Fire severity assessment by using NBR (Normalized Burn Ratio) and NDVI (Normalized Difference Vegetation Index) derived from LANDSAT TM/ETM images

TLDR

The study notes that the derived thresholds can be applied to other fires with similar characteristics, though fire‑specific thresholds may further improve mapping accuracy. The authors investigated fire severity assessment by analyzing three southern Spanish fires. They compared NBR and NDVI derived from LANDSAT TM/ETM images by examining pre‑/post‑fire pixel displacements in NIR‑MIR and NIR‑R bi‑spectral spaces and then evaluated each index’s ability, from uni‑temporal and bi‑temporal perspectives, to discriminate three severity levels. The optimal approach was to segment pixels using dNBR to separate burned from unburned, then classify extreme versus moderate severity with post‑fire NBR, achieving an 86.42 % accuracy (±4.31 %).

Abstract

In this work, the capacity of NBR and NDVI indices derived from LANDSAT TM/ETM images has been analysed for fire severity assessment. For this purpose, three fires occurring in southern Spain were studied. Firstly, the displacements of burned and unburned pixels in the pre‐/post‐fire NIR‐MIR and NIR‐R bi‐spectral spaces were analysed with the aim of establishing which of the two indices was the most sensitive for discriminating severity levels. Then, the capacity of the two indices, both from a uni‐temporal (post‐fire) and bi‐temporal perspective (pre and post‐fire), to discriminate three severity levels was studied. Based on the results, it was decided that the most suitable way to assess wildfire severity by index segmentation was to discriminate between unburned and burned pixels according to their NBR pre‐/post‐fire difference values (dNBR), and, subsequently, to distinguish between pixels with an extreme and moderate severity based on the NBR post‐fire values. The thresholds calculated for these indices permitted fire severity mapping with an accuracy of 86.42% (±4.31%). These thresholds could be extrapolated to other fires with similar characteristics although a calculation of their own specific thresholds could improve the accuracy of the fire severity map obtained.