Different approaches for modeling Swietenia macrophylla commercial volume in an Amazon agroforestry system

Abstract

The agroforestry systems (AFS) in the Amazon stand out in the national and international scenario due to the possibility of cultivating tropical forest species of high commercial value as the Brazilian mahogany, concomitantly, has sought to use more robust approaches to quantifying timber volume that does not require the same assumptions as traditional approaches. In this context, this study aimed to develop volumetric equations, by traditional approaches, by mixed modeling, and by machine learning techniques, that accurately estimate the commercial volume of Brazilian mahogany trees in an AFS in Amazon as well as verify whether there are differences in the estimates of these approaches by univariate analysis. For that, 108 trees were sampled in 36 circular plots of 500 m2 to estimate the commercial volume. Volumetric equations were developed from the fit of the Schumacher and Hall volumetric model and Kozak taper, by nonlinear regression, from the application of nonlinear mixed modeling in the most precise traditional model and training of artificial neural network (ANN) and supporting vector machines. The analysis of variance indicated that there was no significant difference between the mean values estimated by the equations developed by the different approaches tested in the inventoried individual data. Nevertheless, it is recommended to use the equation generated by ANN to perform estimations in other populations, because it presented more precise estimates in the test set.