Modelo geoestatístico espaço-temporal com funções de covariância estacionárias não-separáveis aplicado ao albedo de superfície

• Ano de defesa: 2016
• Autor(a) principal: Alves, Henrique José de Paula
• 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: Funções de covariância; Modelos espaço-temporais; Covariograma; Covariance functions; Spatial and temporal models; Covariogram; Estatística
• Link de acesso: http://repositorio.ufla.br/jspui/handle/1/10966

Several areas of science, such as environmental areas, biological, epidemiological, agrigultura, etc., have data from characterized by variations in space and time. In most of the cases, these variations are measure using statistical methods that take or not take into account the interactions between the dimensions of space and time. Geostatistics is one of those procedures. The goal is to predict observations locations and / or unsampled time. Directed studies for this purpose stand out mainly due to the wide applicability of spatio-temporal models. For many authors in the literature, there is a lack of targeted software for this type of analysis. Gneiting (2002) proposes a model that is based on the construction of valid covariance functions, given the condition of being positive definite and defining separable and inseparable random fields. In this thesis, the objectives are: Present a conceptual and methodological review of the proposed modeling Huang and Gneiting; Analyze using the R language actual data of the surface albedo in the southern region of Minas Gerais, using the methodology proposed by Gneiting. Importantly, the goal here is not to set the best model for the data, but rather investigate the modeling structure. Data for the analysis deal with the daily average incidence of albedo, in southern Minas Gerais, counted in the 31 days of December 2010, obtained by satellite images through remote sensing. Analyses were performed using the geoR packages, CompRandFld and fields available in the free statistical software R.