Modelagem de árvore individual para uma Floresta Estacional Semidecidual utilizando redes neurais

• Autor(a) principal: Castro, Renato Vinícius Oliveira
• Data de Publicação: 2012
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: LOCUS Repositório Institucional da UFV
• Texto Completo: http://locus.ufv.br/handle/123456789/584
• topic: Árvore individual; Florestas naturais; Redes Neurais Artificiais; Floresta estacional semidecidual; Individual tree; Natural forests; Artificial Neural Networks; Forest semideciduous; CNPQ::CIENCIAS AGRARIAS::RECURSOS FLORESTAIS E ENGENHARIA FLORESTAL::MANEJO FLORESTAL

This study aimed to develop a complete model in individual tree basis for a fragment of Semideciduous Montana forest using artificial neural networks (ANN) and validate their application. Data from ten permanent plots in a forest of 17 ha in Viçosa, MG, at the intermediate stage of ecological succession, monitored for 14 years (1994, 1997, 2000, 2004 and 2008) were used for this purpose. The data set was randomly divided into two groups: 1) training networks group, consisting of six parcels totaling 3,556 cases at five measurements, observed 231 cases of death and 238 cases of ingrowth, and 2) the validation model group, consisting of four parcels, totaling 2,062 cases, with 181 cases of death and 146 cases of ingrowth. The work was divided into four chapters where, in Chapter I, we evaluated different competition indexes independent, dependent and semi-independent of distance, and selected those best for studies of growth and mortality in later chapters. In Chapters II and III were evaluated and compared different types and architectures of neural networks to estimate mortality regular trees, as well as projection of diameter, total height and bole height. In Chapter IV, there was the validation of networks employing the set of independent data. Besides the selected networks was considered the average rate at each measurement period, the first data group, the projection of the forest structure (validation). The analysis showed that, overall, the competition indexes were correlated with the probability of mortality and growth of trees. There was also verified that the neural networks efficiency in the estimation of tree mortality and the growth projection, accurate estimates are obtained. In validation, designed to mortality, growth and ingrowth of the first measurement (1994) by the year 2008, which were compared to the observed values. The model for individual tree basis was efficient in simulating forest growth and yield at all levels of detail (stand level, diameter and hight distribution level and individual tree level). The results confirmed the potential use of individual tree basis models through artificial neural networks in natural forests structurally complex, as is the case of the Semideciduous Montana Forest.