Out of context: fire background temperature and methods for its calculation

Abstract

Background temperature is an important component of any fire detection and monitoring method; the use of the contrast between the expected brightness temperature of a location and the brightness temperature associated with fire activity is the basis for most fire detection algorithms. The commonly used method for calculation of fire background temperature involves estimation of the surface characteristics using the immediately adjacent, non-occluded surrounds of the target pixel, in order to provide a contextually – based estimate of temperature. Depending upon conditions such as cloud, smoke, surface water and heterogeneity of land cover, this derivation of background temperature from the surrounding landscape can be vastly different from the measured brightness temperature of a pixel in a non-fire context. This paper examines the relationship between pixel brightness temperature and pixel context, to identify situations where the currently used contextual methods are most likely to perform below the required level of accuracy for fire detection. Initial results show that in many cases the ideal candidate pixels for estimating temperature at any one location may not be those that sit immediately adjacent spatially.