Modelagem da distribuição diamétrica de florestas tropicais

Abstract

This study aimed to apply the artificial neural network technique to model the diametric structure of tropical forests. We used data of 27 areas of native forests situated in the Rio Grande watershed in Minas Gerais state, totaling 979 plots. Three Artificial Neural Networks were tested for diametric modeling, networks 1 and 2 for prediction of probabilities by diameter class and the network 3 for prediction of parameters b and c of the Weibull function. The diameter structure estimated by the networks was compared with the probability density function Weibull with three parameters fitted by the method of moments and with the actual distribution obtained from information collected in the field. The evaluation and comparison of the methods are given from residual analysis calculated by diameter class and the KolmogorovSmirnov adherence test. Neural Networks 1 and 2 had greater accuracy in estimates of probability in the first three diameter classes, compared to estimates produced by Weibull, obtaining smaller total error and over 95% adherence to the actual distribution. This technique can also be applied to the prediction of Weibull function parameters with adhesion higher than 90%. According to the results, the Artificial Neural Networks can be successfully employed in shaping the diameter distribution of uneven-aged forests.