Assessment of the discrimination ability of MERIS spectral data for burned area mapping using ROC curves

P. Oliva Pavón, E. Chuvieco Salinero


Traditionally, the selection of the most appropriate bands to classify the target cover was supported by statistical indices that measured the discrimination ability of the spectral bands based on the Gaussian distribution assumption. However, that assumption might not be fulfilled in every instance. In this study we applied a non-parametric test (receiver operating characteristic, ROC) to measure the discrimination ability of MERIS sensor spectral bands and derived spectral indices to classify burned areas. The discrimination potential of each band was computed from the post-fire image, and from the temporal difference of the images. In both cases, the sources of confusion between burned areas and other covers were identified. The bands with higher discrimination ability were the NIR bands and the best indices were η, GEMI, BAI, α B8, α B10, DGEMI and DBAI.

Palabras clave

separability, non-parametric analysis, forest fires, satellite images

Texto completo:


Licencia Creative Commons

Esta obra está bajo una Licencia Creative Commons Atribución-NoComercial-SinDerivar 4.0 Internacional.

GeoFocus es la revista del Grupo de Tecnologías de la Información Geográfica de la Asociación de Geógrafos Españoles. Recibe soporte institucional y técnico de RedIRIS (Red Española de I+D soportada por el Gobierno de España), de la FECYT (Fundación Española para la Ciencia y la Tecnología) y Grumets (Grupo de Investigación Métodos y Aplicaciones en Teledetección y Sistemas de Información Geográfica).