A Bayesian approach to the modeling of growth and forest production of Eucalyptus
Name: LETÍCIA DA PASCHOA MANHÃES
Publication date: 26/02/2019
Advisor:
Name |
Role |
|---|---|
| GILSON FERNANDES DA SILVA | Advisor * |
Examining board:
| Name |
Role |
|---|---|
| ADRIANO RIBEIRO DE MENDONÇA | Internal Examiner * |
| GILSON FERNANDES DA SILVA | Advisor * |
| WELLINGTON BETENCURTE DA SILVA | External Examiner * |
Summary: xThe use of modeling techniques for the growth and production of forest stands as support for planning and management is of fundamental importance in the forestry sector. One of the ways to perform this type of modeling is by means of models at population level, being the Clutter model the most widespread in Brazil. There are numerous approaches used to quantify the production of forest stands, with emphasis on regression analysis. Thus, this work had the objective of estimating the parameters of the Clutter model by the Bayesian method, to accurately model growth and forest production. The data used come from clonal eucalyptus plantations located in the Midwest region of the state of Minas Gerais. The parameters of the Clutter model were adjusted by the classical method of least squares in two stages and by the Bayesian method using the particle filter of importance sampling and sequential re-sampling (SIR). The data were divided by productivity class and proposed nine treatments, combining number of particles (50, 100 and 500) and model deviation (1%, 5% and 10%), for volume and basal area. The methodologies were evaluated by means of the statistics: root mean square error (RMSE (%)); BIAS (%); and graphical analysis, and for the treatments the analysis of variance was performed together with the Tukey's test. From the evaluative statistics, it was observed the good performance in the classical adjustment of the Clutter model and also using the SIR filter. For all treatments, the filter used was able to estimate the values well, filtering the uncertainties derived from the model deviation, for volume and basal area data. Thus, it is concluded that the SIR filter methodology applied to the Clutter model for volume and basal area estimation was accurate and is a promising tool in the forest area since research using this approach is still scarce.
Keywords: Bayesian Inference; Particles filter; Models at Whole Stand Level; Eucalyptus.
