Known Issues

General Issues

The most up-to-date issues and in-progress fixes are available on the GitHub issues pages for the Legacy Surveys website or the legacypipe pipeline.

Issues with CCDs that may have affected the quality of DECam observations are recorded on the DECam CCDs page.

The DR9sv "Data Release"

DR9sv was a preliminary version of the Legacy Surveys imaging intended to possibly be used for targeting during the DESI Survey Validation (SV) phase. However, DR9sv was never used for DESI SV targeting. Although DR9sv remains publicly available at:

and at NERSC in the directory:

/global/cfs/cdirs/cosmo/data/legacysurvey/dr9sv

DR9sv has been supplanted by the full DR9 release. We recommend ignoring any data from DR9sv, and using DR9 instead.

Bricks that were processed using the Burst Buffer at NERSC

After noticing that some bricks that were run using the Cori Burst Buffer at NERSC were corrupted, we reprocessed all such reductions. A total of 1691 bricks in the southern region of the Legacy Surveys were discovered that displayed changes in some of their coadded image, invvar, or nexp files. These bricks were rerun, and assigned a RELEASE number of 9012 (instead of the RELEASE number of 9010 used for reductions in the rest of DR9 south of the Legacy Surveys).

A list of the affected bricks is available here.

"Blobmodel" images are not compressed

Blobmodel images were intended to be compressed, and are named with a .fits.fz suffix, but they are not actually compressed.

NOBS differs between the Tractor catalogs and random catalogs

Quantities named nobs_x (where x is any of the Legacy Surveys imaging bands) in the Tractor catalogs (and derived products such as the sweep files) are slightly different from the NOBS_X quantities in the random catalogs.

The Tractor catalogs nobs columns count all pixels overlapping each brick pixel, even if the pixel is masked. In contrast, the random catalogs are derived from the nexp coadded stacks, which only count pixels where the inverse variances are positive. This means, for instance, that the nexp stacks ignore both masked pixels and pixels that have zero weight passed forwards from the Community Pipeline.

The result of this discrepancy is that a few percent of objects in the Tractor catalogs have different values of nobs than would be expected when deriving NOBS from the same location in the nexp coadded stacks. Conveniently, the nobs values in the Tractor catalogs are (almost) always larger than what would be derived from the nexp coadded stacks. This means that constraints based on a minimum nexp value in a given filter will always result in reproducible survey geometry, at the expense of rejecting a small fraction (typically <1%) of sources.