PLS prediction as a tool for modeling wood properties

During an earlier study, material responses of shrinkage and deformation during air-drying were measured on small wood specimens cut from the cross section of a Pinus radiata log. In this paper the analysis of the collected data and the modeling of studied responses were done using multivariate methods. A first model with all variables and all 104 observations was created. In this model there were observations that deviated from the rest and some of them were therefore excluded from further modeling. Also, weak variables and undesired variables were excluded from further modeling. After these exclusions, 77 observations of wood responses below fiber saturation point remained. The results showed good modeling of radial, tangential and volumetric shrinkage between 0-22% MC, with explained variance (R-2) and predicted variance (Q(2)) at approximately 0.9, and moderate modeling of longitudinal shrinkage, R-2=0.67 and Q(2)=0.65. It was also shown that longitudinal shrinkage has weak correlation to density-related variables in the studied wood, No model with good predictability of deformation was found. This study showed that PLS prediction modeling of shrinkage and deformation in studied wood samples was found to be an effective and easy-to-use tool for untangling relationships between variables and generating information from data.