Estimativa de diâmetro a 1,30 m do solo utilizando redes neurais artificiais

Abstract

The aim of this study was to evaluate the performance of Artificial Neural Networks (ANNs) to estimate the diameter variable at 1.30 m above ground (DAP), using as independent variables Age, Area per plant and Total height of trees. Two types of training were carried out; for the first, the data referring to the last year of the forest inventory were not considered, and for the second, the data of all the years were used and these were randomized. NeuroForest software was used to train ANNs. After the processing, the Bias, RQME, correlation and mean percent error statistics were calculated; dispersion charts were generated, considering the estimated and observed values; and histogram of residues. It was observed that training considering the randomized data generated the best statistics, with high correlation values both for the training phase as well as for the validation, due to the greater representativeness of the data during the training. The graphs generated for this training obtained less dispersion of the estimated points, and can infer the greater precision of the estimate when compared to the training in which the data of the last year were not considered.