59460

Desafios na classificação automática de fitofisionomias do Cerrado brasileiro com base em mapas de referência na escala 1:250.000

Favoritar este trabalho

Brazilian Savanna, also known as Cerrado, is one of the most important biomes in world in terms of biodiversity. Mapping Cerrado is an important task, considering that deforestation reached almost half of its original area. Two classification systems are currently used to map Cerrado Land Cover and reference maps are available on the scale of 1:250.000. Thus, the aim of this study is to discuss the challenges regarding the automatic classification of Cerrado land cover using the reference maps of 1:250.000 produced by Brazilian Institute of Geography and Statistics (IBGE). Three protected Cerrado areas were used in this study, Brasília National Park (DF), Emas National Park (GO) and Chapada das Mesas National Park (MA). Images from the Landsat-8 satellite, acquired in the dry and wet seasons, were used in the classification. Images were segmented and classified according to Brazilian vegetation classification system. Classification was performed through the random forest algorithm. The classification results pointed out an overall accuracy of 84.4%. The main source of classification error was transition areas among the vegetation formations. Due to the map scale, some areas close to the edges are not distinguished with precision and may be incorrectly classified. The use of high resolution images can improve the classification results in the vegetation boundaries. Another notable problem is that the Brazilian vegetation classification system does not separate Gallery Forests from other classes although the segmentation has distinguished these segments quite well.