Bugs: wcs_match
Table of Contents
Caveats
Caveats
Using method=rst (default) may change the pixel size/scale which introduces problems for some down-stream analysis.
When running wcs_match with the default parameter setting: method=rst (rotate, scale, translate), it may compute an astorometric solution that includes a change to the pixel size (scale). Typically this is then applied to the Level2 event file. The updated event list may then have a non-standard pixel size which is propagated into images created by binning the event file.
However, some data products such as exposure maps, are always created using the standard pixel sizes. This mismatch in pixel size can then trigger errors or unexpected results when trying to use the two data products together.
There are two types of errors. (1) Using the same grid to bin the event file and create the exposure map may result in different image axis lengths; typically they will be off by 1 pixel. (2) Attempting to use the exposure map as a Datamodel mask() filter, will result in an error since it requires the mask and the image being filtered to have the same pixel size; this type of filtering is done with merge_obs when creating a combined PSF map.
Workaround:
Users should only run wcs_match using method=trans which forces the astrometric solution to only include linear translations.
Using method=rst (default) may result in a non-physical astrometric solution.
When running wcs_match with the default parameter setting: method=rst (rotate, scale, translate), it may compute an astorometric solution that is non-physical. Examples of non-physical solutions include large translation offsets which are compensated for with large rotation offsets in the opposite direction. The Correcting Absolute Astrometry thread has example of how to interpret wcs_match's output transform file. Given the accuracy of the Chandra pointing, offsets larger than a few pixels are highly suspicious. In addition, since the aspect camera field of view is much larger than the instruments' field of view, rotation adjustments larger than a fraction of a degree are unlikely physical.
This is caused by having relatively few matching sources and/or all the matching sources are clustered together on one side of the tangent point. Mathematically, to use method=rst requires at least 3 source pairs, however, in practice users should have dozens of matching sources dispersed throughout the entire field of view. With few sources, the under-constrained "rst" solution may be mathematically optimal, but physically impossible.
Workaround:
Users should only run wcs_match using method=trans which forces the astrometric solution to only include linear translations.