Enhancement of a fire detection algorithm by eliminating solar reflection in the mid-IR band: application to AVHRR data

Abstract

Satellite data from the Advanced Very High Resolution Radiometer (AVHRR) have been widely employed for fire monitoring around the world by virtue of the thermal emission in the middle-infrared (mid-IR) channel at 3.7 μm. This channel, however, receives both thermal emission and solar reflection. As far as fire detection is concerned, the solar reflection contaminates the fire emission signal, which can cause significant errors, especially over non-forest biomes. This study presents a method to detect and eliminate the significant contribution of solar reflection to the AVHRR mid-IR band so that the fire detection accuracy is improved. AVHRR data from April to November 2004 were analysed. Twenty-seven percent of commission errors, mainly located in the southwestern part of North America, were found to be caused by the strong solar reflection from the surface. We also found that the calculated solar reflection itself is an effective indicator of false detections for the AVHRR. Introducing a new test to take into account this effect leads to a considerable reduction in commission errors. The new filter can eliminate most commission errors at the expense of minor increases in omission errors. The total number of true fires is missed by 0.3%, and the total number of false fire detections is reduced by 27.1%.

Acknowledgements

This study was supported by the NOAA GOES-R risk reduction project. The AVHRR data were downloaded from NOAA's Comprehensive Large Array-data Stewardship System (CLASS). The MODIS fire product was downloaded from the MODIS web fire mapper. The GOES fire product was downloaded from the National Geophysical Data Center. The HMS data were downloaded from NOAA's National Environmental Satellite, Data, and Information Service (NESDIS). The authors thank I. Csiszar and W. Schroeder for their useful comments.