Hypsometric relationship of Schizolobium parahyba var. amazonicum in plantations integrated with livestock in eastern Amazonia: applications of different modeling methods

Backgrounds: Research on how obtaining basic variables from the forest inventory supports the accurate estimation of planted forest production. Therefore, this work aimed to select the best modeling method for estimating the heights of trees in a Schizolobium parahyba forest and livestock integration system in the countryside of Pará state, Brazil; hence it was established to compare specific and general regression equations for the different management types, and to analyze whether there is a gain in precision with the increased complexity of the regression models and artificial neural networks (ANNs). Three hypsometric regression models were tested: Curtis, Stoffels & Van Soest, and Petterson, using linear, mixed, nonlinear, and covariate models. The ANNs were of the Multilayer Perceptron type with one and two variables in the input layer. Results: The linear Stoffels & Van Soest hypsometric models showed the best regression adjustment, followed by the Curtis model. The linear and nonlinear regression models performed similarly; hence, the linear ones were more efficient based on their simplicity of adjustment. The specific equations performed better than the general equation except for stratum II. The artificial neural networks with two input variables resulted in better estimates of tree heights. Conclusion: The linear equation models were selected, including the specific strata I and III, and the general equation for stratum II. The increase in the complexity of the regression models did not indicate better estimates, unlike the ANNs.