A parametrised bispectral model for savanna fire detection using AVHRR night images

Abstract

Ten years ago Dozier presented a bispectral model for the detection of sub-resolution high temperature features in NOAA AVHRR satellite images. This model builds on Planck's law and theoretically derives brightness temperatures and fractions of sub-resolution fields of high temperatures, such as wildfires, using AVHRR channel 3 and 4 thermal data. The original Dozier model was based on a larger number of assumptions. Most fire detection studies have therefore turned to simpler approaches. Still, the original model has been successfully applied in several fire studies in various ecosystems. Lately, some studies have tried to refine the model. In this study the Dozier model has been modified through inclusion of some steps of parametrisation and applied for savanna fire detection. The parametrisa-tion gives an operational and unbiased approach to extract the temperature of the non-burning part of the ground resolution cell. To bypass the problem of reflected solar radiation in the daytime in the AVHRR channel 3 wavelengths, night-time images are used. The constraint of channel 3 sensor saturation for African savanna environments in the daytime during the hottest period of the year is then also avoided. The development of the approach and its application to AVHRR night-time images are shown for images of the 1989-90 fire season of the Senegal and the Gambia, West Africa. The potential usage of the method for fire detection and analysis here, and in other ecosystems, is discussed.