ESTIMATIVA DE ALTURA PARA POVOAMENTOS DE EUCALIPTO NO RIO GRANDE DO SUL POR MEIO DE DIFERENTES CONFIGURAÇÕES DE REDES ARTIFICIAIS

Abstract

The use of Artificial Neural Networks (RNA) in the forest environment has shown to be an efficient alternative to estimate variables more accurately in the forest inventory, besides having the advantage of reducing the high costs involved. The objective of this work was to estimate the height of Eucalyptus grandis trees in two mesoregions of the state of Rio Grande do Sul through training and validation of RNA with different configurations: variation of the number of neurons in the hidden layer (1, 2, 3 and 4), variation of initial weights (0 and 1) and different learning rates (0,01 and 0,3). For this, 714 pairs of data were used, in which the data set was separated in different treatments: T1 100% of the data set, only training; T2 80% of the data set for training and 20% of the data set for cross-validation and; T3 50% of the data set for training and 50% for cross validation. With statistical criteria evaluated, square root of the mean square of error (RQME), correlation ( ) and graphic analysis of residues in percentage (E%), it was verified that the best RNA configuration is with 3 neurons in the layer hidden, with 1 seed and 0,01 learning rate. Regarding data separation, the best result was obtained with the T3 treatment, where an RQME of ± 1,3938 m and a of 0,9606 were obtained with the best RNA configuration.