Experimental comparison of four remote sensing techniques to map tropical savanna fire-scars using Landsat-TM imagery

A landscape-scale fire experiment, conducted over two consecutive dry seasons in a large tract of tropical savanna in northern Australia, was used to evaluate four methods to map fire scars apparent on Landsat-TM imagery: (i) systematic visual; (ii) semi-automated; (iii) automated; and (iv) change detection. All of the methods showed rapid fading of the fire scars. Overall, the automated and visual methods were able to discriminate burnt areas for longer than the other methods. However, the automated method also falsely identified fire-scars on between 5 and 20% of the unburnt catchments prior to the experimental late dry season fire treatments. One cause of the fading appears related to the increased flushing of tree canopies on burnt areas, although the spatially patchy recovery within and between catchments points to the importance of other factors such as the recovery of the ground layer. It appears that Landsat-TM imagery cannot be used to reliably determine the spatial extent and timing of fires in environments with rapid post-fire recovery, such as tropical savannas, thereby limiting the utility of this data source for fine-scale ecological studies.