Generation of long time series of burn area maps of the boreal forest from NOAA–AVHRR composite data

A time series of burned land areas was generated for a 23 year period (1984–2006) using 10-day composites of AVHRR data. The study area covers 1.6 million km2 of boreal forest in western Canada. The algorithm was intended to be consistent throughout the study period and region, and to avoid commission errors, so as to obtain a reliable sample of temporal trends in burned area in the region. The algorithm relies on temporal comparisons of several spectral indices (GEMI, BAI), as well as near infrared reflectance. It emphasizes the stability of the post-fire signal, to avoid false detections associated with cloud, cloud shadows, missed data and radiometric or geometric calibration between AVHRR sensors.Final results show a very consistent temporal adjustment to official statistics and fire perimeters, with very low commission error (< 10%), but medium to high omission error (50%). Burned areas in the region were predominantly associated with coniferous forest cover, with the Taiga and Boreal Shield ecozones, in latitudes between 56 and 60°N, and predominantly at long distances from populated places.