This document describes the procedure for validating the first non-beta release of the jointcal
package as a replacement for meas_mosaic
. Longer term requirements for jointcal
are described in a separate document (draft here). Because meas_mosaic
only runs on Hyper Suprime-Cam data, the scientific performance tests here are also focused on HSC data. The jointcal
beta already supports additional cameras and we expect it to continue to do so, but as there's nothing to replace for other cameras there are no short-term requirements for its performance on them.
jointcal
shall be runnable as a command-line Task
that takes the src
and ref_cat
datasets as input and produces (at least) a wcs
and photoCalib
dataset as outputs, which provide updated astrometric and photometric calibrations.jointcal
shall perform a different fit for each tract, and allow multiple tracts to be run in single invocation (at least in serial, which is all meas_mosaic
does). Some visits may thus be processed with multiple tracts.jointcal
shall fit models that have a level of sophistication similar to those in meas_mosaic
(if it fits simpler models, it will be because the tests below demonstrate that they are sufficient to generate similar-quality results). That involves:As of and , the validate_drp
package can now (optionally) utilize the wcs
and photoCalib
datasets to calibrate the src
dataset prior to computing its astrometric and photometric accuracy metrics. To replace meas_mosaic
, we require jointcal
to match (no significant difference in metric value) or improve upon the following metrics as computed with meas_mosaic
on HSC data.
validate_drp
's check_astrometry
plot).validate_drp
's check_photometry
plot).No significant difference is a bit tricky to define here because most of these metrics are not accompanied by uncertainty estimates; to estimate them we'll use the RMS of the meas_mosaic
metric (henceforth σ) results over different tracts of HSC Wide. Differences less than 0.1 mmag for photometry or 0.1 mas for astrometry (even if larger than the cross-tract RMS, though I doubt the RMS across tracts will be that small) can also be ignored.
We will run the metric comparison independently on each of at least 19 tracts of HSC-SSP Wide data, as well as the (single tract) HSC-SSP UDeep COSMOS dataset. No more than 1 of these 20 tracts shall have a regression from the meas_mosaic performance in that tract by more than 2σ, and none may deviate by more than 3σ. The number of tracts may be extended if initial failure to achieve these rates appears to be due to statistical fluctuations.
The WCS mappings and photometric scalings output by meas_mosaic and jointcal should also be manually inspected for at least 5 visits in each of 5 tracts (including COSMOS) to look for unusual differences.
Scheduling work is obviously a T/CAM responsibility, but it may be best to try to share the generation of the input data repositories for these tests with re-run of the full HSC-SSP PDR1 (NCSA may have already scheduled some time for this later this cycle).
Scientific validation of the results is the responsibility of the Pipeline Scientist.