Evaluating spectral indices for estimating burned areas in the case of Izmir/Turkey

Abstract

Mapping and determination of fire damaged areas in an accurate and prompt way is essential for identifyingenvironmental losses caused by fires, post-fire management activities and planning strategies. In this context,this study aims to evaluate the performance spectral indices for discriminating burned and unburned areas in theimmediate post-fire environment in the case of Gaziemir, Buca and Karabağlar districts of Izmir metropolitancity where one of the forest fires occurred in the 18rd August 2019. For this, whilst a Sentinel 2A (26th August2019) was used to map burned / unburned areas as the reference dataset, two Landsat 8 satellite images (7th and28th August 2019) were used for the calculation of spectral indices. The spectral indices of normalised differencevegetation index (NDVI), atmospherically resistant vegetation index (ARVI), two versions of normalised burnratio (NBR and NBR2) and burnt area index (BAI) were calculated for the selected two dates as well as pre-fireand post-fire temporal differences in those indices. For the performance comparison of spectral indices, binarymaps of burned and unburned areas were created and separability index (SI) was calculated for pre/post-firedifferenced spectral indices. Our results suggest that NBR2, NDVI and ARVI had the highest potential fordiscriminating burned areas, respectively. Even though the value of separability indices was different from eachother where NBR and BAI had the lowest values, that doesn’t necessarily mean these indices cannot discriminateburned areas since the separation of burned and unburned areas highly depend on the spatio-temporalcircumstances e.g. vegetation types and time lags between image acquisition dates.