Global near infrared models to predict lignin and cellulose content of pine wood

Global near infrared models to predict lignin and cellulose content of pine wood were developed using 517 samples for lignin and 457 samples for cellulose. Samples came from seven different pine species, including tropical species (Pinus caribaea, P. oocarpa, P. maximinoi, P. patula and P. tecunumanii) and temperate species (P. radiata and P. taeda) from five different countries (Brazil, Colombia, Chile, South Africa and the USA). The global models were tested on an independent validation data set and had excellent fits for lignin [correlation coefficient (r(2))=0.97 and standard error of prediciton (SEP)=0.44] and good fits for cellulose (r(2)=0.82 and SEP=1.08). Subsets of the data were used to develop smaller multi-species, multi-site calibrations that could be tested on independent datasets containing different species not included in the calibration model. For calibrations based on four or more species, predictions from those models on independent datasets were generally good, with only slight degradation in r(2) and SEP relative to the calibration R-2 and SECV. The results suggest that global calibrations could be valuable in tree breeding programmes to rank species and genotypes for lignin and cellulose content. Species-specific models were developed for two species (P. tecunumanii and P. taeda) which had sufficient numbers of observations; the global calibrations gave predictions as good as the species-specific calibrations.