Potencial de redes neurais artificiais e regressão na estimativa de volume de espécies nativas

Abstract

Understanding the productive potential of native species contributes to sustainable management. The objective of this work was to evaluate the potential of artificial neural networks and regression in the estimation of wood volume in homogenous stands of Anadantera macrocarpa, Genipa americana and Mimosa caesalpiniflolia. The regression methods and artificial neural networks presented applicability for estimating the individual volume of homogeneous stands at different spacings and at seven years of age. The Spurr regression model obtained better statistical results and dispersion of non-biased errors for the species Anadantera macrocarpa and Genipa americana, whereas the model proposed by Shumacher and Hall obtained greater precision of the volumes of the species Mimosa Caesalpinifolia. The RNAs with two neurons in the middle layer obtained the best adjustments for the species. Artificial neural networks are recommended to estimate the individual volumes of the evaluated species and showed greater precision in relation to the regression in the volume estimation.