Algorithms: Target Selection
Detailed descriptions of the selection algorithms for the different
categories of SDSS targets are provided in the series of papers
noted below under Target Selection References.
Here we provide short summaries of the various target selection
algorithms. Also compare the target selection
quality page.
In the SDSS imaging data output tsObj files, the
result of target selection for each object is recorded in the 32-bit
primTarget flag, as defined in
Table 27 of Stoughton et al. (2002). For details, see the Target Selection References
Note the following subtleties:
- An object can be targeted simultaneously by more than one
algorithm.
- The photometric catalogs contain a target selection flag for
every single object,
- but not all objects which are
flagged as a spectroscopic target will actually be observed with the
spectrograph. The assignment of spectrograph fibers to targets
from the photometry catalogs is called tiling.
- Perhaps most importantly, the target selection flags used in
order to create the spectroscopic plates were (necessarily) based
on an earlier processing of the data. Thus, objects that were
targets in the original rerun may not be targets now, and vice
versa. For the Main Galaxy Sample, this amounts to changes in the r
band flux limit; for Quasars it means wholesale changes in the
algorithms; for Luminous Red Galaxies, it means that the effective
color selection differs from place to place on the sky.
The following samples are targetted:
- Main Galaxy Sample
- Luminous Red Galaxies (LRG)
- Quasars
- Stars
- ROSAT All-Sky Survey sources
- Serendipity targets
Main Galaxy Sample
The main galaxy sample target selection algorithm is detailed in
Strauss et al. (2002) [link to AJ site, subscription required;
also at astro-ph/0206225] and
is summarized in this schematic
flowchart.
Galaxy targets are selected starting from objects which are detected
in the r band (i.e. those objects which are more than 5σ
above sky after smoothing with a PSF filter).
The photometry is corrected for Galactic extinction using the
reddening maps of
Schlegel, Finkbeiner, and Davis (1998).
Galaxies targeted from DR2 and later data are separated from stars using the following cut on the difference between the r-band
PSF
and
cmodel
magnitudes:
rPSF - rcmodel >= 0.24
Note that this cut is more conservative for galaxies than the
star-galaxy separation cut used by Photo (galaxies targeted off
DR1 data used a slightly different cut, rPSF - rmodel >= 0.3).
Potential targets are then rejected if they have been flagged by Photo
as SATURATED, BRIGHT, or (BLENDED and not NODEBLEND) - see
the descriptions of flags.
The
Petrosian
magnitude limit rP = 17.77 is then applied, which
results in a main galaxy sample surface density of about 90 per deg2.
A number of surface brightness cuts are then applied, based on
mu50, the mean surface brightness within the Petrosian half-light
radius petroR50. The most significant cut is mu50
<= 23.0 mag arcsec-2 in r, which already
includes 99% of the galaxies brighter than the Petrosian magnitude
limit. At surface brightnesses in the range 23.0 <=
mu50 <= 24.5 mag arcsec-2, several other
criteria are applied in order to reject most spurious targets, as
shown in the flowchart. Please see the
detailed discussion of these surface brightness cuts, including
consideration of selection effects, in
Section 4.4 of Strauss et al. (2002).
Finally, in order to reject very bright objects which will cause
contamination of the spectra of adjacent fibers and/or saturation of
the spectroscopic CCDs, objects are rejected if they have (1) fiber magnitudes
brighter than 15.0 in g or r, or 14.5 in
i; or (2) Petrosian magnitude rP <
15.0 and Petrosian half-light radius petroR50 < 2 arcsec. Main
galaxy targets satisfying all of the above criteria have the GALAXY
bit set in their primTarget flag. Among those, the ones with
mu50 >= 23.0 mag arcsec-2 have the GALAXY_BIG
bit set. Galaxy targets who fail all the surface brightness selection
limits but have r band fiber magnitudes brighter than 19 are accepted
anyway (since they are likely to yield a good spectrum) and have the
GALAXY_BRIGHT_CORE bit set.
Luminous Red Galaxies (LRG)
SDSS luminous red galaxies (LRGs) are selected on the basis of
color and magnitude to yield a sample of luminous intrinsically red
galaxies that extends fainter and farther than the SDSS main galaxy
sample. Please see
Eisenstein et al. (2001) for detailed discussions of sample
selection, efficiency, use, and caveats.
