EVALUATION OF ALGORITHMS FOR LAND USE AND LAND COVER CLASSIFICATION IN THE CENTRAL PORTION OF RIO GRANDE DO SUL STATE FROM HIGH AND MEDIUM SPATIAL RESOLUTION IMAGERY

• Autor(a) principal: Silva Oliveira, Helena
• Data de Publicação: 2020
• Outros Autores: Marchesan, Juliana, Alba, Elisiane, Honnef, Dionatas Henrique, Frigo Wolfer, Matheus, Soares Pereira, Rudiney
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: Geo UERJ
• Texto Completo: https://www.e-publicacoes.uerj.br/geouerj/article/view/43259
• topic: Landsat 8. RapidEye. MLE. Bhattacharyya. Kappa.; Landsat 8. RapidEye. MaxVer. Bhattacharyya. Kappa.

Analyzing classification algorithms of land use and land cover as well as images from sensors on satellites with different spatial resolutions are essential to determine the most suitable for each location. The objective of this study was to evaluate the efficiency of supervised classification algorithms, Maximum Likelihood (MLE) and Bhattacharyya, using medium spatial resolution (OLI/Landsat 8) and high (REIS/RapidEye) images in localized municipalities in the central of Rio Grande do Sul state. For this were used OLI/Landsat 8 and REIS/RapidEye sensor images with spatial resolution of 30 and 5 m, respectively. The classification of both images was performed by the MLE and Bhattacharyya algorithms with the definition of six classes of land use and land cover, these being Native Forest, Planted Forest, Exposed Soil, Agriculture, Field and Water. To evaluate the efficiency of the classification were used 120 points distributed randomly stratified in each municipality, 20 points in each class of land use and land cover. The quality of the classification was analyzed by Kappa and global accuracy indices, and the error of omission and commission was calculated. According to the results, the kappa index was higher for the classifications using the REIS/RapidEye sensor images for both algorithms, totaling 85.33% (MLE) and 83.67% (Bhattacharyya). In this context, it was possible to conclude that the REIS/RapidEye images and the MLE algorithm stand out for the best results, which are more adequate for the study area.