Issues with CCDs that may have affected the quality of DECam observations are recorded on the DECam CCDs page.
The columns nobjs, npsf, and so on are signed 16-bit integers. One brick, 2296p020, containing the globular cluster M5, contains 32,943 PSFs (and no other types), which overflows a 16-bit signed integer. It therefore appears as -32,593 (off by 65,536).
Some morphological quantities are inconsistent for PSF sources in DR8, due to a bug in the model selection function that was introduced on May 13th, 2019 to fix a different bug in the reduction process.
This issue occurs when the REX model is better than the PSF model, but not by more than a DCHISQ value of 1%. In that case, instead of reverting to PSF, the modeling code reverts to NONE.
The upshot of this issue is that when REX is better than PSF, but not by a sufficiently large margin, then the EXP or DEV model would be chosen instead of the PSF model.
This issue has been mostly addressed by looking at the DCHISQ values in the Tractor catalogs, determining which sources are affected, and patching the Tractor files with the PSF fit values, which are then propagated to all files that rely on the Tractor catalogs. Patching the catalogs in this manner has the inconsistency problem that the pipeline would have already subtracted the EXP or DEV model for a patched source during extraction, affecting the fitting of other sources in the image.
- Quantities that should be unaffected:
- RA, DEC, fluxes and their IVARs; TYPE, BX, BY, BRICK_PRIMARY.
- Quantities that could be inconsistent (if the PSF-fit centroid was in a different brick pixel than the DEV- or EXP-fit centroid):
- ANYMASK, ALLMASK, MASKBITS, BRIGHTBLOB, NOBS.
- Quantities that will be inconsistent as they will still use the DEV or EXP model shape:
- FIBERFLUX, FRACIN, FRACFLUX, FRACMASKED, RCHISQ and the WISE forced photometry.
DR8 was processed with several different code versions. Morphological models were only patched for a subset of the files that were processed with version DR8.2.1 of the code. These files can be identified by the header card: PATCHED (= integer number of sources patched).
The cores of very bright stars are often highly saturated. This produces negative fluxes in model fits, which causes very bright stars to be dropped completely during model selection. An example is Tania Australis (V~3, RA~213.88°, Dec~19.18°).
The fix, which will be included in DR9, is to force all reference stars (such as Gaia stars) to be retained during model fitting.
Bricks that didn't finish processing, or that only partially completed because we "bailed out" of source-fitting can fail after the coadded images have been produced but before sources have been extracted.
For such bricks, we remove any images from the coadd/ directory so that the pixel-level files are consistent with the catalog-level files. This means that there are a small number of bricks in DR8 which have images loaded in the viewer but that do not have corresponding files, coadd files or Tractor catalogs.
Sky levels at Declinations south of -30° were computed using CCD zero-points of 0 mag rather than the correct zero-points. This produced incorrect values for the following quantities in the annotated ccds file:
These values, which have been corrected for DR9, need to be updated for DR8 as in this code snippet. In addition, the units stated for these 4 quantities in the DR8 documentation were originally incorrect. These quantities are all in nanomaggies, not ADU.