Contextual and multiple‐threshold algorithms for regional active fire detection with AVHRR data

This paper compares the detection of active fires using NOAA‐AVHRR 1 km HRPT data by two different methods, one a multi‐channel threshold algorithm and the other a contextual algorithm. The comparison is undertaken with the aim of assessing if the two methods of data extraction perform equally and to document logistical limitations of each method. The comparison is undertaken using 15 NOAA‐11 afternoon passes over the Central African Republic (RCA) from January 1994, a region in which there is extensive fire activity, occurring across a variety of different ecosystems. The work shows that the two methods detect the same spatio‐temporal trends in active fire distribution, with nearly 70% of the fires detected being co‐incident. The contextual algorithm, however, appears to extract some 20% more fires than the threshold method. Visual analysis of the data was carried out to see if differences could be explained. On the whole the differences were found to be mainly a question of degree, with many more pixels detected by one method for larger fires. The main exception to this general trend was found to be in the areas where the threshold algorithm seems to have been set too low, extracting warm surfaces, and highlighting the drawback of such methods. Given the greater amount of preparation required for the threshold method, it is suggested that for this region, the contextual method is a more efficient and consistent method for the extraction information on fire activity.