Abstract
A multilayer fire detection model that offers improved sensitivity to smaller fires with a significantly lower false alarm rate has been proposed for fire automatic recognition using the thermal bands of MODIS data. Non-fires are gradually eliminated by multi-contextual criteria derived from the brightness temperature in the 4pm wavelength band and the brightness temperature difference between 4 mu m and 11 mu m bands. Experiments with the high quality data from MODIS covering Heilongjiang region of China through validation have suggested several improvements to the original MODIS active fire detection algorithm.