Satellite-based assessment of fire hotspots in the Brazilian Amazon: implications of sensor differences and climatic drivers

Reliable monitoring of fire activity in the Brazilian Legal Amazon is essential for understanding fire dynamics and supporting environmental management. Satellite-derived fire hotspot products are widely used for this purpose; however, differences among sensor and the occurrence of redundant hotspot detections may affect the consistency and interpretation of these datasets. This study evaluates the spatiotemporal dynamics of fire hotspots in the Brazilian Legal Amazon from 2012 to 2021, with emphasis on the assessment and refinement of satellite-derived datasets. Fire hotspot records from the Brazilian National Institute for Space Research were obtained from the AQUA (MODIS), SUOMI-NPP (VIIRS) and NOAA-20 (VIIRS) satellites. A spatial filtering procedure was applied to merge redundant hotspot detections associated with the same fire event according to the spatial resolution of each sensor. The filtered datasets were evaluated using Kernel Density Estimation, normalized kernel density maps, Principal Component Analysis (PCA), and Pearson correlation with the Oceanic Niño Index (ONI). Results indicate substantial reductions in hotspot counts after redundancy filtering (up to ~ 66% for VIIRS sensors), highlighting the importance of preprocessing satellite-derived hotspot datasets before spatial analyses. Despite this reduction, normalized kernel density maps revealed highly consistent relative spatial patterns between the original and filtered datasets, indicating that the filtering procedure preserves the principal fire-prone regions while providing a more realistic representation of spatially aggregated active-fire detections. The highest hotspot concentrations remained consistently located along the Arc of Deforestation, whereas the SUOMI-NPP and NOAA-20 products exhibited greater spatial agreement than AQUA due to their similar sensor characteristics. Precipitation-related variables were identified as the climatic variables most strongly represented in the PCA explaining data variance, and a moderate positive correlation with ENSO variability was observed. Overall, the results demonstrate that redundancy filtering improves the reliability of spatial analyses based on satellite-derived hotspot products without compromising the regional distribution of fire occurrence. This approach provides a robust methodological framework for wildfire monitoring and enhances the interpretation of long-term active-fire datasets in tropical regions.