Avaliação da estrutura de florestas tropicais a partir da simulação e inferência bayesiana

Abstract

The main dynamics study tool of trees is the diameter distribution. Different methods are used for the adjustment functions, but only the Bayesian method considers the experience of the researcher in the setting through the use of information a priori. Thus, the aim of this study was to assess the importance of different prioris in the setting of Weibull distribution and its influence on the estimated volume in simulated scenarios of a rainforest. The dataset was obtained from an area located predominantly in Linhares (ES). The sample was stratified, each diameter class regarded as a layer 2 tree, yielding a set comprised 200 individuals. They were resampled three scenarios with 20 plots of 30 trees each, in which the resampling is reversed, equiprobable and proportional to dap. The resulting set of each resampling was divided into 7 classes of diameters, each 5 cm intervals. For each scenario were adjusted likelihood 4 hypsometric models (polynomial of the 1st, Stoffels, Richards and Weibull) and three volumetric models (Spurr, Schumacher-Halland Ogaya) for further estimate of the volume by diameter class, whose the best fit selection criteria were the AIC and BIC. In the setting of Weibull distribution by Bayesian method was used four levels of information for the model parameters and only likelihood. For adjustment we used the MWG algorithm, 3 Markov chains and 100,000 early iterations. The criteria for Geweke, BMK, ESS and MCSE to analyze the convergence of Markov chains and used the DIC and the Bayes factor for the selection of best fit the Weibull function were used. Estimates of the number of trees per hectare was made through the accumulated density of the Weibull distribution function. The volume by diameter class was estimated using the parameters of the best adjusted biometric templates. For scenario 1, the top set biometric templates were to Richards and Schumacher-Hall and the Weibull distribution to best fit the criterion DIC with the use of priori 3 (a, b e c ∼ Gama(5; 5)), and the difference between the total volume sampled and the volume estimated total of 20.18%. For scenario 2, the top set biometric templates were to Richards and Schumacher-Hall, and the best fit of the Weibull distribution, according DIC was obtained with the use of priori 3, and the difference between the total volume sampled and the estimated total volume of 25.93%. For scenario 3, the best set biometric templates were to Richards and Ogaya and the best fit of the Weibull distribution, the discretion DIC was obtained using the likelihood, and the difference between the total volume sampled and the estimated total volume 7%. Using the Bayesian method was not effective for adjusting the Weibull distribution because the sample size is very large, favoring the likelihood and also because the distribution of the data.