LRGs are selected using a variant of the photometric redshift technique
and are meant to comprise a uniform, approximately volume-limited sample of
objects with the reddest colors in the rest frame. The sample is
selected via cuts in the (g-r, r-i, r)
color-color-magnitude cube. Note that all colors are measured using
model magnitudes, and all quantities are
corrected for Galactic extinction following
Schlegel, Finkbeiner, and Davis (1998).
Objects must be detected by Photo as BINNED1 OR BINNED2 OR BINNED4
(see
flag descriptions)
in both r and i, but not necessarily in g,
and objects flagged by Photo as BRIGHT or SATURATED in g, r,
or i are excluded.
The galaxy model colors are rotated first to a basis that is aligned with the galaxy
locus in the (g-r, r-i) plane
according to:
c⊥ = (r-i) - (g-r)/4 - 0.18
c|| = 0.7(g-r) +
1.2[(r-i) - 0.18]
Please note that some earlier versions of SDSS
documentation (notably among them this page and the print version of
the EDR paper) have incorrect signs in the definition of cperp; the
above with just minus signs are now correct, as are those in the LRG
target selection paper (referenced
below).
Because the 4000 Angstrom break moves from the g band to the r band
at a redshift z ~ 0.4, two separate sets of selection criteria are needed
to target LRGs below and above that redshift:
Cut I for z <~ 0.4
- rPetro < 13.1 + c|| / 0.3
- rPetro < 19.2
- |c⊥| < 0.2
- mu50 < 24.2 mag arcsec-2
- rPSF - rmodel > 0.3
Cut II for z >~ 0.4
- rPetro < 19.5
- c⊥ > 0.45 - (g-r)/6
- g-r > 1.30 + 0.25(r-i)
- mu50 < 24.2 mag arcsec-2
- rPSF - rmodel > 0.5
Cut I selection results in an approximately volume-limited LRG sample to
z=0.38, with additional galaxies to z ~ 0.45.
Cut II selection adds yet more luminous red galaxies to z ~ 0.55.
The two cuts together result in about 12 LRG targets per deg2
that are not already in the main galaxy sample (about 10 in Cut I, 2 in Cut II).
In primTarget, GALAXY_RED is set if the LRG passes either Cut I or Cut II.
GALAXY_RED_II is set if the object passes Cut II but not Cut I.
However, neither of these flags is set if the LRG is brighter than the
main galaxy sample flux limit but failed to enter the main sample
(e.g., because of the main sample surface brightness cuts).
Thus LRG target selection never overrules main sample target selection
on bright objects.
Changes in LRG target selection for DR2 and later data
With the change in the model magnitude code between DR1 and DR2
data (see improvements to image
processing in DR2), the mean g-r and r-i
model colors of galaxies have shifted by about 0.005 magnitudes.
Because the LRG is very sensitive to color, this would have increased
the number density of targets by about 10%. Instead, we shifted the
LRG color cuts to compensate; in addition, improved star-galaxy
separation allows tighter cuts on the model-PSF quantity
by which stars are rejected. Here we give the updated equations for
Cut I applied to data reduced with the DR2 version of the photometric
pipeline (5_4; criteria not listed here are taken over unchanged from
above):
The definition of the new color basis changes as follows:
-
- c⊥ = (r-i) - (g-r)/4 - 0.177
- c|| = 0.7(g-r) +
1.2[(r-i) - 0.177]
Cut I for z <~ 0.4 becomes
- rPetro < 13.116 + c|| / 0.3
- rPSF - rmodel >= 0.24
Cut II for z >~ 0.4 becomes
- c⊥ > 0.449 - (g-r)/6
- g-r > 1.296 + 0.25 (r-i)
- rPSF - rmodel >= 0.4
This new version of LRG target selection is applied to the
best region of sky reduced with the latest version of the
imaging pipeline. It is of course not applied retroactively to the
target version of the sky, which used older versions of
the pipeline.
Due to other subtle differences in the photometric pipeline and the
calibration, these changes will not exactly reproduce the selection
criteria actually used when spectroscopy was carried out. Indeed,
defining an LRG sample based on the best reductions will
result in large spectroscopic incompleteness because so many objects
are close to the boundaries. Instead, one should use the
target photometry and adjust the calibrations of that
relative to the best calibration. Of course, if one is
interested in photometric properties of single objects, then we
recommend the best photometry.
