Funções de densidade e probabilidade e métodos de predição de parâmetros para povoamentos de Khaya ivorensis no Brasil

Abstract

Stand horizontal structure in a stand is normally given by a probabilit density function, described by its diameter or seccionalarea distribution. This way, one is able to plan the effect of management practices, silvicultural interventions and wood trade. The fit of a probability density function describes the chance of finding a tree at a determined diameter class, and to relate its parameters with stand attributes, which ease prognosis systems. Diameter class modeling is specially used for those forest with high wood value, as Khayaivorensis for example. It is species with potential for commercial plantations because of its high-priced hardwood. However, its management practices are unknown. This work aims to characterize horizontal structure of Khayaivorensis stands in Brazil, located at Minas Gerais, Goias and Para states by probability density functions and relate the parameters of the best function with stand attributes. On the first paper, Johnson´s SB, Weibull 2 and 3 parameters, Beta and Gamma functions are tested. Johnson´s SB was fitted by 5 methods (maximum likelihood, moments, Knoebel-Burkhart, linear regression and mode). Weibull was fitted by 3 methods (percentis, maximum likelihood and moments). Beta and Gamma functions were fitted by moments method. The adherence was assessed by Kolmogorov-Smirnov test with 5% level of significance. Functions performances were ranked based on the Kolmogorov-Smirnov test. Besides, fittings were compared by the error and mean absolute error. The best fitting method wasJohnson´s SB fitted my maximum likelihood and moments. Weibul fitted by percentis was good as well. The second paper aims to predict Weibull parameters fitted by maximum likelihood and percentile methods using linear models. One of the models tested was proposed by Cao in 2004 in a paper published at Society of American Foresters the others are models constructed by stepwise methods. The best prediction was made by maximum likelihood method, using Cao´s model.