Fire disturbance mapping in a northern boreal forest using AVHRR/NDVI imagery : comparing techniques of change detection and substrate correction

Abstract

Ontario’s landmass north of 51° latitude has been designated as the least intensively managed zone with respect to fire protection. Consequently, large natural fires dominate the boreal landscape. This northern zone, comprising nearly 43-million hectares of primarily unmanaged boreal forest, is virtually inaccessible by road. Understanding the fire dynamics of such an extensive, unmanaged, and inaccessible forest is valuable for further ecological research. Three methods of analyzing Advanced Very High Resolution Radiometer (AVHRR)/ Normalized Difference Vegetation Index (NDVI) imagery were compared for post-fire detection and mapping in northern Ontario. Exposed ground conditions, resulting from the removal of forest cover, can inflate derived AVHRR/NDVI values. Corrections for these effects were applied to test whether fire-mapping accuracy could be increased. The suitability of three threshold-driven change detection methods developed in Alaskan boreal forests were tested under northern Ontario conditions along with three corrections for substrate reflectance for 1992, 1993, and 1995 fire seasons. A factorial ANOVA statistical design was implemented to test equality among various spatial coincidence variables. Accuracy was assessed using a spatial database. Existing fire detection and mapping methods using AVHRR/NDVI are not directly suitable for use in northern Ontario. Results indicate the superiority of a strict, single-threshold method for reducing false detection and the lenient double-threshold method for increasing mapped area for each fire. Furthermore, strong correlation was found between fire size and the area detected to represent them; however, corrections for substrate reflectance did not significantly increase detection and mapping accuracy.