Forest fire scar detection in the boreal forest with multitemporal SPOT-VEGETATION data

摘要

Disturbance events, such as fire, have a major impact on boreal forest dynamics, succession, and the global carbon cycle. Methods using satellite imagery are well established for detecting forest fires in real time and mapping the burned area (fire scars) within one year of the fire. This paper focuses on the detection of older fire disturbance-regeneration patterns in the boreal forests of Canada. Previous work found that shortwave-infrared image segmentation proved particularly good at creating uniform regions that were easy to associate with fire scars. Our findings suggest it is possible to detect fire scars up to ten years old using SPOT-VEGETATION data from a single year and that the use of a vegetation index based on near- and shortwave-infrared reflectance is critical to this success. We demonstrate how the use of short-term multitemporal imagery can enhance segmentation results and present a threshold-based procedure for a posteriori identification of fire scar segments. The resulting fire scar probability map showed a good correspondence with records of fire scars mapped by the Canadian Forest Service for 1980-1992 and "hot spots" from the FireM3 Information System for 1994-1998.