Mayra Luiza Marques da Silva Binoti

Aplicação de redes neurais artificiais para estimação da altura de povoamentos equiâneos de eucalipto

RESUMO – Este trabalho teve como objetivos aumentar a precisao das estimativas de altura de arvores e diminuir a necessidade de afericao da altura em campo, levando a reducao dos custos no inventario florestal atraves da construcao e validacao de um modelo de estimacao da altura de arvores em povoamentos de eucalipto com a utilizacao de redes neurais artificiais. Os dados utilizados consistiram em tres clones, compreendendo cerca de 3.000 arvores em 145 parcelas permanentes com area media de 215 m², mensuradas em seis ocasioes (idades). As variaveis utilizadas para estimar a altura total das arvores dividiram-se em quantitativas: idade (meses), diâmetro com casca a 1,30 m de altura a partir da superficie do solo (dap) e altura dominante media da parcela; e qualitativa: tipo de solo com as respectivas classes. Para validacao e aplicacao da metodologia proposta, foram consideradas duas situacoes: (a) quando ha a introducao de um novo material genetico e nao existem informacoes sobre a relacao hipsometrica deste; (b) quando ja se conhece a tendencia de crescimento em altura dos povoamentos implantados, obtida pela existencia de medicoes em parcelas de IFC. Com as metodologias testadas, obtiveram-se valores de coeficiente de correlacao superiores a 0,99. As metodologias mostraram-se eficientes para alcancar os objetivos propostos, garantindo alta precisao das estimativas obtidas atraves das redes neurais artificiais.

REDES NEURAIS ARTIFICIAIS PARA ESTIMAÇÃO DO VOLUME DE ÁRVORES

The objective of this study was to propose a methodology by using Artificial Neural Networks (ANN) to reduce the number of trees to be scaled during the process of generating volumetric equations. The data used in this study were originated from measurements of 2,700 trees of clonal eucalyptus plantations located in southern Bahia State, Brazil. The training of ANN was performed in order to obtain networks for estimating the volume with and without bark. As input variables, we used the diameter at breast height (dbh) - 1.30 m, height, and diameter at 0.0, 0.5, 1.0, 1.5, 2.0 and 4.0 m above the ground and the volumes obtained until 2 and 4 m. The accuracy of the method was carried out using the test L & O. We also evaluated the dispersion of percentage errors, frequency histogram of the percentage error and the root mean square error (RMSE). The methodology proposed in this study proved to be efficient for estimating the volume of trees, and is indicated to obtain the total volume with and without bark of eucalyptus, enabling the reduction of costs for the construction of volumetric equations.

PROGNOSE EM NÍVEL DE POVOAMENTO DE CLONES DE EUCALIPTO EMPREGANDO REDES NEURAIS ARTIFICIAIS

Objetivou-se, neste estudo, treinar, aplicar e avaliar a eficiencia de redes neurais artificiais (RNA) para realizar a prognose da producao de povoamentos equiâneos de clones de eucalipto. Os dados utilizados foram provenientes de povoamentos localizados no sul da Bahia, totalizando cerca de 2.000 hectares de floresta. Foram utilizadas variaveis numericas, como: idade, area basal, volume e variaveis categoricas, como classe de solo, textura, tipos de espacamento, relevo, projeto e clone. Os dados foram divididos aleatoriamente em dois grupos: treinamento (80%) e generalizacao (20%). Foram treinadas redes de tres tipos: perceptron, perceptron de multiplas camadas e redes de funcao de base radial. As RNA que apresentaram os melhores desempenhos no treinamento e generalizacao foram selecionadas para realizar a prognose com dados, a partir do primeiro inventario florestal. Conclui-se que as RNA apresentaram resultados satisfatorios, comprovando o potencial e aplicabilidade da tecnica na solucao dos problemas de mensuracao e manejo florestal.

Modelagem da distribuição diamétrica de povoamentos de eucalipto utilizando a função Gama

Neste estudo, avaliou-se o comportamento da funcao densidade de probabilidade Gama com 2 parâmetros para a descricao da distribuicao diametrica de um povoamento de eucalipto em diferentes idades. Um modelo de projecao da distribuicao de diâmetros foi construido e ajustado aos dados das parcelas permanentes. Os ajustes da funcao Gama foram avaliados pelo teste de Kolmogorov-Smirnov (KS) e o modelo de distribuicao diametrica foi avaliado por meio dos coeficientes de determinacao e graficos de residuos das estimativas geradas pelas equacoes que compuseram o modelo. As distribuicoes diametricas estimadas pelo sistema de equacoes foram comparadas com as distribuicoes observadas pelo teste KS. A maioria dos ajustes apresentou aderencia pelo teste KS. O modelo foi capaz de projetar as distribuicoes diametricas de maneira satisfatoria, acompanhando a tendencia de achatamento da curva da distribuicao em uma sequencia de idades. Conclui-se que a funcao Gama pode ser utilizada em um modelo de projecao da distribuicao dos diâmetros de povoamentos de eucalipto

