Spectral and Spatial Filter

The Spectral and Spatial Filter is intended to identify all pixels with spectral and spatial signals that suggest they may potentially contain an actively burning vegetation fire. The aim is to successfully include all the true fire pixels within the potential fire pixel set, whilst minimising the number of non-fire pixels included so as to minimise data processing overheads and avoid later false alarms. The final set of potential fire pixels is the intersection of these two sets of pixels.

The spectral filter identifies potential fire pixels using two adaptive thresholds. Firstly, the scene is divided into equal sized sub-scenes. For each sub-scene being tested, the number of land and cloud free land pixels is calculated. The sub-scene is classified as valid for use in adaptive threshold determination if the number of land pixels is greater than 10% of the total number of pixels in the sub-scene, and if the number of cloud free land pixels is greater than 1% of the total number of pixels in the sub-scene. In this case, the mean values of S7 and S8 brightness temperatures, and the difference between them for all cloud-free land pixels in that sub-scene, are tested according to a specific threshold.

However, the spectral filter process is designed to be very liberal, in order to catch any possible fire pixels. The disadvantage of this method is that it can return very large numbers of potential fire pixels which are not fires (containing large areas of solar-heated bare rocks, soil or other "warm" surfaces). In order to minimise the inclusion of such areas in the final potential fire pixel set, a series of spatial thresholds is employed in conjunction with an edge detection filter. This test is used to identify locations where the difference between the S7 and S8 signals show a marked spatial change, such as is found at fire pixels but not at areas of homogeneous warm land.

Multiple filter kernel sizes are used to ensure that the spatial filter is appropriate for detecting both single fire pixels and those belonging to larger spatial clusters of fire pixels. The edge detection filter is applied to the entire scene, and pixels passing the test are those where the filter output exceeds a specific threshold.