Determining the use of Sentinel-2A MSI for wildfire burning & severity detection

Accurate, reliable, and timely burn severity maps are necessary for planning, managing and rehabilitation after wildfires. This study aimed at assessing the ability of the Sentinel-2A satellite to detect burnt areas and separate burning severity levels. It also attempted to measure the spectral separability of the different bands and derived indices commonly used to detect burnt areas. A short investigation into the associated environmental variables present in the burnt landscape was also performed to explore the presence of any correlation. As a case study, a wildfire occurred in the Sierra de Gata region of the province of Caceres in North-Eastern Spain was used. A range of spectral indices was computed, including the Normalized Difference Vegetation Index (NDVI) and the Normalized Burn Ratio (NBR). The potential added value of the three new Red Edge bands that come with the Sentinel-2A MSI sensor was also used. The slope, aspect, fractional vegetation cover and terrain roughness were all derived to produce environmental variables. The burning severity was tested using the Spectral Angle Mapper (SAM) classifier. European Environment Agency's CORINE land cover map was also used to produce the land cover types found in the burned area. The Copernicus Emergency Management Service have produced a grading map for the fire using 0.5 m resolution Pleiades imagery, that was used as reference. Results showed a variable degree of correlation between the burning severity and the tested herein spectral indices. The visible part of the electromagnetic spectrum was not well suited to discern burned from unburned land cover. The NBRb12 (short-wave infrared 2 - SWIR2) produced the best results for detecting burnt areas. SAM resulted in a 73% overall accuracy in thematic mapping. None of the environmental variables appeared to have a significant impact on the burning severity. All in all, our study result showed that Sentinel-2 MSI sensor can be used to discern burnt areas and burning severity. However, further studies in different regions using the same dataset types and methods should be implemented before generalizing the results of the current study.