Modelos de distribuição diamétrica para povoamentos desbastados de eucalipto utilizando a função hiperbólica

The objectives of this study were to adjust, to employ and to propose diameter distribution models with different hyperbolic function shapes for the projection of the diameter distribution of thinned stands. It was used data from an experiment on thinning set up in three stands of an Eucalyptus grandis × Eucalyptus urophylla hybrid, located in the northeastern Bahia state. The permanent plots were measured at 27, 40, 50, 58, 61, 76, 87, 101 and 112, 125, 135, 147 and 160 months, with a thinning carried out at 58 months of age and a second thinning conducted at 130 months of age. The two-parameter function showed the best results of root mean square error (RMSE), bias and bias%.

Aplicação da função hiperbólica na construção de curvas de índice de local

This study evaluated the efficiency of the hyperbolic function for construction of anamorphic and polymorphic site index curves. The site index curves were constructed using the methods: guide-curve, difference equation, parameter prediction, Hammer and site index preliminarily defined. To apply these methods data from permanent plots in stands of eucalyptus were used. The efficiency was confirmed, being the function indicated to describe anamorphic and polymorphic trends that can be used in the classification of productivity capacity of eucalypt stands.

Descrição da distribuição diamétrica de sistemas agrossilvipastoris utilizando as funções Weibull e hiperbólica

The objectives of this study were to adjust and to evaluate different forms of the Weibull and hyperbolic function to describe the diameter structure of agroforestry system. The forestry component was established with clones of eucalyptus hybrids in the 10 x 4 m spacing, aimed at producing wood for electric power and sawmill. Annual crops as rice and soybeans were planted on the interows in the first year and soybean in the second year. Pastures with Brachiaria were formed in the following year and with the purpose of fattening beef cattle.It was tested the two functions with three parameters, two parameters, and parameter of location as the minimum diameter, and right truncated. Adhesion of function of the data was assessed by Kolmogorov-Smirnorv test. The functions were also evaluated by the Sum of Deviations Squares (SDS) and by graph analysis among the observed and predicted values. The results indicate the possibility of using those functions to describe the diameter distribution of thinned populations.

Utilização de redes neurais artificiais para a projeção da distribuição diamétrica de povoamento equiâneos

The aim of this study is to develop and evaluate the application of artificial neural networks to forecast parameters of the Weibull distribution. We used data from permanent plots of eucalyptus, measured on eight occasions. We adjusted the Weibull function with two parameters for all plots and occasions, by the method of maximum likelihood. The projection of the diameter distribution was made through artificial neural networks. We compared the proposed method to the method traditionally used in modeling the diameter distribution. The models using ANN presented improvements in the graphic dispersion of the residues, as well as the evaluated statistics. The proposed method was superior to the commonly used method.

Modelagem da distribuição de diâmetros utilizando autômatos celulares e redes neurais artificiais

This study presents a diametric distribution model based on a one-dimensional cellular automata model (CA) and artificial neural network (ANN). Each cell of CA represents a dbh class, with the future state predicted in function of the present state of this cell, of the four neighboring cells and of its present and future age. An ANN was used as rule of evolution. Accuracy was evaluated by applying: statistical procedure proposed by Leite and Oliveira (2002); relation between observed and estimated frequency; and biological realism of the built model. Of the trained networks, were selected the ten representing the evolution of the diameter distribution with greater accuracy. Among these ten ANN, seven had estimated values statistically equal to observed (p>0.01). The proposed modeling approach estimates accurately future diameter distributions.

Modelagem da distribuição diamétrica em povoamentos de eucalipto submetidos a desbaste utilizando autômatos celulares

This study presents a diametric distribution model based on a one-dimensional cellular automata (CA) model and artificial neural network (ANN). Each CA cell proposed stands for a dap class, with the future state being predicted in function of the present state of this cell, the state of its surrounding cells and of its present and future ages. An ANN was used as a rule of evolution. Accuracy was evaluated by applying the statistical procedure L, a relationship between observed and estimated frequency and biological realism of the model built. Of the trained networks, 10 representing diametric distribution evolution with more accuracy were selected. Of these 10 ANR, seven represented estimated values statistically equal to those observed (p>0.01). The focus of the modeling proposed allows estimating future diametric distributions accurately.