An algorithm for extracting burned areas from time series of AVHRR GAC data applied at a continental scale

This study describes the methodology developed to detect burned surfaces us ing a long-time series of low-resolution satellite data. NOAA-AVHRR-GAC 5 K m images were used because they constitute a very complete historical data set of satellite imagery over Africa. The Burned Area Algorithm (BAA) relie s on a multitemporal multithreshold approach, based on the spectral changes of the land surface after a fire occurrence. By using different sets of AV HRR channels and derived Indices, spectral signatures have been determined for burned and unburned surfaces. Indices that make use of the information contained in Channel 2 and Channel 3 are the best for detecting burned area s. The results showed excellent agreement at the continental scale with kno wn temporal and spatial patterns of active fires. Validation of the algorit hm by comparison with a number of Landsat TM images, which were classified in terms of burned and unburned surfaces, showed an overall accuracy of 71% . (C) Elsevier Science Inc., 1999.