Study of stem form using artificial neural networks and taper functions

Abstract

Artificial neural networks (ANN) have great potential as an alternative to conventional regression analysis because of its learning capacity of data set information and the generalization of learning to unknown data. So, the aim of this study was to apply RNAs to estimate relative diameter, total and commercial volume, as well as to compare their performance in relation to conventional taper functions. Data from 47 trees of Eucalyptus sp. were used in the training and validation of ANNs and in adjusting Hradetzky and Garay taper functions. The performance of ANNs were similar to the taper functions for diameter estimative, furthermore the estimative of commercial and total volume applying ANNs were more accurate and presented less residues scattering than Garay and Hradetzky function. ANNs were accurate and appropriate for the estimation of volume and relative diameter.