Prediction of the bending stiffness of Uruguayan loblolly and slash pine timber applying different statistical techniques

Abstract

Regression trees, random forests, and generalized additive models (GAM) are statistical techniques often used in several disciplines, but rarely in wood technology. This study presented a novel approach to predicting the modulus of elasticity of Uruguayan pine timber by applying three statistical techniques and using visual parameters and non-destructive testing. For this purpose, two sample groups of beams (50 mm × 150 mm × 2800 mm) were selected from two commercial plantations, one comprised of 122 specimens from 14-year-old loblolly pine (Pinus taeda) and the second comprised of 111 specimens from 27-year-old slash pine (P. elliottii). The visual parameters and dynamic modulus of elasticity for each specimen were obtained and associated with their experimental static bending stiffness. The number of annual rings per centimeter, twist, crook, and knot size were the most relevant visual variables for the modulus of elasticity prediction. The inclusion of the dynamic modulus of elasticity in the modeling improved the stiffness prediction by reducing the prediction error by 46% on average. The GAM had the best prediction, with a 10% prediction error, and explained 88% of the variability. These results suggested that GAM is a useful tool for stiffness prediction of Uruguayan pine timber.