Files
Abstract
Wood quality is receiving increasing attention as the world gradually relies on fast-growing trees to supply growing demand for wood and fiber. The Wood Quality Consortium was established to address these concerns. Several basic wood properties were determined for sample trees spanning the natural range of loblolly pine in the U.S. This dissertation centers on developing prediction models for two important wood properties, wood specific gravity and microfibril angle, and quantifying their variation across physiographic regions, stands, trees and within individual trees. Specific gravity and microfibril angle were determined on a ring-by-ring basis for approximately equally spaced disks cut from each sample tree. In addition, cross-sectional specific gravity was measured for these sample disks. These wood properties reveal consistent patterns between trees in both tree longitudinal and radial directions. A mixed effects model approach was employed to model these wood properties, accounting for autocorrelation of multiple measurements within the same tree, unbalance design and heteroskedasticity. The specific gravity sheath model is a non-hierarchical two-level mixed effects model that predicts specific gravity as a function of ring number from pith, relative height and stem taper as its predictor variables. Disk height random effects and growth year random effects nested within a tree were assumed to be a log linear spatial model with a first-order autocorrelation and first-order moving average model. In conjunction with tree level random effects, this specification adequately accounted for spatial autocorrelation of within a tree and between-tree variation of wood specific gravity. The wood sheath specific gravity model will yield relatively accurate, unbiased prediction of wood specific gravity both longitudinally and radially at any position within a tree. The cross-sectional disk specific gravity provides specific gravity prediction for any heights in a tree. We tried three approaches: a normality-based mixed effects model, a smoothing distributed random effect approach and a Gibbs sampler approach. All three approaches lead to similar specific gravity models and conclusions. The microfibril angle models are mixed effect smooth spline analysis of variance (SS ANOVA) models. We proposed a new algorithm which dramatically reduced computing time and memory requirement to fit a SS ANOVA model and made it feasible to fit to the microfibril angle data. Both earlywood and latewood microfibril angle models were developed in the framework of the mixed effects SS ANOVA model. Microfibril angle spatial autocorrelations in these models were accounted for by assuming correlated disk random effects and growth year random effects within a tree. Earlywood and latewood microfibril angle models were fit simultaneously to account for autocorrelation of earlywood and latewood measurements from the same growth ring. This methodology represents the first model to account for individual model autocorrelation, between model autocorrelation in a system of equations for longitudinal data. Our wood property models will improve wood quality prediction and utilization.