Quasars
The final adopted SDSS quasar target selection algorithm is
described in
Richards et al. (2002). However, it should be noted that the
implementation of this algorithm came after the last
date of DR1 spectroscopy. Thus this paper does not technically
describe the DR1 quasar sample and the DR1 quasar sample is
not intended to be used for statistical purposes (but see
below). Interested parties are instead encouraged to use the catalog
of DR1 quasars that is being prepared by Schneider et al (2003, in
prep.), which will include an indication of which quasars were also
selected by the Richards et al. (2002) algorithm. At some later time,
we will also perform an analysis of those objects selected by the new
algorithm but for which we do not currently have spectroscopy and will
produce a new sample that is suitable for statistical analysis.
Though the DR1 quasars were not technically selected with the
Richards et al. (2002) algorithm, the algorithms used since the EDR
are quite similar to this algorithm and this paper suffices to
describe the general considerations that were made in selecting
quasars. Thus it is worth describing the algorithm in more detail.
The quasar target selection algorithms are summarized in this
schematic flowchart. Because the
quasar selection cuts are fairly numerous and detailed, the reader is
strongly recommended to refer to
Richards et al. (2002) (link to AJ paper; subscription required)
for the full discussion of the sample selection criteria,
completeness, target efficiency, and caveats.
The quasar target selection algorithm primarily identifies quasars as outliers
from the stellar locus, modelled following
Newberg & Yanny (1997)
as elongated tubes in the
(u-g, g-r, r-i)
(denoted ugri) and
(g-r, r-i, i-z)
(denoted griz)
color cubes.
In addition, targets are also selected by matches to the FIRST
catalog of radio sources
(Becker, White, & Helfand 1995).
All magnitudes and colors are measured using
PSF magnitudes, and all quantities are
corrected for Galactic extinction following
Schlegel, Finkbeiner, and Davis (1998).
Objects flagged by Photo as having either "fatal" errors (primarily those
flagged BRIGHT, SATURATED, EDGE, or BLENDED;
see
flag descriptions) or "nonfatal" errors
(primarily related to deblending or interpolation problems) are rejected
from the color selection, but only objects with fatal errors are rejected
from the FIRST radio selection. See
Section 3.2 of Richards et al. (2002) for the full details.
Objects are also rejected (from the color selection, but not the radio selection)
if they lie in any of 3 color-defined exclusion regions which are dominated
by white dwarfs, A stars, and M star+white dwarf pairs; see
Section 3.5.1 of Richards et al. (2002) for the specific exclusion region color boundaries.
Such objects are flagged as QSO_REJECT. Quasar targets are further restricted to
objects with iPSF > 15.0 in order to exclude bright objects
which will cause contamination of the spectra from adjacent fibers.
Objects which pass the above tests are then selected to be quasar targets
if they lie more than 4σ from either the ugri or griz
stellar locus. The detailed specification of the stellar loci and of the
outlier rejection algorithm are provided in Appendices
A and
B of Richards et al. (2002). These color-selected quasar targets are divided
into main (or low-redshift) and high-redshift samples, as follows:
Main Quasar Sample (QSO_CAP, QSO_SKIRT)
These are outliers from the ugri stellar locus and are selected in
the magnitude range 15.0 < iPSF < 19.1.
Both point sources and extended objects are included, except that
extended objects must have colors that are far from the colors
of the main galaxy distribution and that are
consistent with the colors of AGNs; these additional color cuts for
extended objects are specified in
Section 3.4.4 of Richards et al. (2002).
Even if an object is not a ugri stellar locus outlier, it may be selected as
a main quasar sample target if it lies in either of these 2 "inclusion" regions:
(1) "mid-z", used to select 2.5 < z < 3 quasars whose colors cross
the stellar locus in SDSS color space; and (2) "UVX", used to duplicate
selection of z <= 2.2 UV-excess quasars in previous surveys.
These inclusion boxes are specified in
Section 3.5.2 of Richards et al. (2002).
Note that the QSO_CAP and QSO_SKIRT distinction is kept for historical
reasons (as some data that are already public use this notation)
and results from an original intent to use separate
selection criteria in regions of low ("cap") and high ("skirt")
stellar density. It turns out that the selection efficiency is
indistinguishable in the cap and skirt regions, so that the
target selection used is in fact identical in the 2 regions
(similarly for QSO_FIRST_CAP and QSO_FIRST_SKIRT, below).
