Because the SDSS spectra are obtained through 3-arcsecond fibers during
non-photometric observing conditions, special techniques must be employed
to spectrophotometrically calibrate the data.
On each spectroscopic plate, 16 objects are targeted as spectroscopic standards.
These objects are color-selected to be F8 subdwarfs, similar in spectral
type to the SDSS primary standard BD+17 4708.
The flux calibration of the spectra is handled by the Spectro2d pipeline.
It is performed separately for each of the 2 spectrographs, hence each half-plate has its own calibration.
At least 3 exposures are obtained of each plate, sometimes under very different
sky conditions, and occasionally on multiple nights. In order to effectively
reject bad pixels when the data is combined, the gross exposure-to-exposure
differences must first be removed. This is accomplished by the smear procedure which is described in detail separately.
The highest S/N exposure is used to derive the flux calibration. Spectro2d runs
a PCA algorithm on the 8 spectra of flux standards and generates an
eigenspectrum. This procedure eliminates bad pixels and any star spectra that
deviate too much from the others. The eigenspectrum is then ratioed to a model
The result is fit with a spline to produce the flux correction. Since the
red and blue halves of the spectra are imaged onto separate CCDs, separate
red and blue flux calibration vectors are produced. These will resemble
the throughput curves under photometric conditions.
Finally, the red and blue halves of each spectrum on each exposure are multiplied
by the appropriate flux calibration vector. The spectra are then combined
with bad pixel rejection and rebinned to a constant dispersion.
Comparisons of the calibrated spectra with the SDSS photometry allow us to
assess the spectrophotometric quality.