Inferência bayesiana na parametrização do modelo hipsométrico de Stoffels-Soest para eucalipto

Abstract

Height–diameter modelling is widely used in forest inventories; however, model parameterization still relies predominantly on frequentist methods, with limited exploration of alternative probabilistic approaches. In this context, Bayesian inference was evaluated for parameterizing the Stoffels and Soest model to estimate total height in a clonal Eucalyptus urophylla S. T. Blake stand in central Minas Gerais, Brazil. The inventory was conducted at 60 months of age, with 100 trees sampled across 10 plots. The model was fitted using ordinary least squares (OLS) and a Bayesian approach with different numbers of iterations (500 to 4,000), using diameter at breast height as the predictor variable. The results indicated equivalent predictive performance between the approaches, with no improvement from increasing the number of iterations after convergence. However, Bayesian inference explicitly represented parameter uncertainty through posterior distributions, enhancing the probabilistic interpretation of the estimates. Thus, it proved to be a complementary alternative to OLS for the parameterization of height–diameter models, providing explicit uncertainty quantification without compromising predictive performance.