High-Redshift Quasar Sample (QSO_HIZ)
These are outliers from the griz stellar locus and are selected in
the magnitude range 15.0 < iPSF < 20.2.
Only point sources are selected, as these quasars will lie at
redshifts above z~3.5 and are expected to be classified
as stellar at SDSS resolution. Also, to avoid contamination from
faint low-redshift quasars which are also griz stellar locus
outliers, blue objects are rejected according to eq. (1) in
Section 3.4.5 of Richards et al. (2002).
Moreover, several additional color cuts are used in order to
recover more high-redshift quasars than would be possible using only
griz stellar locus outliers.
So an object will be selected as a high-redshift quasar target if it
lies in any of these 3 "inclusion" regions:
(1) "gri high-z", for z >= 3.6 quasars;
(2) "riz high-z", for z >= 4.5 quasars; and
(3) "ugr red outlier", for z >= 3.0 quasars.
The specifics are given in eqs. (6-8) in
Section 3.5.2 of Richards et al. (2002).
FIRST Sources (QSO_FIRST_CAP, QSO_FIRST_SKIRT)
Irrespective of the various color selection criteria above,
SDSS stellar objects are selected as quasar targets if they have
15.0 < iPSF < 19.1 and are matched to within 2 arcsec of
a counterpart in the FIRST radio catalog.
Finally, those targets which otherwise meet the color selection
or radio selection criteria described above, but fail the cuts
on iPSF, will be flagged as QSO_MAG_OUTLIER
(also called QSO_FAINT).
Such objects may be of interest for follow-up studies,
but are not otherwise targeted for spectroscopy under routine
operations (unless another "good" quasar target flag is set).
Other Science Targets
A variety of other science targets are also selected; see also
Section 4.8.4 of Stoughton et al. (2002). With the exception of
brown dwarfs, these samples are not complete, but are
assigned to excess fibers left over after the main samples of
galaxies, LRGs, and quasars have been tiled.
Stars
A variety of stars are also targeted using color selection criteria,
as follows:
- blue horizontal-branch stars (STAR_BHB)
- both dwarf and giant carbon stars (STAR_CARBON)
- brown dwarfs (STAR_BROWN_DWARF) - this is the only tiled sample of stars
- low-luminosity subdwarfs (STAR_SUB_DWARF)
- cataclysmic variables (STAR_CATY_VAR)
- red dwarfs (STAR_RED_DWARF)
- hot white dwarfs (STAR_WHITE_DWARF)
- central stars of planetary nebulae (STAR_PN)
ROSAT Sources
SDSS objects are positionally matched against X-ray sources from
the ROSAT All-Sky Survey (RASS;
Voges et al. 1999), and SDSS objects
within the RASS error circles (commonly 10-20 arcsec) are targeted
using algorithms tuned to select likely optical
counterparts to the X-ray sources. Objects are targeted
which:
- are also radio sources (ROSAT_A)
- have SDSS colors of AGNs or quasars (ROSAT_B)
- fall in a broad intermediate category that includes stars
that are bright, moderately blue, or both (ROSAT_C)
- are otherwise bright enough for SDSS spectroscopy (ROSAT_D)
Objects are flagged ROSAT_E if they fall within the RASS
error circle but are either too faint or too bright for SDSS
spectroscopy.
Serendipity
This is an open category of targets whose selection
criteria may change as different regions of parameter space
are explored. These consist of:
- objects lying outside the stellar locus in color space
(SERENDIP_RED, SERENDIP_BLUE, SERENDIP_DISTANT)
- objects coincident with FIRST sources but fainter
than the equivalent in quasar target selection;
also not restricted to point sources (SERENDIP_FIRST)
- hand-selected targets (SERENDIP_MANUAL)
Target Selection References
-
Spectroscopic Target Selection in the Sloan Digital Sky Survey: The Main Galaxy Sample,
Strauss, M., et al. 2002, AJ, 124, 1810
-
Spectroscopic Target Selection for the Sloan Digital Sky Survey: The Luminous Red Galaxy Sample,
Eisenstein, D., et al. 2001, AJ, 122, 2267
-
Spectroscopic Target Selection in the Sloan Digital Sky Survey: The Quasar Sample,
Richards, G., et al. 2002, 123, 2945
-
Sloan Digital Sky Survey: Early Data Release, Stoughton, C., et al. 2002, AJ, 123, 485
Last modified: Wed Jun 9 16:13:43 CDT 2004
|