Abordagem metodológica baseada nos dados multitemporais MODIS EVI/NDVI para a classificação da cobertura vegetal na região do Parque Nacional da Chapada dos Veadeiros/GO
Savannas are the main vegetation type in Central Brazil, covering approximately 23% of the national territory. Locally known as Cerrado, Brazilian savannas are formed by a mosaic of different physiognomies such as grassland, shrubland and woodland that have a typical phenological cycle. In this context, the MODIS data provide daily measurements well suited to monitor the seasonal phenology of vegetation. The present work aims to evaluate the advantages of the temporal signatures to detect Brazilian Savanna vegetation types in the Chapada dos Veadeiros National Park. The adopted methodology was subdivided into the following steps: (a) elaboration of the 3D cube of NDVI and EVI from MODIS temporal images, where the z profile corresponds to temporal signature, (b) noise elimination, (c) endmember detection, and (d) classification. The noise elimination utilized three methods: (a) exclusion of the image with high percentage of noise, (b) application of the three-point moving median filter to smooth temporal spectrum, and (c) employement of the Minimum Noise Fraction Transformation. Endmembers? automatic identification encompasses the following steps: a) spectral reduction by the Minimum Noise Fraction (MNF) transformation, (b) spatial reduction by the Pixel Purity Index (PPI), and (c) manual identification of the endmembers using the N-dimensional visualizer. In the classification we used two methods: (a) Iterative Self-Organizing Data Analysis Techniques (ISODATA), and (b) tree decision. The results demonstrated that the savanna physiognomies present typical temporal signatures. The endmembers corresponded to grassland, woodland and cultivate area. The grassland signature characterized by lower values in the study period; the woodland by higher values and agriculture areas by a higher variation with higher values in the raining season and lower values in the dry season. The unsupervised classification by ISODATA method allowed a priori knowledge of the data, which the map classification obtained a defined distribution. The tree decision enables an approach classification obtained the classes of the grassland, woodland and agriculture. Comparison with Landsat 7/ETM+ image demonstrated the classification efficiency of the temporal series. The study concluded that the NDVI and EVI series is useful in differentiation amount vegetation types. The methodology efficiency has been proved for regional delimitation of savanna physiognomies even considering the low spatial resolution of the 250m MODIS sensor and consequently with high spectral mixture.
Advisor:Renato Fontes Guimarães; Edson Eyji Sano; Osmar Abílio Carvalho Júnior; Yosio Edemir Shimabukuro
School:Universidade de Brasília
Source Type:Master's Thesis
Keywords:análise multitemporal savanna change detection analysis
Date of Publication:07/03/2007