Burned-area mapping of the Serengeti-Mara region using MODIS reflectance data

Fire is a key factor for vegetation structure and ecosystem functioning in the Serengeti-Mara region in East Africa. However, there is a lack of accurate and consistent information on fires. We developed an algorithm for mapping burned areas in the wider Serengeti-Mara region from remote-sensing data. The algorithm is automated and, once trained for one year, runs independently for all years. It uses daily measurements of red and near-infrared (NIR) reflectance acquired by the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor at a spatial resolution of 250 m. The MODIS time series was first converted to ten-day minimum NIR composites. Each composite was then classified into new and old burned areas, by thresholding burned-area index and temporal difference of NIR reflectances. The algorithm adjusts detection thresholds dynamically using measures related to atmospheric and vegetation conditions. Having trained the algorithm once on 2003 data, it was applied to MODIS data from April 21, 2000 to November 10, 2005. Accuracy was improved by size- and location-sensitive filters. Overall accuracy was 90.3% as determined from Advanced Spaceborne Thermal Emission and Reflection Radiometer satellite imagery from 2005 and 87.1% as determined using field data from 2005. The algorithm holds the potential to be applied to other savanna areas. This letter provides a reliable product useful for future investigations of fire ecology and fire management.