Análise do padrão sazonal de imagens de índice de vegetação do sensor modis para culturas agrícolas

• Ano de defesa: 2016
• Autor(a) principal: Becker, Willyan Ronaldo
• Orientador(a): Johann, Jerry Adriani
• Banca de defesa: Johann , Jerry Adriani , Correa, Marcus Metri , Mercante, Erivelto
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual do Oeste do Paraná; Cascavel
• Programa de Pós-Graduação: Programa de Pós-Graduação em Engenharia Agrícola
• Departamento: Centro de Ciências Exatas e Tecnológicas
• País: Brasil
• Palavras-chave em Português: Ciclo fenológico; EVI; Timesat; Mineração de Dados
• Palavras-chave em Inglês: Phenological cycle; EVI; Timesat; Data Mining
• Área do conhecimento CNPq: CIENCIAS AGRARIAS::ENGENHARIA AGRICOLA
• Link de acesso: http://tede.unioeste.br/handle/tede/2975

Orbital remote sensing techniques have proved to be a valuable tool, since they enable the agricultural monitoring of the vigor and the type of vegetation coverage in a regional scale, bringing results with greater anticipation and precision, and lower operational cost when compared to traditional techniques. Automatic identification of cultivated areas is one of the most important steps in the crop forecasting process. The improvement in the estimate of area cultivated with each crop directly influences the result of the forecast of each crop year, since the agricultural production is a function of the cultivated area. The general objective of this research was to create an automatic methodology for the separation of agricultural crops from soybean and maize by means of data mining (Article 1) and a methodology for forecasting the harvest date from the date of maximum vegetative development (Article 2). The methods used corresponded to the application of the seasonal trends analysis and data mining for soybean and corn agricultural areas in the state of Paraná, with images of the EVI vegetation index of MODIS sensors, TERRA and AQUA satellites. The results obtained in Article 1 show that, through the decision tree, one of the techniques of data mining, it was verified that, among eleven variables that characterize the spectral-temporal pattern of the EVI of each culture, five were enough to perform the separation of soybean and maize crops, in the year 2014/2015, with an accuracy of 96.3% and a kappa index of 0.92, being the maximum value of EVI, the date of sowing (DS), the Date of maximum vegetative development (DMDV), Cycle, and Major Integral. In Article 2 the DS, DMDV and Harvest Date (DC) of the EVI temporal profile were estimated for each mapped soybean and maize pixel in the crop years 2011/2012 to 2013/2014. Then, for each crop and crop year, the variables Delta1 (DMDV minus DS) and Delta2 (DC minus DMDV) were created. The results of the differences (DCDifference) between DC estimated by EVI (DCEVI) and predicted by mean time (DCDelta2) show that, for soybeans, it is possible to use only the mean value of the interval between DMDV and DC in the three harvested years studied, with 55 days for soybeans. For corn, the mean interval between DMDV and DC was 60 days, but it is verified that there is a large difference between the results obtained with DCEVI and DCDelta2. For corn DCDelta2 there were large variations among the mesoregions. Differences in DC (DCDifference), when using the means by mesoregions, presented better results than for Paraná as a whole.