| Abstract | This paper is the second part of our investigation on the remote sensing of boreal forest fires in Canada. Part I presents the development and application of a fire detection algorithm that was
developed for use with NOAA 14 across Canadian boreal forests. Part II, this paper, deals with the validation of the algorithm and comparison with another popular algorithm. Validations against air-borne fire observations indicate that almost all
fire events are captured by the satellite-based technique. In terms of burned area, the aerial accumulation of fire pixels is generally smaller than ground reports. The discrepancy is caused both by an underestimation of satellite detection due to
the presence of clouds and the low sampling frequency, and by an overestimation of ground reports due to the existence of unburned islands, water bodies, etc. inside the polygons of fire data. Nevertheless, the outliners of fire pixels match rather
well with the boundaries of the fire polygons. In addition, the performance of another fire detection algorithm used by the International Geosphere-Biosphere Programme (IGBP) was evaluated by comparing the results of ensuing fire detection with
those derived from our algorithm and with the ground reports. Note that the IGBP algorithm was designed for global application. Its performance for the specific regions and biome of Canadian boreal forests is assessed. It was found that the IGBP
algorithm tends to produce a lot of false fires over the fire scars from previous burns. A step-by-step comparison between the two algorithms helps reveal the causes of the problem and recommendations are made accordingly to improve its performance.
|