Uso de redes neurais artificiais e métodos tradicionais na estimativa do volume do fuste de Eucalyptus spp., na região do Polo Gesseiro do Araripe

Abstract

One of the ways to estimate volumes of the Eucalyptus trees with efficiency is to apply formulae that can make the estimates with precision. The objective of this work was to compare two ways of volume estimations: mathematical models through regression methods and the artificial neural networks. The data set used came from the second rotation of the experiment containing 15 clones, of Eucalyptus spp., implanted at the Experimental Station of the Agronomic Institute of Pernambuco (IPA), located in the Araripe region, Pernambuco, Brazil. The data set was composed of 2199 trees, cubed by the Smalian method. It was measured the total height, stem height and diameters at several position in the bole. For modeling, the trees were grouped using the Skott-Knott test, per clone and for all the clones together. The evaluation of Artificial Neural Network (ANN) adjustments considered two forms of input of the independent variables: a) volume sections and b) diameter at breast height (DBH) and total height (Ht). The performance of the mathematical volumetric models and ANN was based on Schlaegel fitting index (IAaj), root mean squared error in percentage (RMSE%) and graphic residuals analysis. According to the results, the Schumacher-Hall model presented better statistical performance for estimating the individual volume of Eucalyptus spp. The Artificial Neural Network considering the volume of the sections as input variables was superior when comparing with DBH and Ht as input variables. The Silva-Borders model and Artificial Neural Networks considering the volume of the sections as input variables are more practical to measure volumes because they do not consider the variable Ht, which is often difficult to measure in the field and costly.