Characterizing sub-pixel landsat ETM plus fire severity on experimental fires in the Kruger National Park, South Africa

Abstract

Burn severity was quantitatively mapped using a unique linear spectral mixture model to determine sub-pixel abundances of different ashes and combustion completeness measured on the corresponding fire-affected pixels in Landsat data. A new burn severity index was derived that is shown to map three categories of burn severity on three experimental burn plots in the southern Kruger National Park, South Africa. Those pixels which corresponded to a greater abundance of white ash were found to be significantly related to the pre-burn above-ground fuel biomass and an indicator of burn efficiency. Landsat ETM+ combustion completeness was most significantly related to the abundance of post-burn residual, non-photosynthetic fuel biomass. For the same reflectance change in pre- and post-fire imagery, a greater magnitude of fire severity was measured on corresponding ETM+ pixels. This implies that fire severity depends more on the colour of the ash than on the magnitude of change in reflectance. Burned area mapping methods that rely on reflectance change in multi-temporal imagery may not reliably characterize burn effects such as fire severity and efficiency in semi-arid savannas.