Uso de modelos não lineares no estudo do crescimento diamétrico de cedro (Cedrela fissilis)

• Ano de defesa: 2018
• Autor(a) principal: Frühauf, Ariana Campos
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Lavras; Programa de Pós-Graduação em Estatística e Experimentação Agropecuária; UFLA; brasil; Departamento de Ciências Exatas
• Palavras-chave em Português: Curva de crescimento; Erros autocorrelacionados; Regressão não linear; Cedro - Crescimento; Growth curve; Autocorrelated errors; Nonlinear regression; Cedar - Growth; Estatística
• Link de acesso: http://repositorio.ufla.br/jspui/handle/1/29741

Forests play a key role in maintaining life. They provide essential environmental services, conserve biodiversity, alleviate climate change, among others. However, forests are declining and many tree species, mainly native forests such as cedar (Cedrela fissilis), are becoming extinct due to intense exploitation. Cedar has great economic impact in Brazil due to its great use in carpentry, civil engineering, among others. Therefore, it is necessary to study its growth as an aid to obtain better management plans for the forests that contain it. According to Encinas, Silva e Pinto (2005), most tree growth can be characterized by a sigmoid curve, which is well fit by nonlinear models. Mean accumulated DBH (diameter at breast height) collected over time were used in this study. These were collected from nondestructive samples of 85 trees in a part of seasonally dry forests within an area of environmental preservation in the north end of Minas Gerais. Logistic, Gompertz, von Bertalanffy, Brody and Richards non-linear models were fit using the free access R software. However, the iterative method used (GaussNewton) did not converge in the Richards model. The structure of first-order autoregressive errors (AR1) was considered for the adjustment of the other models. The Generalized Least Squares method was used for parameter estimation and the residual standard deviation, the corrected Akaike’s Information Criterion (AIC C ), the Bayesian Information Criterion (BIC) and the Adjusted Coefficient of Determination (R 2 ad j ) were used as criteria for model selection. All models fit well, but the Brody model was the best to describe the cedar diametric growth (Cedrela fissilis) in the study region over time.