Favoritar este trabalho
Como citar esse trabalho?
Resumo

Spectral Indices (SI) are widely used for remote sensing application because they enhance targeted features in optical images through the algebraic combination of spectral bands. There is a large variety of SI, in which the performance varies depending on the user's application. Considering the different emphases that spectral indices may offer, here we present a test-case based on the combination of 10 SI in a three channels remote sensing image (Red; Green; Blue - RGB) aiming to highlight burned areas from other targets such as vegetation and water. This process generated 120 possible combinations without repetition. With spatial resolution of 30m, the proposed method was able to achieve an accuracy between 0,21 and 0.86, according to Cohen's Kappa coefficient. The two groups of indices MIRBI, NBR2, EVI, MNDWI and CSI; and BAI, NBR and NDVI were the most inaccurate and accurate indices, respectively, identified for the study site.

Instituições
  • 1 Instituto Nacional de Pesquisas Espaciais - SJC
  • 2 Centro de Nacional de Monitoração e Alertas de Desastres Naturais
Eixo Temático
  • Classificação e mineração de dados
Palavras-chave
Rainforest
Landsat-8
Spectral Index
forest fires
Burnt Area