Abstract: Knots are common defects in wood, the size of which has a major influence on mechanical performance and visual quality. To elucidate the longitudinal growth patterns of knots, we examined 27 individuals of Larix olgensis from the Mengjiagang Forest Farm in Heilongjiang Province. Based on 1137 knot samples, we simulated the vertical growth dynamics of knots along the stem and developed a predictive model for knot width and sampling strategy. The results showed that Hossfeld model was the best baseline among the seven commonly used models of growth. We further constructed a reparameterized model by incorporating tree-level and knot-level variables, as well as a mixed-effects framework improved with random effects. The mixed-effects model had the best performance, with R² increased to 0.6051 and RMSE reduced to 2.3865. We tested four sampling strategies to calibrate the mixed model, and the results showed that sampling design strongly influenced predictive accuracy. Scheme 2, randomly selecting seven knots from the upper stem, achieved the best balance between accuracy and efficiency. Model parameters indicated that knot width increased with branch insertion height and angle but decreased with increasing height diameter ratio of L. olgensis. We recommended to use the mixed-effects model in forest management combined with sampling of seven upper-stem knots for prediction. Moreover, priority should be given to pruning upper-stem branches to effectively reduce knot width and improve timber quality.

Key words: knot width, mixed-effects model, model calibration