Improved Detection of Nighttime Fire and its Combustion Efficiency for Wildfires from VIIRS during FIREX-AQ Campaign

Abstract:

Biomass burning plays a significant role in the Earth’s atmosphere and climate system. They emit radiatively important greenhouse gases and smoke particles, which disturb the atmospheric radiative balance and affect climate and air quality regionally and globally. An accurate estimation of biomass burning emissions is partially limited by the missing detection of fire pixels and the lack of knowledge of fire burning phase. Current operational nighttime detection algorithms usually use a bi-spectral approach which takes advantage of infrared channels such as ~4 and ~11 um to separate the hot source from the background due to their different sensitivities to fire and background. However, for cool or small-scale fires, the signal from the 4 um is usually weak and will fail to pass the fined-tuned thresholds for fire detection. In this research, we develop a hybrid nighttime fire detection algorithm by utilizing the Visible Infrared Imaging Radiometer Suite day-night band (DNB) and moderate band (M band) to derive the visible energy fraction and estimate the combustion efficiency. First, a seasonal nighttime visible light and thermal temperature climatology are generated based on multiple years of nighttime observations. Second, possible fire pixels are identified by fitting the real-time DNB and M band observations into statics tests formed based on the derived climatology. Third, relax contextual tests adopted from the operational algorithms are applied to the potential fire pixels to exclude false alarms. Fourth, the visible energy fraction is determined as the ratio of the derived visible light power to the fire radiative power. Lastly, based on the linear coefficients obtained in our previous studies, modified combustion efficiency of each fire pixels is estimated. We applied our algorithms to the FIREX-AQ Campaign dataset. We found that during the William Flat fire, the number of nighttime fire pixels detected by our refined algorithm was ~500 while the operational algorithm was only ~200. For small scale fires, such as Cow fire, ~200 fire pixels were detected in a total of 13 days while the operational algorithm only detected ~30 fire pixels out of the total 8 days. The new algorithm improved the detection of the fire pixels which allows for a more detailed analysis of the fire progression and combustion phases.