Day and night-time active fire detection over North America using NOAA-16 AVHRR data

Presents an investigation of the applicability of NOAA-16/AVHRR (N-16) satellite data for detecting and mapping active wildfires across North American forest ecosystems. Two fire detection algorithms were developed for application to N-16 day and night-time daily imagery. The algorithms exploit both the multi-spectral and thermal information from the AVHRR daily images. For daytime data collection N-16 provides 3 reflective bands, namely, red, near infrared and short-wave infrared (SWIR), in addition to two thermal brightness temperature bands centred at 10.8 /spl mu/m and 12 /spl mu/m. During nighttime data collection the SWIR band is switched to the thermal brightness temperature band centred at 3.7 /spl mu/m. This band in particular has been extensively used for detecting active fires. The fire algorithms have two major steps: detection of potential fires followed by false fire elimination. The threshold tests developed for fire identification and false fire removal were optimised through a trial-and-error approach using a database of active fire pixels over the whole of North America. The database was generated from a large number of single-daytime and night-time AVHRR scenes, where the fire pixels were identified visually on the images aided by the associated daytime smoke plumes. The SWIR band was found to be sensitive only to large burning fires leading to a noticeable spike in the measured reflectance. However, a serious limitation with the SWIR band is that small burning fires are not easily detectable and quite a number of active fires go undetected. Night detection does not pose such a limitation and has allowed for significantly higher detection rates. Overall night detection provides reasonably good alternative for the reduced sensitivity of the SWIR band. The set of day and night algorithms was used to generate daily active fire maps across North America. Such a combined approach for fire detection leads to an improved detection rate. Selected validation sites in western Canada and the United States, showed reasonable correspondence with the location of fires mapped by the Canadian Forest Service and the USDA Forest Service using conventional means.