Active fire detection using Landsat 8 reflective bands

Abstract:

Vegetation fires can alter landscapes and are a significant source of atmospheric emissions, particulates and greenhouse gases. Currently, only coarse spatial resolution sensors with high temporal coverage, such as MODIS or VIIRS, are used for routine global active fire mapping. Higher spatial resolution satellites have not been used due to their low temporal coverage and so are less useful for monitoring fires at the time of satellite overpass. However, looking forward, combination of the recently launched Landsat-8 (2013), Sentinel-2A (2015) and upcoming Sentinel-2B (2016) sensor data will provide 10-30m global coverage multi-spectral reflective wavelength observations approximately every three days. Therefore the development of reflective wavelength active fire detections to take advantage of these new data is highly attractive. Conventional detection algorithms use the elevated thermal emission of fire to detect the location of fires burning at the time of satellite overpass and apply contextual checks to remove commission errors by examination of neighboring pixels. A Landsat 8 active fire detection algorithm that takes advantage of the improved 12-bit radiometric resolution and high reflectance saturation of the Landsat 8 OLI detectors is presented. The algorithm uses the 1.6 μm and 2.2 μm bands without the need for a contextual implementation, or thermal bands, and was parameterized using six months of Landsat 8 data over the conterminous United States. Active fire detection results for Landsat 8 scenes acquired over a range of fire sizes and temperatures in Canada, Brazil and Southern Africa are presented and compared to detections found using an existing Landsat 7 contextual algorithm adapted to the Landsat 8 bands. Results show that the Landsat 8 algorithm has potential for global application, with relatively low omission and commission errors, and is suitable for application to the corresponding Sentinel 2 reflectance wavelength bands.