The probability of mortality for an individual tree with certain characteristics growing under certain conditions was modeled. A particular algorithm SCREEN was used to fmd the best set of predictors variables. This algorithm was specially created to be used when the dependent variable can take only two values like in this binary case (dead or alive tree). The logistic model with different independent variables, which were found to be significant through the SCREEN algorithm, was fitted to the data. For the unthinned plots the logistic model with the following variables, CR (crown ratio), HH (total height/height of dominant and co-dominant trees) and CI (competition index) was compared with the survival model applied in a published distance-dependent model PTAEDA. The logistic model with CR, HH and DD (quadratic mean diameter/dbh) was compared with the survival model already used in a distance-independent model TRULOB. In both cases the behavior of the logistic model was quite similar to the published models. For the thinned plots the predictor variables DDt HH, CI and CR raised to 1.5 were used in the logistic model to predict mortality for individual trees. Mortality is difficult to predict. In this particular study the logistic model was used. The final distance-dependent model for unthinned plots includes as predictor variables CR, HH and CI. For thinned plots the final logistic model employs HH, CI and CR raised to 1.5 as independent variables. The final distance-independent model for unthinned plots includes as predictor variables HH, DD and CR. For thinned plots the final logistic model uses HH, DD and CR raised to 1.5 as independent variables. Differences between deterministic and the stochastic treatments of mortality were also studied. No practical differences in several stand characteristics such as average height, total volume, basal area were found when using these two approaches. Further, no significant differences were found in the diameter distribution for dead or alive trees. / Master of Science

Several nonlinear, individual tree crown height increment equations were tested for their ability to predict annual crown height increment in loblolly pine plantations. The selected model contained tree height (HT), tree crown ratio (CR) raised to the one-half power, age (A), and a competition index (CI) for the distance-dependent model and the ratio of quadratic mean diameter to tree dbh (DR) for the distance-independent model. The distance-dependent and the distance-independent models were the same form, except for the expression for competition. Hypothesis tests revealed that thinning, both its intensity and the elapsed time since its occurrence, had a significant effect on crown height increment. A thinning variable, THIN1, which accounted for thinning intensity and the interval since thinning, was developed and incorporated into the final individual tree increment models. Predictions of crown height increment were improved using models with the THIN1 variable as compared to those with no thinning allowance. In another approach, existing crown height equation was modified to account for the effect of thinning on crown recession. Another thinning variable, THIN2, similar to THIN1, was added to the crown height model. This model yielded better results than its counterpart with no thinning variable; however, the improvement was not as great as for the increment models. The individual tree increment models were also used to form a stand level crown height increment model. The independent variables were collapsed to stand-level statistics; the final model contained average height of dominants and codominants (HD), average crown ratio (R), age (A), and the THIN1 variable. Unlike the individual tree models, raising the average crown ratio to .5 did not improve the fit; however, including THIN1 did improve the results. From this study it was concluded that better data, a standard definition of height to the live crown, and other crown variables, such as crown diameter, will be required to produce more refined individual tree crown height increment models. / Master of Science

The minimum number of tree heights that are necessary, with a probability of 0.95, to obtain a merchantable plot volume estimate of loblolly pine within ± 3, 5, and 10% of the volume observed if all plot trees had been measured for height were determined for all combinations of volume estimation techniques and sample designs examined in this study. The volume estimation techniques examined in this study were: 1) a volume equation using measured tree diameters and either measured heights or height estimates obtained from a plot height-diameter relationship, 2) a volume equation using strata average diameter and average height, and\ 3) a strata volume/basal area ratio estimator. The examined sampling designs were: 1) a simple random sample, 2) a stratified random sample, 3) a stratified systematic sample, and 4) a purposive sample. Both combined and separate stratified estimators were used for volume estimation techniques 2 and 3 when a stratified sample design was used. Of all the possible combinations of volume estimation techniques and sample designs, two combinations, volume estimation technique 1 and a stratified random sample, and volume estimation technique 1 and a purposive sample, are the only combinations that have sample sizes of no more than 30 trees for all three accuracy levels and require the smallest or nearly the smallest number of sample tree heights at these accuracy levels. / Master of Science