IDENTIFICATION OF BURNED AREAS BY SPECIAL INDEX IN A CERRADO REGION OF THE STATE OF TOCANTINS, BRAZIL

Resumo

Obtaining information on burned areas has been studied and improved in the last decades, and the biggest question is the acquisition of consistent and detailed information about the occurrence of burnings in a simple and effective way. In view of this, remote sensing is a very interesting tool because it allows obtaining information in large areas of difficult access. The identification of areas burned by orbital data is directly related to their spectral behavior. The objective of this study was to analyze the performance of spectral indices in the identification of burned area in OLI/Landsat-8 satellite images. The indices for the before and after fire images were calculated using bands of red and near infrared: NDVI, MSAVI, SAVI, and GEMI, and bands of near infrared and short wave infrared: NBR, BAIMmod, and MIRBImod. The difference between pre and post-fire index was also calculated: dNDVI, dMSAVI, dSAVI, dGEMI, dNBR, dBAIMmod, and dMIRBImod. From these indices, six different compositions (RGB) were created and later they were segmented and classified in a non-supervised way and soon after made the extraction of the area of interest. The results of this classification were validated with the reference data obtained through the visual interpretation of the image. The methods had shown a good quality of classification, with a percentage of accuracy ranging from 85.54 to 92.46% and Kappa value of 0.70 to 0.89. The best method was the dNBR, NBRpost-fire, and dMIRBImod indices in the RGB composite.