Known Issues and Workarounds

DR3.1 Update to sweeps and Tractor Files


During final processing checks on DR3 it was noticed that 36 bricks had been corrupted in a read/write error. The corrupted files corresponding to these 36 bricks were:

tractor-1763p240.fits tractor-1821p295.fits tractor-1888p270.fits tractor-1916p295.fits
tractor-1934p255.fits tractor-1942p130.fits tractor-1942p255.fits tractor-1942p320.fits
tractor-1951p237.fits tractor-1896p300.fits tractor-1985p267.fits tractor-1986p237.fits
tractor-1986p272.fits tractor-1965p300.fits tractor-2030p270.fits tractor-2040p287.fits
tractor-2049p237.fits tractor-2054p187.fits tractor-2063p300.fits tractor-2087p305.fits
tractor-2087p272.fits tractor-2109p317.fits tractor-2115p317.fits tractor-2116p302.fits
tractor-2117p307.fits tractor-2176p097.fits tractor-2204p252.fits tractor-2224p285.fits
tractor-2254p290.fits tractor-2261p257.fits tractor-2290p292.fits tractor-2295p320.fits
tractor-2310p320.fits tractor-2318p302.fits tractor-2334p200.fits tractor-2496p257.fits
These files have been reprocessed and are included in a "DR3.1" directory structure, which is now the default to which DR3 directories point. The one exception to the overall directory structure is the sweeps files, for which the uncorrupted versions are available at legacysurvey/dr3/sweep/3.1 (instead of at legacysurvey/dr3/sweep/3.0). Note that users who want access to the original version of the legacysurvey directory structure can still find it at legacysurvey/dr3.0.


Tractor-on-Bricks is a slightly different implementation of the Tractor source extraction code in which a fraction of bricks that don't touch are run through Tractor, and then sources that are BRICK_PRIMARY are fixed so that additional bricks run through Tractor don't produce duplicate "edge" sources where bricks overlap. This has the advantage of producing rigorous brick boundaries, but the disadvantage of slowing down processing through being harder to parallelize. In DR3, tractor-on-bricks was enforced until June 6.
Bricks with timestamps before June 6 were processed slightly differently to subsequent bricks. All of these "pre-6/6" bricks are within a region bounded by Right Ascensions of 315 to 45 degrees and Declinations of -5 to +7 degrees (a region in which the eBOSS survey is targeting Emission Line Galaxies). Not every brick in that region was run with Tractor-on-Bricks but no brick outside of that region was run with Tractor-on-Bricks.

Blacklisted Frames

We attempt to process all available DECam imaging, regardless of whether such data was specifically taken as part of the legacy survey. Some imaging surveys revisit certain areas of the sky many times, and processing the coaddition of such data can severely slow down the Tractor pipeline. Because of this, some proposal IDs are "blacklisted" to prevent Tractor from dwelling on those regions. The upshot of this is that if you're looking for coadded DECam imaging from a specific set of surveys in a certain region of the sky, it may not be available to the full depth that you expect.
Which proposal IDs were considered in which brick can be obtained from the ccds files that are documented on the catalogs page. Unfortunately, however, the final dr3e code used to generate DR3 catalogs has a blacklist cut that does not match the blacklist_ok column in the DR3 ccds-annotated files. This implies that the ccds-annotated files were generated with an earlier version of the code (seemingly the dr3c version). The easiest way to handle these discrepancies is to use the newer Data Release 5 of the Legacy Surveys.

NANs in Tractor Catalogs

Occasionally, measurements of an object in a particular band will have NaNs in all of the following columns: DECAM_RCHI2, DECAM_FRACFLUX and DECAM_FRACMASKED. No other columns ever have NaNs. The vast majority of NaNs correspond to measurements which have DECAM_NOBS of 0. A small fraction of the NaNs (~1%) correspond to measurements with non-zero DECAM_NOBS. In these cases, the relevant band's flux is identically zero, suggesting that no such measurement was actually performed.
We suggest that users simply ignore all measurements which correspond to NaN values of DECAM_RCHI2/DECAM_FRACFLUX/DECAM_FRACMASKED, although we have not determined whether there are unintended negative consequences of this prescription. ~7% of all Tractor catalog rows contain a nonzero number of NaNs. The root cause of these NaNs is being investigated.