A stochastic model for active fire detection using the thermal bands of MODIS data

Active fire detection using satellite thermal sensors usually involves thresholding the detected brightness temperature in several bands. Most frequently used features for fire detection are the brightness temperature in the 4-μm wavelength band (T₄) and the brightness temperature difference between 4- and 11-μm bands (/spl Delta/T=T₄-T/sub 11/). The task of active fire detection is examined in the context of a stochastic model for target detection. The proposed fire detection method consists of applying a decorrelation transform in the (T₄,/spl Delta/T) space. Probability density functions for the fire and background pixels are then computed in the transformed variable space using simulated Moderate Resolution Imaging Spectroradiometer (MODIS) thermal data under different atmospheric humidity conditions and for cases of flaming and smoldering fires. The Pareto curve for each detection case is constructed. Optimal thresholds are derived by minimizing a cost function, which is a weighted sum of the omission and commission errors. The method has also been tested on a MODIS reference dataset validated using high-resolution SPOT images. The results show that the detection errors are comparable with the expected values, and the proposed method performs slightly better than the standard MODIS absolute detection method in terms of the lower cost function.