Modelagem do volume de espécies florestais da Amazônia: comparação entre equações genéricas e específicas por espécie.

Abstract

The general objective of this research was to quantify the effect in adopting species-specific models for estimating wood volume of trees in Amazonian forests. A dataset of 789 boles of 13 merchantable tree species were utilized. In Chapter 1, the aim was to measure the gain in accuracy and precision in wood volume estimates, when species-specific equations are applied rather than general equations. Husch and Schumacher-Hall volume models were fit to 14 datasets. The models were fit through the Ordinary Least Square method, under an approach based on the bootstrap technique (nrep =5.000), generating 5,000 vectors of coefficients by model and dataset. The bootstrap-based confidence interval for angular coefficients were calculated for the global fitting, that is including all species. The comparison of the specific equations to the generic ones obeyed the following criterion: if 95% of the angular coefficients of a specific equation fitted into the CI of the coefficients of the generic equation, then both equations are statistically equal. The accuracy, precision (Syx%), and deviation of estimates (assessed via residual plots) also were examined. This chapter revealed that the specific equations and are statistically different from the generic ones. In addition, the generic equations were, in general, less accurate and precise than the specific equations. In Chapter 2, the aim was compare the performance of volume equations fitted with and without using the bootstrap technique. As in the previous chapter, the bootstrap technique was applied to the regression analysis, in which 5,000 vectors of coefficients were generated by model and calibration dataset (by species and mixed). Such vectors were used to determine the bootstrap-based CI. To compare equations fitted with and without bootstrap, it was assessed if the coefficients obtained without bootstrap fitted into the CI for those obtained with bootstrap. Precision, accuracy and deviation of estimates were examined as in the previous chapter. The coefficients were statistically equal for both regression approaches. As a conclusion, both regression approaches yielded coefficients statistically equal.