Use of artificial neural networks for predicting volume of forest species in the Amazon Forest

The objective of this study was to evaluate the efficiency of artificial neural networks to predict wood volume of forest species in the Tapajós National Forest (Amazon) compared to the Schumacher and Hall volumetric model. We used data from two Units of Annual Production (UAP) of a Sustainable Management Plan in an Ombrophilous Dense Forest in the municipality of Belterra, Pará, where the Tapajós National Forest (FLONA Tapajós) is located. The data were obtained from 3607 trees of 31 species, felled and processed to obtain their commercial volume. The logarithmic form of Schumacher and Hall's model was adjusted for each UAP and dbh class (two dbh classes). The input variables for training the artificial neural networks (ANN) were dbh and h as continuous variables and UAP and tree species as categoricals. The output was the commercial wood volume of a tree. We trained Multi-Layer Perceptron (MLP) neural using a resilient propagation algorithm and the sigmoid activation. The efficiency and accuracy obtained using ANN was proved based on some usual validation statistics and residual plots. The statistical analysis of the regression and ANN showed that both methods were satisfactory, but the ANN generated lower RMSE% and higher correlation in the training and generalization. The estimates of commercial tree volume of forest species in the Amazon Forest can be obtained accurately and without bias using artificial neural networks (ANNs). © 2021 University of Sao Paolo. All rights reserved.