Modelagem da altura, volume e afilamento do fuste de Calycophyllum spruceanum Benth. empregando regressão e redes neurais artificiais

Abstract

The estuarine wetlands are riparian ecosystems that are associated with white-water rivers, with large input of sediment from the daily cycle of flood and ebb of the tides. In these ecosystem are few approved management plans have been approved, mainly due to the lack of legalization ownership. In addition, there is a need for studies and methods of enabling the proper management of this ecosystem in order to promote their livelihoods and rational use of resources. In general, height, volume and taper estimates are obtained by linear and non-linear regression. Alternatively, the artificial intelligence is a promising tool that has been used successfully in the forestry sector, assisting the decision-making. This study aimed to adjust, through regression analysis and artificial neural networks, hypsometric equations, volumetric and tapering to Calycophyllum spruceanum, an abundant tree species in secondary forests in the Amazon várzea estuary. We randomly selected 695 trees of the species in four stands with 60, 72 and 120 months ages located in the municipalities of Gurupá, Pará state and Mazagão, Amapá state. Hypsometric, volumetric and taper were adjusted considering all data and population stratified by age, and identity tests were then applied to verify the possibility of reducing the number of equations for the volume, height and taper. One-hundred artificial neural networks were trained for height, volume and taper using the Resilient Propagation algorithm with eight neurons for the hidden layer. For the total height, the three models tested by regression (Exponential, Gompertz and Logistic) resulted in high precision estimates. The nonlinear model of Schumacher and Hall generated without bias precise volume estimates. The Garay model showed the best precision to estimate the species taper shaft. Although we did not find any differences in the precision of the models, considering all the data from the models stratified by age, identity testing also has shown that it is more appropriate to take into account the ages. Based on the precision parameters and residual distributions our results demonstrate that the estimates generated by the artificial neural networks are as efficient as the regression models.