A global feasibility assessment of the bi-spectral fire temperature and area retrieval using MODIS data

Abstract

The use of the sub-pixel bi-spectral fire temperature and area retrieval with moderate and coarse spatial resolution satellite data has grown in recent years despite the numerous significant limitations of the method. Many of these limitations arise from a well-known sensitivity to errors in the 11-μm background radiance estimate used in the retrieval. Since this estimate is almost always obtained by averaging neighboring pixels, the accuracy of the bi-spectral retrieval is intrinsically coupled to the local 11-μm surface variability at the scale of the sensor footprint. In this paper, we explore the impact of this variability on the accuracy of the retrieval using ten years of 1-km MODIS fire data. In addition, we propose a simple a priori rejection test to identify and eliminate cases in which the retrieval is destined for failure (i.e., has no physically valid solution), or has a greater likelihood of yielding highly inaccurate fire area and average temperature estimates as a result of this variability. Finally, we examine the implications of our rejection scheme on the feasibility of performing the bi-spectral retrieval globally using MODIS and other comparable 1-km sensors. Under our proposed rejection criterion based on 11-μm background variability, the bi-spectral retrieval could be performed with sufficient accuracy (here defined as limiting the average bias in retrieved fire fraction to approximately a factor of two) for only 7% of all MODIS fire pixels detected globally during the ten-year study period. Consideration of additional error sources is likely to further reduce this fraction.