BHM Caveats

Optical Spectra Caveats

As with any dataset, there are several caveats to watch out for when using the optical spectra released by SDSS. Some affect only a few spectra or a few data columns, while some others have wider impacts. Caveats are in many cases specific to the release and will be corrected in an upcoming release. This page contains a list of known caveats valid for SDSS Data Release 18 optical spectroscopic data. For caveats for SDSS-IV and earlier spectra see

PyXCSAO Outputs

PyXCSAO is a package used to measure radial velocities for stars by cross-correlating with stellar templates. Although pyXCSAO was run for all BOSS targets, the outputs are only valid for stellar targets.

Redshift Status

The quality flags for the redshift fitting procedure are stored in the ZWARNING bit mask. Most redshift warnings indicate a likely substantial problem with the data, or an indication that the best-fit classification or redshift is not reliable (due, e.g., to low S/N, or the unusual nature of the spectrum). An exception is MANY_OUTLIERS, which flags when many pixels are poorly explained in a statistical sense by the best-fit redshift model. This bit is typically set for very high signal-to-noise ratio stars (where errors are small, so χ2 is high), or galaxies with broad lines (the redshift fitting model includes only narrow lines); in such cases, the redshift is usually fine. About 2% of non-sky spectra have some warning set other than MANY_OUTLIERS. The redshifts of the remainder are virtually always correct. Many of the spectra flagged with problems also have correct redshifts and classifications, but we recommend care before using them. Note that the ZWARNING flag bits in BOSS are similar, but not identical, to those used in SDSS-I/II.

Night sky emission lines

The night sky emission lines at 5577Å, 6300Å, 6363Å (when there is auroral activity), and in the OH forest in the red can be very strong and leave significant residuals in the spectra whose amplitude is occasionally underestimated by the noise model. Be cautious about interpreting the reality of weak features close to these lines.

Sky Subtraction Bias

The sky spectrum estimates in BOSS (and in fact in SDSS) that are subtracted from each object are biased slightly low. This is due to the well-known bias associated with fitting an error-weighted model to data when the errors are estimated from the data itself (e.g. in the case of Poisson estimates of errors). These residuals can be detected by taking the average of the sky-subtracted sky fibers, which yield a slightly positive spectrum ranging from 7×10-20 erg/cm2/s/Å at around 8000 Angstroms to up to 10-18 erg/cm2/s/Å at the bluest and reddest end of the spectra.

Since DR13, the extraction algorithm was modified as described in Bautista et al. (2017). This new algorithm uses a constant weighing scheme, reducing considerably the sky subtraction bias. As a consequence, the extraction is slightly less optimal, yielding 5 to 10% larger errors in flux estimates per pixel.

Coadd errors are not perfect

The default BOSS/eBOSS spectra distributed in DR18 are coadded from several individual exposures. Each individual exposure has a slightly different relationship of pixel number to wavelength. Thus, errors in the coadded spectra have covariance between neighboring spectral bins; however, we do not calculate or track this covariance. As a result, there is a 10-20% “error-on-the-error” in the coadd noise model. If discrepancies at this level matter for your analysis, you should use the individual exposures, which have much better accuracy in their noise model (1-2%).

Galaxy velocity dispersion measurements

We do not recommend using SDSS velocity dispersion measurements for:

  • spectra with median per-pixel (S/N)2 < 10
  • velocity dispersion estimates smaller than about 70 km s-1 given the typical S/N and the instrumental resolution of the SDSS spectra

Also, note that the velocity dispersion measurements are not corrected to a standard aperture size. See the archived velocity dispersion algorithm for details.

Clipped Spectral Lines

Some emission lines intrinsic to the target are erroneously clipped because they were identified as cosmic rays. In SDSS-V, our clipping algorithm was modified to prevent the clipping of sky lines used for secondary wavelength calibration. This modification limits the clipping of spectral lines, but that does not preclude the possibility that some strong (typically narrow) emission lines might still be clipped. If an emission line is so bright that it is saturated in the individual 15-minute exposures of the spectrograph, it can suffer this effect. Unfortunately, such saturated pixels are not flagged as such, although usually, that region of the spectrum has an inverse variance equal to zero. Luckily, objects with such strong emission lines are very rare, but the user should be aware of the possibility of objects with extremely strong emission lines and unphysical or unusual line ratios.


A dominant source of bad classification/redshift fits to galaxy spectra is QSO templates with unphysical parameters, e.g. negative terms so that QSO emission lines “fit” galaxy absorption lines. To correct for this, galaxy spectra also have a ZWARNING_NOQSO mask, Z_NOQSO redshift, etc. which excludes QSO templates when performing classification/redshift fits. For studies with galaxy spectra, these *_NOQSO values should be used instead of the original ZWARNING mask, Z redshift, etc.

Bad CCD column results in bogus high-z quasars

Some unmasked intermittently bad CCD columns result in spurious identifications of spectra as z>5 quasars. These affect fibers 40, 59, and 60. Treat any redshifts this large with caution, and especially these fibers.

QSO pipeline redshifts

The QSO pipeline redshifts are generally reliable for >95% of the quasars, with a small fraction of catastrophic redshift failures and misclassifications. The reported uncertainty in the pipeline redshift for quasars, however, is underestimated due to intrinsic velocity shifts of quasar emission lines with respect to the systemic velocity. More details about the pipeline redshift uncertainties of SDSS quasars can be found in Wu & Shen (2022).

BOSS Flux Calibration

The flux calibration of individual exposures has an observing hour-angle and fiber dependence, especially below 4200 Angstroms. Analyses which rely upon accurate flux calibration of individual exposures should perform additional systematics cross-checks for the consistency between different exposures of the same object, and avoid data observed at large hour-angles. This issue may also affect SDSS spectra but that has not been confirmed.

BOSS Stellar Classifications

BOSS object classifications are primarily focused on the identification of galaxy vs. quasar vs. star. Although sub-classifications are provided, they are not optimized for accuracy. In particular, the CV star templates have more degrees of freedom than other stellar templates, which can result in unphysical solutions where negative PCA components of the CV templates can fit the absorption features of White Dwarfs. Fixing this has not been a high priority since the primary classification of “star” vs. “galaxy” or “qso” is still correct.

Artificial dichroic transitions at 6000 Å due to cross-talk from bright stars

A small number of spectra are affected by cross-talk from bright stars (generally spectrophotometric standards) in neighboring fibers. This is often manifested in a strong break feature at the dichroic transition around 6000 Å, resulting from different levels of cross-talk between the red and blue arms of the spectrograph. These effects appear to occur less frequently at later survey dates, which would be consistent with the improvements in the focus of the BOSS spectrograph cameras that have been achieved with routine operation. We intend to mitigate these effects in future BOSS data releases through improvements in the extraction codes and to flag any spectra that remain compromised. No masking of this effect is implemented for BOSS DR10 data, however, except to the extent that it triggers a ZWARNING bit in certain instances.

Back to Top