Optimum systems for satellite fire detection

Significant improvements on the detection of thermal anomalies in multispectral satellite data can be obtained when both the false alarm rate and the probability of detection are known. A desirable, optimum system should have constant false alarm rate and maximum probability of detection. While proper hypotheses can be done on the background statistical distribution, on target for constant false alarm rate, a statistical model for thermal anomalies is not easily available, and, consequently, the detection probability is not available in a closed form. Therefore, an appropriate description of the physics of the phenomenon is required: the mixed pixel model provides a valid answer to this need. Once a physical model is available, a Monte Carlo simulation can be performed for evaluating the probability of detection. Simulated data of thermal anomalies have been compared with the fire pixel temperatures from NASA-DAAC MOD14 showing the good agreement between simulated and experimental data. Finally, the receiver operating characteristic of a constant false alarm rate system has been derived through simulation, and comparisons between optimum and non-optimum systems as well as between systems with different rules for the channels fusion have been carried out.