The Rupestrian Cerrado (Cerrado Rupestre) corresponds to a shrub-like vegetation of the Brazilian tropical savanna biome, occurs mainly over hilly topography and rocky outcrops, presents high biodiversity and several endemic species, and acts as barrier for agricultural expansion. In the State of Goias, large occurrences of this type of vegetation are found in the Chapada dos Veadeiros National Park (PNCV - Parque Nacional da Chapada dos Veadeiros). Rupestrian Cerrado mapping using remotely sensed data is difficult because of the spectral confusion with other phytophysiognomies, especially with Dry Forest in dry season images and with Cerrado strictu sensu in wet season images. The goal of this study was to develop a new approach to discriminate Rupestrian Cerrado based on multitemporal Landsat satellite images. The study area was the PNCV. Seven Landsat scenes were converted into surface reflectances with support of FLAASH (Fast Line-of-Sight Atmospheric Analysis of Spectral Hypercubes) atmospheric correction algorithm. The reflectances from each ETM+ band were summed and the resulting images were processed through the image segmentation technique by growing region. The segments were exported into shapefile format and the polygons related to the Rupestrian Cerrado were identified by visual analysis of the false color composite of bands 3, 4 and 5 in the computer screen. We found 24,451 hectares of Rupestrian Cerrado, which corresponds to 37% of total area of the park. The global accuracy of the mapping was 85%. As ongoing research, we recommend the inclusion of a higher number of scenes, mainly from wet season, the data integration of satellite images with digital elevation models, and the synergistic analysis of cross-calibrated Landsat ETM+ and TM data sets.
remote Sensing; Atmospheric Correction; Reflectance; Image Segmentation
