Volumetric estimates in eucalyptus stands using support vector machines and artificial neural networks

This study aimed to evaluate the performance of artificial neural networks (ANN) and support vector machines (SVM) in volumetric modeling in eucalyptus stands. Data from commercial plantations, located in four municipalities in the southern mesoregion of the state of Amapa, were used and were provided by a private company. Volumetric models established in the literature were adjusted and compared with the SVM and ANN techniques. Data were divided into 80% for training and 20% for model validation. The same dendometric variables used by the regression models (DBH and height) were used by the SVM and ANN. For training and generalization of the SVM, four configurations were used, formed from two error functions and two Kernel functions. For configuration, training, and generalization of the ANN, the NeuroForest-Volumetric software was used, in which network configurations such as Adaline (Adaptive Linear Element) were used; Multilayer Perceptron (MLP) and Radial Base Functions (RBF). The quality of the adjustments of the regression models, and of the methodologies using ANN and SVM, were evaluated using the correlation coefficient between the observed and estimated individual volumes (ryy), the root mean square error, expressed as a percentage of the mean (RMSE%), and graphical analysis of residues (Res%). Considering the results, SVM and ANN performed slightly better, compared to the traditional methodology, in individual volume estimates, demonstrating that they are techniques that are well suited for applications in the area of measurement and forest management.