Remote sensing of Canadian boreal forest fires: hotspots, burned area, and smoke plumes

A comprehensive investigation of Canadian boreal forest fires was conducted using NOAA-AVHRR. Algorithms were developed for detecting active fires (hotspots), burned areas, and smoke plumes, which employ single images and 10-day Normalised Difference Vegetation Index (NDVI) composites. The hotspot algorithm was applied to four years (1994-97) of imagery, producing a daily fire mask for Canada. Almost all fire events were detected, but cumulative hotspot area was significantly smaller (/spl sim/30%) than burned area reported by fire agencies. To provide more accurate estimates of burned area, a burn mapping algorithm was developed that synergistically combines the hotspot product with anniversary date NDVI composites. The hybrid technique produced estimates of Canada-wide burned area that were within 5 percent of official statistics. A neural-network classifier was also developed that allows smoke plumes to be effectively separated from cloud cover at a regional scale.