Detection of savanna fires in Bolivia using ATSR-2 data

The overall aim of this work is to evaluate ATSR-2 data for savanna fire monitoring. Vegetation fires are generally understood have major impacts on the biosphere and understanding fire dynamics is necessary to develop models of global environmental change. Initially, the spatial and temporal distribution of fires needs to be characterised. The ATSR-2 has the potential to improve our ability to detect fires and, in particular, to discriminate recently burnt areas. This work is concerned with savanna fires in Bolivia, which contains different savanna types representative of those found elsewhere in South America. Bolivia is a region of widespread fire and fire regimes there have not been studied using Earth Observation data. Fire and fire scar detection algorithms were developed and applied to data obtained over Bolivia during the 1995 fire season. The results were evaluated and used to derive an estimate of biomass consumption.;Using ATSR-2 data it was possible to detect active fires and scars in the dry savanna landscapes found in the study area. Errors of commission were low but errors of omission resulted from an incomplete image dataset. Fire scars were best discriminated by a combination of the NIR (0.87m), SWIR (1.6m) and thermal (11.0m) channels. The forward-view data and data from the visible channels (0.55m and 0.67m) did not provide useful or additional information about burnt surfaces. The scar detection algorithm identified 88% of the total detected fire events. The algorithm results were applied to a fuel-load map, derived from Landsat TM data combined with field data, to estimate burnt biomass. The detected fires were found to have consumed 4% of the total available fuel-load. This figure was sixteen-fold lower than an estimate derived using a traditional approach without EO data. These results are in agreement with the trend in such studies of lower estimates of burnt biomass.