Testing the potential of multi-spectral remote sensing for retrospectively estimating fire severity in African Savannahs

The remote sensing of fire severity is a noted goal in studies of forest and grassland wildfires. Experiments were conducted to discover and evaluate potential relationships between the characteristics of African savannah fires and post-fire surface spectral reflectance in the visible to shortwave infrared spectral region. Nine instrumented experimental fires were conducted in semi-arid woodland savannah of Chobe National Park (Botswana), where fire temperature (Tmax) and duration (dt) were recorded using thermocouples positioned at different heights and locations. These variables, along with measures of fireline intensity (FLI), integrated temperature with time (Tsum) and biomass (and carbon/nitrogen) volatilised were compared to post-fire surface spectral reflectance. Statistically significant relationships were observed between (i) the fireline intensity and total nitrogen volatilised (r2=0.54, n =36, p \u3c0.001), (ii) integrated temperature (Tsum-µ) and total biomass combusted (r2=0.72, n =32, p \u3c0.001), and (iii) fire duration as measured at the top-of-grass sward thermocouple (dtT) and total biomass combusted (r2=0.74, n =34, p \u3c0.001) and total nitrogen volatilised (r2=0.73, n =34, p \u3c0.001). The post-fire surface spectral reflectance was found to be related to dt and Tsum via a quadratic relationship that varied with wavelength. The use of visible and shortwave infrared band ratios produced statistically significant linear relationships with fire duration as measured by the top thermocouple (dtT) (r2=0.76, n =34, p \u3c0.001) and the mean of Tsum (r2=0.82, n =34, p \u3c0.001). The results identify fire duration as a versatile measure that relates directly to the fire severity, and also illustrate the potential of spectrally-based fire severity measures. However, the results also point to difficulties when applying such spectrally-based techniques to Earth Observation satellite imagery, due to the small-scale variability noted on the ground. Results also indicate the potential for surface spectral reflectance to increase following higher severity fires, due to the laying down of high albedo white mineral ash. Most current techniques for mapping burned area rely on the general assumption that surface albedo decreases following a fire, and so if the image spatial resolution was high enough such methods may fail. Determination of the effect of spatial resolution on a sensor’s ability to detect white ash was investigated using a validated optical mixture modeling approach. The most appropriate mixing model to use (linear or non-linear) was assessed using laboratory experiments. A linear mixing model was shown most appropriate, with results suggesting that sensors having spatial resolutions significantly higher than those of Landsat ETM+ will be required if patches of white ash are to be used to provide EO-derived information on the spatial variation of fire severity