AGRADECIMENTOS Aos meus pais Raimundo e Antônia que contribuíram para a formação do meu caráter e me fizeram chegar até aqui. À minha mulher Liana com quem compartilhei sonhos, entusiasmo, expectativas e frustrações e recebi apoio, incentivo e compreensão. Aos pesquisadores Isabel Belloni Schmidt e Alexandre Bonesso Sampaio que me incentivaram a conhecer um pouco mais acerca da dinâmica das queimadas no Cerrado e me apresentaram um pedacinho do maravilhoso Jalapão. Ao meu professor orientador

more »

... mar Abílio de Carvalho Júnior, pelo acompanhamento e pelo aprendizado. Aos colegas do ABSTRACT Forest fires have been significant in the Cerrado for millions of years, shaping the landscape and biotic components, at least since the Miocene, about 10 million years ago. Burnt area mapping using time series data is critical for understanding the impacts fires have on the carbon cycle and climate change. The time series of satellite images have a high potential for understanding and monitoring fire events in the Cerrado. Spectral indices have received extensive attention, due to their conceptual simplicity and good computational performance, which reduce processing time and eliminate the subjectivity of visual interpretation of burned areas. This study aims to compare the detection of burnt areas by standard time series of four spectral indices: NDVI, NBR, CSI and MIRBI. The area of study is the ecological corridor of Jalapão, which comprises different vegetation types of the Cerrado biome. The data used are images of a surface reflectance MODIS/Terra sensor, MOD09 product, corresponding to an eight day composite. The analysis considered a period of twelve years (2001-2012), totaling 552 images for each spectral index. The standardization method aims to highlight the burnt areas between different types of vegetation and avoid the mistakes present in the seasonal difference method. The normalized time series has an average of zero and a standard deviation equal to one. The definition of the limit values for the indexes between burnt and unburned areas were carried out in an automated fashion. The method compares a reference classification from Landsat TM images at various limits. The best threshold value is the one with the highest value for the Kappa coefficients and Overall. The ratings of the different spectral indices underwent accuracy analysis. Because the spectral indices showed similar results, a definitive assessment about which is best for the mapping of burnt areas could not be reached. Future research should be carried out using a greater number of images at the stage of validation and accuracy analysis.