Given that some of the ARF components vary as a function of position, the spatial variation of a source should be considered in the generation of an ARF. This information can be passed to arfgen by specifying a dataset containing an image of the source distribution in DET coordinates (a so-called `detector map'). arfgen would then compute a weighted ARF based on the variation of the ARF for each pixel in the detector map, using the values in the detector map as a weighting. Ideally this map should represent a model of the incident source distribution, convolved with the telescope PSF, but with no other instrument effects applied (it fact it would be more preferable if the input source distribution did not include the PSF convolution either; but PSF convolution is unfortunately beyond the scope of arfgen). However, in most cases it is expected that the observed source imprint will suffice.
The detector map also specifies the spatial points at which arfgen extracts ARF information from the CAL. These points correspond to the centre of each pixel.
NB: The detector map image must be created in detector coordinates in the current implementation. It is a priority to allow this file to be created in sky coordinates in a future release.