U-BAD: ULTIMATE BURNED AREA DETECTION

Here, we evaluated the combination of five satellites/sensors for improving burned area (BA) detection over the Amazon basin. Using a data set of 2,400 burn/no burn points by visual inspection in 2016, we investigated several spectral indices and ingested them into data mining algorithms to evaluate their burning area detection performance. Better results were provided with attribute selection combining Sentinel-2 (S2) and MODIS indexes (96 %), which were not significantly better than S2 indexes alone (95 %). The worst was the Sentinel-1 SAR data with 85 % accuracy. This is the first large-scale data research to evaluate the potentiality of combined temporal, spectral, and spatial resolutions for BA detection across the Amazon.