Integration of principal component analysis and artificial neural network to modeling productive capacity of eucalypt stands from biophysical attributes

Abstract

Modeling the productive capacity of forest sites from biophysical factors is important when site-dominant height data is not available. For this reason, we aim with this study to model the mean annual volume increment at age 7 (MAI7) of Eucalyptus clonal plantations, to evaluate the accuracy of the modeling, and to fit an empirical equation. For this, we used data from twenty-two variables collected in 51 plots distributed randomly in three classes of MAI7 predicted. Initially, Spearman’s rank correlation was used for primary mining of these variables, then principal components analysis (PCA) was used to create orthogonal latent variables that were used as input in the artificial neural network (ANN) to MAI7 predict. Spearman’s and PCA analysis proved to be excellent for data mining because when used together were enabled to reduce the number of variables and create a variable that represented the maximum variance of the variables which were significantly associated with the MAI7. All ANN trained exhibited high learning capability when using as input to the latent variables created in conjunction with the classes of MAI7, but the ANN trained with architecture 6-9-1 made prediction with greater precision for the test set and too showed high accuracy to the train set. Therefore, combining PCA with ANN was an excellent approach to developing an empirical equation to accurately predict MAI7 of clonal plantings of hybrid of Eucalyptus located in the southeast of Pará State, Brazil from biophysical variables.