Mapeamento de culturas anuais no estado de Goiás com séries temporais de imagens Modis

• Ano de defesa: 2014
• Autor(a) principal: Oliveira, Bernard Silva de
• Orientador(a): Ferreira, Manuel Eduardo
• Banca de defesa: Ferreira, Manuel Eduardo, Clementino, Nilson Ferreira, Coutinho, Alexandre Camargo
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Goiás
• Programa de Pós-Graduação: Programa de Pós-graduação em Geografia (IESA)
• Departamento: Instituto de Estudos Socioambientais - IESA (RG)
• País: Brasil
• Palavras-chave em Português: MODIS; Séries temporais; Agricultura; Processamento digital de imagens; Métricas fenológicas
• Palavras-chave em Inglês: Time-series; Agriculture; Digital image processing; Phenological metrics
• Área do conhecimento CNPq: CIENCIAS HUMANAS::GEOGRAFIA::GEOGRAFIA HUMANA::GEOGRAFIA AGRARIA
• Link de acesso: http://repositorio.bc.ufg.br/tede/handle/tede/12376

Agricultural expansion in Brazil is still quite intense, especially in the Midwest region, especially in the states of Mato Grosso and Goiás, both with a large representation of the Cerrado biome. The main agricultural crops, with emphasis on the international market, are soybeans, corn and sugarcane. Thus, it becomes absolutely necessary the development and application of new techniques based on remote sensing to map areas of crops at a regional scale, as quickly and accurately as possible. Using data from MODIS (Moderate Resolution Imaging Spectroradiometer) aboard the Terra platform, such land use in the country can be systematically monitored, especially in grazing areas, forests, savanna and agriculture. In this context, the main objective of this research was to improve techniques for mapping the soybean and corn in the middle region of Goiás (Centro Goiano macro-region), with the use of Vegetation Index images (EVI) obtained from MODIS time-series between 2002 and 2010. The EVI images, despite the recognized high quality, contain some atmospheric interference, inherent to the process, as the presence of clouds; in this sense, a set of methods to minimize such noise were applied to datasets. Overall, among the methodological procedures of this research, were adopted (1) the application of pixel reliability band, in order to remove pixels contaminated by clouds, and (2) the use of estimates of contaminated pixels (excluded from each image), and (3) the application of interpolation filters to each scene, to obtain continuous temporal-spectral profiles for the land use class analyzed over the time. Due to digital processing, it was possible to characterize the phenological response for agriculture, followed by its classification through a decision-tree method in IDL language, with the aid of phenological metrics and statistical analysis of the pixel response. The results demonstrate the efficiency of the method for the temporal monitoring of agricultural areas in the Cerrado (Goiás), although an omission error for regions with small areas of planting occurred due to the pixel size of MODIS (6.25 hectares, favoring a spectral mixture). In areas with large plantings of soybeans, was achieved an accuracy of 78%, while corn remained below 48%, due mainly to the few areas intended for this crop in Goiás. As part of this research, an image processing tool for MODIS/EVI dataset (developed for ENVI/IDL) was created and put available for agriculture mapping in the Cerrado biome.