An Improved Algorithm of Wild Fire Detection for Modis Imagery

As well known, wildfire emits carbon into atmosphere for 1.7 to 4.1GtC/yr in entire earth (IPCC, Mack et al. 1996, Andreae et al. 2001). This is not negligible amount, corresponding to one quarter to one half of anthropogenic greenhouse gas emission. Thus, accurate wild fire detection is important in estimation of the impact of wild fire and in disaster management. In fact, accuracy of wild fire detection algorithm was remarkably improved by MOD14 algorithm (Giglio et al. 2003). However, we still have many false alarm and omission errors in boreal forest. One of the reasons is mixture of reflection and emission in burning area in short wavelength infrared. Thus, considerations of a variable for stochastic test and of choice of candidate of fire pixels are very important. At first, authors modified MOD14 algorithm to improve sensitivity using 2-dimensional stochastic test (Nakau et al. 2008). This algorithm detects about 15% more hotspots. Secondary, we propose an algorithm to detect wild fire using estimated radiation by wild fire. As a result of these improvement, these two improved algorithms detected 16% and 67% more hotspots in 408 MODIS imagery covering Alaska in 2004. Precise validation including in Africa using ASTER imagery will be presented in the presentation.