This study employs remote sensing techniques and machine learning models to monitor and analyze Amazon forest fires, one of the most significant environmental issues in the contemporary world. The methodology relied on the integration of multi-sensor data, including high-spatial resolution Sentinel-2 and Landsat-8 images, along with thermal data from VIIRS and MODIS sensors, to achieve a balance between spatial and temporal resolution. Comprehensive preprocessing was performed, including geometric and spectral correction, and cloud and smoke removal using cloud masks.

The study used spectral indices such as the Normalized Burn Ratio (NBR) and ΔNBR, in addition to the Random Forest model to classify burn intensity, while a U-Net network was used to segment burned areas at the pixel level. The results showed that the integration of optical and thermal data significantly improved early detection efficiency, and the use of machine learning algorithms increased classification accuracy to over 85%. Deep models proved effective in delineating affected areas, while radar data showed promising potential for enhancing monitoring in extreme weather conditions.

The study confirms that the integration of space sensing and artificial intelligence represents an effective approach for monitoring fires with high spatial and temporal resolution, paving the way for the development of more reliable early warning systems to support environmental management strategies in the Amazon Basin.