Modeling the longitudinal variation of sawn timber grades in Norway spruce (Picea abies (L.) Karst.)

Grades derived from visual assessments of sawn timber are determined by the worst part of each piece. Since grade varies longitudinally in timber, grade yield will decrease if the average length of timber increases. The variation in grade is caused by longitudinal variation in knot properties and other features as they appear on the sawn surface taken into account during grading. The objective of this study is to describe and analyze this variation in Picea abies. The study consisting of 768 boards for which all features that could lead to downgrading were recorded noting position, type and size. Based on this information, all boards were graded according to appearance by Nordic Timber, and strength by INSTA 142. Logistic regression models of grade as a function of position in the stem were developed, and the dependence between responses was taken into consideration by using General Estimating Equations (GEE). The models showed a decreasing trend in grade from the butt end toward the top end of the trees, and the effect was more pronounced in strength grading than in appearance grading. Models with binomial response and different correlation structures were tested, and it was shown that both independent and autoregressive correlation structures could be used. This suggests that a multinomial ordinal logistic regression with a GEE-approach with an independent correlation structure is appropriate for modeling grade in this study.