Perfil longitudinal do fuste de cedro australiano (Toona ciliata M. Roem.) estimado por modelos não lineares e redes neurais artificiais

Abstract

With the increasing demand for the use of Australian cedar wood (Toona ciliata M. Roem), we verify the need to understand the longitudinal profile of the stem as well as its individual volume by means of applying new methods. In recent years, techniques which allow the estimative of height, radius and volume of the trees with ANNs rather than with the traditional regression method have been used. The data to perform this work derived from 72 Australian cedar trees, belonging to the Bela Vista Florestal company, located in the municipality of Campo Belo, Minas Gerais, Brazil. This study aimed at modeling the longitudinal profile of the stem of Australian cedar (Toona ciliata M. Roem.) using non-linear models and Artificial Neural Networks (ANNs). The taper function adjustments were performed and the Garay function best estimated the height, considering the entire data set. Following, the Garay function was adjusted to estimate height, radius and individual volume of the trees. In order to compare the ANNs, the training data were divided into five groups: 90%, 70%, 50%, 30% and 10%, and were evaluated using the test data. According to statistical analyses, the ANNs presented satisfactory results for square root error for the entire set of tests in relation to the regression method. When analyzing the residue, satisfactory results were observed for the ANNs method. Thus, it is considered possible the use of ANNs to represent the longitudinal profile of the stem and the estimate of the individual volume of the Australian cedar trees.