The obs_subaru camera geometry currently uses an effective focal plane scaling of "unity" (i.e. plateScale = 1 arcsec/mm, pixelSize = 1 mm). All other LSST camera geometries are defined in actual physical units: millimeters on the focal plane. For the work being done in DM-19943 (and see
RFC-616), we need to have the correct CCD plate scale in the camera geometry to properly reflect the pixel-to-focal plane conversion.
In addition to this unit conversion being a good idea in its own right, the investigations in DM-19943 rendered us fairly confident that there is an inconsistency in the current HSC cameraGeom itself that was actually responsible for most of the problems. Updating the units thus has the added bonus of eliminating the inconsistency without us having to debug it further. Furthermore, HSC's non-conventional unit system has caused mild headaches for a few developers (e.g. erykoff's calibration work on fgcmcal), so the switch will benefit them as well.
There are a few package affected and will require accommodation. These have already been exposed and tested on
DM-20289 for singleFrame and skyCorrection and, in the context of ci_hsc, through coaddition and multiband with the cursory tests there (but without running jointcal, which is not yet included in any of our CI scripts). A more rigorous battle testing is being performed on DM-20548 (the ticket triggered by this RFC) which will include a full run of at least one tract from the RC2 dataset (i.e. running through singleFrameDriver, skyCorrection, jointcal, coaddDriver, and multibandDriver) to make sure the results are consistent with that of the latest weekly reprocessing run.
One item of note: this change will effectively break meas_mosaic. While updating it to accommodate the unit update is not completely out of the question, it would take significant effort which does not seem worthwhile given the adoption of jointcal as our “uber calibration” algorithm. As such, this RFC is also effectively proposing the official retirement of meas_mosaic.