A comparative analysis of burned area datasets in Canadian boreal forest in 2000.

The turn of the new millennium was accompanied by a particularly diverse group of burned area datasets from different sensors in the Canadian boreal forests, brought together in a year of low global fire activity. This paper provides an assessment of spatial and temporal accuracy, by means of a fire-by-fire comparison of the following: two burned area datasets obtained from SPOT-VEGETATION (VGT) imagery, a MODIS Collection 5 burned area dataset, and three different datasets obtained from NOAA-AVHRR. Results showed that burned area data from MODIS provided accurate dates of burn but great omission error, partially caused by calibration problems. One of the VGT-derived datasets (L3JRC) represented the largest number of fire sites in spite of its great overall underestimation, whereas the GBA2000 dataset achieved the best burned area quantification, both showing delayed and very variable fire timing. Spatial accuracy was comparable between the 5 km and the 1 km AVHRR-derived datasets but was remarkably lower in the 8 km dataset leading, us to conclude that at higher spatial resolutions, temporal accuracy was lower. The probable methodological and contextual causes of these differences were analyzed in detail.