Volume de madeira: determinação, amostragem e métodos de estimativa

Abstract

Finding strategies to establish the best practice of volume determination, the best modeling method and the minimum number of samples to compose a robust volumetric equation is important for the manager since it generates reliable Information about the volumetric forest inventory. In this sense, the objective of this thesis was to search for strategies that can improve the quality of the adjustment of statistical models to estimate the volume of standing trees in the southeast of Pará. The volume estimation was studied from three analyzes: the first analysis was raised three objectives: the first was to compare the three practices of measuring tree volumes (real volume, usable volume and geometric volume); the second one was to verify the influence of the species and the diameter classes in the composition of the volume of wood obtained by the three practices of volume measurement; and finally it was quantified the volumetric losses after the cutting of the trees in a forest management unit in Anapu-PA. Fifty-three sample-trees belonging to five species were analyzed in three diametric classes. A factorial (5x3x3) ANOVA was used to verify the effect of the species, diameter classes and measurement practices, on the different volumes obtained. The geometric volume presented, on average, less volume than the actual volume for ali species in the different classes, with no difference between the averages of the commercial volume and the actual volume. The second objective aimed to determine the minimum number of trees required to estimate the volume for six species in Novo Repartimento-PA. A total of 318 sample-trees belonging to six species were measured. To determine the minimum number of sample-trees, 1000 random simulations were performed for different sample sizes (4, 5, ..., n- 1). In each simulation, the Husch and Schumacher-Hall volumetric models were adjusted and the Reliability Index (IC%) and the Permissible Error Limit (LEa%) were obtained for each sample size. For the LEa%, we considered six limits varying between 10 and 35%. The t test was used to compare the mean LEa% with the mean IC%, for each sample size, when the IC% was lower than the Lea%, this was considered the minimum sample-tree point. For Schumacher-Hall, at a 20% error limit, 29 sample-trees per species and 81 trees for ali species are required. For Husch, at 35% limit, 39 sample-trees per species and 158 trees for ali species are required. The third objective aimed to compare the volumetric precision of the different forms of tree volume modeling (form factor 0.7, estimated form factor, linear regression and artificial neural networks), for different sizes of sample-trees in Novo Repartimento-PA. A total of 681 sample-trees belonging to 52 species were considered. Trees were randomly selected to compose different sample sizes (25, 50, 75, 100, 150, 200, 250, 300) for each modeling method. The precision of the models was analyzed by the Sy.x%, the BIC and the Graybill F test. The models with the lowest values of the precision parameters were selected. The factorial ANOVA (7x8) was used to compare the estimated volume by the methods in the different sample-trees sizes. The best models in terms of accuracy and precision were Schumacher-Hall, adjusted from 75 sample-trees, Husch, from 150 sample-trees and the double entry RNAs, from 100 sample-trees. Therefore, in order to generate a robust volumetric equation, 150 sample-trees must be rigorously measured for the fitting of the Husch model using linear regression.