Publications

Siqueira, RG; Moquedace, CM; Silva, LV; de Oliveira, MS; Cruz, GD; Francelino, MR; Schaefer, CEGR; Fernandes, EI (2025). Do finer-resolution sensors better discriminate burnt areas? A case study with MODIS, Landsat-8 and Sentinel-2 spectral indices for the Pantanal 2020 wildfire detection. INTERNATIONAL JOURNAL OF REMOTE SENSING, 46(10), 3968-3991.

Abstract
In the last decades, the Brazilian Pantanal biome has been experiencing an increase in the frequency, duration, and intensity of fire regimes, culminating in the catastrophic wildfires of 2020. Due to its high fire susceptibility, strategies for prevention, monitoring, and mitigation, supported by remote sensing products, are pressingly needed in the biome. Generally, it is assumed that finer resolution sensors tend to produce better results in detecting burnt areas due to their greater spatial resolution. To test this assumption, we assessed the sensitivity of seven spectral indices (EVI, NBR, NBR2, NDMI, NDVI, MSAVI, SAVI) to separate burnt and unburnt areas, generated from three satellite sensors: Terra/Aqua MODIS, of 500 m of spatial resolution; Landsat-8 OLI, of 30 m, and Sentinel-2 MSI of 10/20 m. For this, we used two sets of validation samples: fire foci and fire scars, aiming to provide confidence in our evaluation. The spectral indices based on the shortwave infrared - near-infrared relationship presented higher M separability values, with the fire index NBR standing out, followed by NBR2 and NDMI. Overall, the spectral indices generated from the MODIS sensor showed the best performance, with some exceptions where a few indices from Sentinel-2 presented better results. The finer spatial resolution sensors identified a considerable larger burnt area, mainly due to the inclusion of small fires located in the southwestern part of the study area. However, the greater commission errors from these sensors indicated that a significant portion of these fires do not correspond to actual burnt areas but were probably small zones of deforestation. Our results reveal that more detailed remote sensing products used for fire monitoring do not necessarily improve burnt area detection while being more computationally expensive.

DOI:
10.1080/01431161.2025.2496000

ISSN:
1366-5901