A Comparative Study of Three Algorithms for Forest Fire Detection in Iran

Fire as a natural disaster plays a major role in deforestation; is a major source of trace gases, aerosols and carbon fluxes. Approximately 7 percent of Iran area is covered by forests/and wooded lands. According to the ECE/FAO database on forest/other wooded land fires in Iran (1982-1995), the number of fires per year is 130 and the average area burnt per year is 5400 ha, but fires are not largely monitored and enough detection facilities are not available. The use of remote sensing is now a useful alternative, especially in vast and remote areas. The objective of this paper is to inspect the applicability of using MODIS imagery for forest fire detection in Iran, so a comparative study between a graph-based algorithm proposed by Byun et al. (2005) and two contextual algorithms developed by Giglio et al. (2003) and Wang et al. (2007) using MODIS data acquired over Kerman shah province in summer 2006 is present. The results were compared with ground-based observations collected with GPS. It implies that the algorithm developed by Giglio et al. (2003), MODIS version 4 contextual algorithm, performed best with the least commission and omission errors; has potential to be applied in detecting the fire pixels that start in or around forests and are mainly surface fires and seldom crown fires. As a result, using daily MODIS data, it is possible to track fire development in Iran.