Algorithm Descriptions
DR2 Help
 Archive Intro
 Table Descriptions
 Schema Browser
 Introduction to SQL
 Form Query User Guide
 Query Limits
 How To
 SkyServer Sites
 SkyServer Traffic Page
 Web Browsers
 Contact Help Desk

Photometric Redshifts

There are no photometic redshifts available for data releases 2 through 4 (DR2-DR4). Starting with DR5, there are two versions of photometric redshift in the SDSS databases, in the Photoz and Photoz2 tables respectively. The algorithms for generating these are described below.

Photoz Table

This set of photometric redshift has been obtained with the template fitting method. Please also see this link for more detailed information about this method..

The template fitting approach simply compares the expected colors of a galaxy (derived from template spectral energy distributions) with those observed for an individual galaxy. The standard scenario for template fitting is to take a small number of spectral templates T (e.g., E, Sbc, Scd, and Irr galaxies) and choose the best fit by optimizing the likelihood of the fit as a function of redshift, type, and luminosity p(z, T, L). Variations on this approach have been developed in the last few decades, including ones that use a continuous distribution of spectral templates, enabling the error function in redshift and type to be well defined.

Since a representative set of photometrically calibrated spectra in the full wavelength range of the filters is not easy to obtain, we have used the empirical templates of Coleman Weedman and Wu extended with spectral synthesis models. These templates were adjusted to fit the calibrations (see Budavari et al. AJ 120 1588 (2000))

For more detailed information see Csabai et al. AJ 125 580 (2003) and references therein.

The table contains the estimated redshift, the best matching template's spectral class, K-corrections and absolute magnitudes. There are also some parameters of the chi-square fitting. Caveats: The quality of photometric redshift estimation of faint objects (or to be prcise with large photometric errors) is weak. The "quality", "zErr" and "tErr" values are just estimates, they are not always reliable. For this estimation we have used galaxy templates for all objects. Except for a few misidentified galaxies which were categorized as star in the photopipeline, the values fornon-galaxies shouldn't be used.

objIDbigint 8 Unique ID pointing to PhotoObj table
Estimated parameters:
zreal 4 Photometric redshift
zErrreal 4 Marginalized error of the photometric redshift
treal 4 Photometric SED type between 0 and 1
tErrreal 4 Marginalized error of the photometric type
dmodreal 4magDistance modulus for Omega_M = 0.3,
Omega_lambda = 0.7 cosmology
rest_ugreal 4magRest-frame u-g color
rest_grreal 4magRest-frame g-r color
rest_rireal 4magRest-frame r-i color
rest_izreal 4magRest-frame i-z color
kcorr_ureal 4magk-correction
kcorr_greal 4magk-correction
kcorr_rreal 4magk-correction
kcorr_ireal 4magk-correction
kcorr_zreal 4magk-correction
absMag_ureal 4magRest-frame u0 absolute magnitude
absMag_greal 4magRest-frame g0 absolute magnitude
absMag_rreal 4magRest-frame r0 absolute magnitude
absMag_ireal 4magRest-frame i0 absolute magnitude
absMag_zreal 4magRest-frame z0 absolute magnitude
Parameters of the chi-square fit
classint 4 Number describing the object type (galaxy = 1)
pIdint 4 Unique ID for photoz version
rankint 4 Rank of the photoz determination; default is 0
versionvarchar 6 Version of photoz code
chiSqreal 4 The chi^2 value for the fit
c_ttreal 4 tt-element of covariance matrix
c_tzreal 4 tz-element of covariance matrix
c_zzreal 4 zz-element of covariance matrix
fitRadiusint 4 pixels Radius of area used for covariance fit
fitThresholdreal 4 Probability threshold for .tting, peak normalized to 1
qualityint 4 Integer describing the quality (best:5, lowest 0)

Photoz2 Table

The photometric redshifts from the U. Chicago/Fermilab/NYU group (H. Oyaizu, C. Cunha, M. Lima, E. Sheldon, H. Lin, and J. Frieman) are calculated using a Neural Network method that is very similar in implementation to Collister and Lahav (2004, PASP, 116, 345), using a a 4:15:15:15:1 network. The photo-z training set consists of 140,000 spectroscopic redshifts and single-pass SDSS photometry measurements. These spectroscopic redshifts come primarily from the SDSS (110,000; including SDSS main, LRG, and southern survey samples), with the remainder from the deeper galaxy surveys CNOC2, CFRS, DEEP, DEEP2, GOODS/HDF-N, and 2SLAQ. Note that the training set includes independent, repeat SDSS photometric measurements of the same objects. The trained network is tested on a larger validation set consisting of 1,700,000 SDSS photometric measurements of objects for which spectroscopic redshifts are available.

Please also see this link for more detailed information, including quality plots.

Our data model is

objid -- 64 bit objid (join to main photoobjall.objid or specobjall.bestobjid)
photoz -- 32 bit float
photozerr -- 32 bit float
flag -- int

The photo-z errors are computed using the Nearest Neighbor Error estimate method (NNE; Oyaizu et al., in preparation). NNE is a training set based method that associates similar errors to objects with similar magnitudes, and is found to accurately predict the error when the training set is representative.

The photo-z "flag" values and their meanings are listed in the table below. We recommend using only flag=0 or flag=2 objects.

Value Description
0 Normal photo-z
1 At least one of the magnitudes is undetected. Undetected magnitudes are treated as having magnitudes of 99 but are still included in the training procedure.
2 Objects with r > 20. We find that the photo-z's are less reliable for such faint objects
3 Objects in which both flags 1 and 2 apply

The input catalog for the photo-z calculations were derived from the SDSS photo pipeline outputs with a few additional quantities calculated to improve the star galaxy separation. These include the PSF probability "objc_prob_psf" and the lensing smear polarizability "m_r_h" (Sheldon et al. 2004, AJ, 127, 2544). The probabilities were cut at a galaxy probability greater than 0.8 which is very stringent, and smear polarizability less than 0.8. Further cuts on magnitude were made, as given below. For the list of input runs and reruns used please see this file.

The cuts:

counts_model[2] != -9999 &&
objc_prob_psf >= 0.0 &&
objc_prob_psf < 0.2 &&
(m_r[1] > 0.0 && m_r[1] < 0.8) ||
(m_r[2] > 0.0 && m_r[2] < 0.8) ||
(m_r[3] > 0.0 && m_r[3] < 0.8) ||
) &&
cmodel_counts[2] < 22.0 &&
cmodel_counts[2] > 14.0 &&
counts_model[2] < 22.5 &&
counts_model[2] > 13.0