The BOSS Data Reduction Pipeline (idlspec2d) produces a single best fit redshift and classification for each visit or coadd spectrum. While these are generally highly reliable, the parameters (like redshift) and spectral classification can sometimes be adversely impacted by e.g. spectral artifacts, noise characteristics, incomplete model template sets, etc. Various efforts have been made over the generations of SDSS to provide refined and unified redshift estimates. These have generally been provided in the form of Value Added Catalogs. Here we will discuss the new DR20 effort, but previous versions, such as the DR19 quasar spectral property catalog and the DR16 Quasar Catalog (DR16Q) are also available separately.
We have gathered visual inspections for a substantial subset (9%) of the ~500k DR20 ‘allepoch’ coadded BOSS spectra. A team of more than 20 consortium members have visually inspected a sample of the DR20 spectra, with an emphasis on a subset of spectra that were expected to have the highest failure rates. We supplemented this with previous results from other visual inspection efforts (e.g. inspections from SDSS DR16Q). These results are compiled into SDSS-V VAC #37, which provides a row-matched supplement to the DR20 allepoch spAll file. For each entry, sources are flagged if they have been visually inspected and are supplied with their redshift from the visual inspection, the redshift confidence, and the visual classification. In addition, using the visually inspected spectra as a training sample, and taking only features from the spAll catalog, we built a machine learning (ML) classifier to predict the reliability of the remaining un-inspected spectra. We judge a pipeline redshift (zpipe) to be ‘reliable’ when it is within 1% (specifically |zpipe − zVI|/(1 + zVI) < 0.01) of a secure visually inspected redshift (zVI). Those ML Classifier quality flags are included in SDSS-V VAC #37.
It is highly recommended that you use the redshift and spectral classification produced by this visual inspection effort over the raw pipeline results supplied in the BOSS pipeline spAll files.