Modeling thinning effects on ring width distribution and wood specific gravity of loblolly pine (Pinus taeda L.)

An appropriate accounting for thinning effects on growth rate and wood quality requires a clear understanding and quantification of these effects. In this regard, four basic interrelated issues were addressed in this study: 1) thinning effects on ring specific gravity 2) thinning effects on ring width distribution 3) thinning effects on stem form, and 4) prediction models for these quantities. The study showed that thinning does not significantly affect ring specific gravity, whereas its effects on ring width distribution and stem form were significant. Thinning increases ring width significantly over most of the tree bole and increases the earlywood and latewood components proportionally maintaining an approximately constant latewood proportion. As a consequence, thinning effects on latewood proportion is not significant; confirming the results obtained in the specific gravity study and further dispelling the concern that thinning may substantially reduce wood specific gravity. Thinning affects stem form by increasing the form exponent especially near the tree base accentuating the neiloid form expected in that area. High up in the stem, the form exponent changes little within a tree and among thinning treatments, with a general tendency towards a paraboloid shape. Differences due to thinning intensities, in general, were not significant indicating the applicability of results within a wide range of densities. Prediction models for ring specific gravity, ring width, latewood proportion and stem profile based on ring, tree, stand and site factors were developed Influences of stand level factors, density measures in particular, in prediction models are minor probably because tree level factors such as, stem diameter at breast height, crown ratio, etc. themselves manifest stand conditions. The mixed-effects analysis technique was used in data analysis to account for correlation among observations from the same subject. Direct covariance modeling yielded better fits than accounting for correlation indirectly using random effects covariates in many cases; however, both could not be accommodated simultaneously. Structures which assume decreasing correlation with increasing distance between observations, such as the first-order autoregressive structure, performed better than alternative specifications. Results consistently showed that accounting for correlation among observations substantially improves the fits over ignoring correlation; effectively addressing the issue of bias in the standard errors of estimates.