Estimation of the height of thinned eucalyptus trees in agrosilvopastoral system with artificial neural network

Abstract

An alternative approach for modeling hypsometric relationships involves the application of artificial neural networks (ANNs). The general objective of this work was to verify the feasibility of using an artificial neural network (ANN) to estimate the total height of Eucalyptus urophylla x Eucalyptus grandis trees in an agrosilvopastoral system and compare its performance in relation to the generic hypsometric regression model. For the estimation of the total height of the trees, by the ANN and by the regression model, the variable diameter at breast height (DBH) and age of the trees were used and, for its evaluation, the correlation coefficient, graph of dispersion of percentage errors, histogram of the frequency of percentage errors, root mean square error percentage, root mean square error, bias and normality of the percentage errors. The use of the Levenberg-Marquardt algorithm and the architecture of four intermediate layers of 10 neurons in each one of the layers, in the ANN, provided an estimate of good accuracy and was superior to the hypsometric regression model.