples producing a calibration with a crossvalidation r2 of 0.87 and standard error of 1.29%. The correlation between NIR-predicted CC and NIR predicted KPY was 85%. A wood density calibration was developed using data generated from wood chips. A total of 1 138 samples were separated randomly into calibration and test sets, representing between 19% and 38% of the sample set. The SEP values were between 34 and 40 kg m-3 and explained between 72% and 84% of the variance. The final calibration including all samples had a crossvalidation r2 of 0.82 and standard error of 29.5 kg m-3.