Target Selection: Quality and Efficiency
The Sloan Digital Sky Survey uses its imaging data to select objects
for spectroscopic followup. There is a description of the target selection
algorithms in the EDR paper. Also compare the targetting algorithms page.
Broadly speaking, targetted objects fall into three categories:
Note: Just because an object is targetted does
not necessarily mean that it will be assigned a spectroscopic fiber.
The tiling algorithm (see also Blanton et
al. 2003) assigns targets to fibers; two fibers cannot be placed
more closely than 55". This means that roughly 10% of all galaxy
targets are not observed spectroscopically (somewhat less in the
overlaps between plates). The majority of close pairs of galaxies are
essentially at the same redshift, an approximation that works
reasonably well for, e.g., correlation function studies (cf., Zehavi
et al. 2002).
Tiled targets
The main galaxy sample
This is described in some detail in Strauss et al. (2002); a
briefer description is on the target selection algorithm
page. In brief, it includes all galaxies with Petrosian r
magnitude brighter than 17.77. Here and throughout target selection,
magnitude limits are based on quantities corrected for Galactic
reddening a la Schlegel, Finkbeiner, and Davis (1998). There is a
Petrosian half-light surface brightness cut of 24.5 magnitudes in one
square arcsec, which rejects less than 1% of all galaxies. There is
also a cut in 3" fiber magnitude at g=15, r=15, and i=14.5 (to prevent
saturation and cross-talk in the spectrographs). There are roughly 90
such targets per square degree.
Bright galaxy magnitude limits
The galaxy sample is limited at the bright end by the fiber
magnitude limits, to avoid saturation and excessive cross-talk in the
spectrographs. Similarly, the image deblending software had a
tendency to occasionally shred large, bright galaxies with
substructure, causing problems for galaxy target selection at the
bright end (this problem is essentially fixed in the version of the
pipeline that produced the "best" imaging reductions, but was not in
place when the targetting and spectroscopy was carried out. These two
effects together cause the spectroscopic sample to become noticeably
incomplete for galaxies brighter than r=14.5 or so. We do plan
eventually to supplement our galaxy catalog with redshifts from the
literature for those bright galaxies for which we do not have
redshifts.
Faint galaxy magnitude limit
The galaxy target selection is magnitude-limited to a Galactic
extinction-corrected Petrosian magnitude of r=17.77. However, this
limiting magnitude has varied through the survey, between 17.5 and
17.77 (it turns out that one needs to survey an enormous region of sky
to determine the number density of galaxies to a few percent
accuracy!). The information on what magnitude limit is used in what
region of sky is quite complex (as is the information on the exact
geometry of the sky coverage). The safest thing to do at this point
is to limit a galaxy spectroscopic sample for statistical studies to
r=17.5.
Galaxy target selection efficiency
The galaxy target selection is very efficient. Only one percent
of the objects targetted turn out not to be galaxies. This one
percent is made up of close pairs of stars, and occasional sprays of
scattered light. The signal-to-noise ratio in the spectra is such
that reliable redshifts are available for essentially all galaxies,
even those at quite low surface brightness levels.
The Luminous Red Galaxy Sample
This sample is described by Eisenstein et al. (2001); a briefer
description can be found on the target selection algorithm
page. This targets the most luminous red galaxies at each
redshift; for redshifts z>0.2, there are clean cuts in
color-magnitude space that effectively isolate objects with the
properties of Brightest Cluster Galaxies. There are two classes,
flagged with the target bits GALAXY_RED and GALAXY_RED_II,
respectively; they select a roughly volume-limited sample of objects
with 0.2<z<0.38 (down to r(Petrosian) = 19.2), and a
flux-limited sample (to r(Petrosian) = 19.5), which extends to z=0.55.
There are roughly 12 objects targetted per square degree.
The LRG sample comes in two parts: a roughly volume-limited
subsample to z=0.38, and a flux-limited sample that extends to z=0.55
or so. The spectra are of high enough quality that only 1-2% of the
objects targetted fail to have a redshift successfully measured. The
color cuts that define the LRG sample break down for redshifts below
0.2. That's OK, all such objects are bright enough to be included in
the main galaxy sample, but it is up to individual scientists to pull
such a sample out (i.e., based on luminosities and colors). There have
been small changes in the LRG
target selection criteria from DR1 to DR2/DR3.
The Quasar sample
This sample is described in Richards et al. (2002); a briefer
description can be found in the target selection algorithm
page. Quasar candidates are defined as objects with colors
distinct from those of ordinary stars (which form almost a
one-dimensional distribution in SDSS color space). Certain regions of
color space are explicitly excluded, as they are contaminated by rarer
types of stars: hot white dwarfs, A stars, and M dwarf/white dwarf
pairs. The algorithm does not explicitly require objects be
unresolved in the region of color space in which ultraviolet excess
objects lie (this allows resolved Seyfert galaxies to be targetted),
but redder objects (i.e., those with z>3, reddened by the Lyman
alpha forest) must be stellar.
Magnitude limits
The sample is
magnitude-limited to a PSF magnitude of i=19.1 in the ultraviolet
excess region of color space, and i=20.2 elsewhere. Roughly 18
objects per square degree are targetted; of order 2/3 of these are in
fact quasars.
In addition, unresolved objects brighter than
i=19.1 with counterparts in the FIRST radio survey are also
targetted. Very few of these are not already selected by the color
algorithm described above.
There is a bright limit of i=15 for similar reasons as
in the galaxy target selection
Quasar target selection changes
Quasar target selection has also evolved somewhat, due both to
changes in the targetting algorithm itself, and improvements in the
photometric pipeline; quasar target selection was the focus of a great
deal of fine-tuning during SDSS commissioning. None of the DR1 data
use the very latest version of the target selection algorithm, as
described in the Richards et al. (2002) paper. As a consequence, the
quasar targetting efficiency in the data themselves varies a fair
amount. With the current versions of the selection algorithm and the
imaging pipeline, the selection is clean: roughly 2/3 of the targets
are indeed quasars, and essentially all the contamination is there for
astrophysical reasons (principally compact star-forming galaxies, A
and F stars, white dwarfs, and late M stars). However, with old
versions of the imaging pipeline, stars tended to scatter out of the
stellar locus when the seeing changed rapidly, contaminating the
quasar selection and reducing the efficiency of the code. As a
consequence, there are plates, fortunately relatively few, with quasar
target selection efficiency as low as 30%.
Quasar selection completeness
The completeness of the quasar selection algorithm depends somewhat on
redshift. In particular, the completeness is low for 2.4<z<2.9,
where the quasar and stellar loci cross; it is similarly low at
redshifts around 3.5 and 4.5. These incompletenesses are more severe
in previous versions of the code than at present.
Quasar spectrum signal-to-noise The signal-to-noise
ratio of the quasar spectra is high, and the
redshift accuracy, based on extensive tests, is similarly high, of
order 99%. The exceptions are astrophysically interesting: BL
Lacertae objects (notoriously difficult to distinguish
spectroscopically from very hot stars) and extreme Broad Absorption Line
quasars (see the paper by Hall et al. 2002) are the most important.
Users of the spectra should be sure to pay close attention to the
flags indicating uncertainty in redshift determination.
Brown dwarf candidates These represent fewer than one object per
square degree; they are targetted by their very red colors in i-z, and
relatively blue colors in r-i.
Other targets
The other categories of scientific targets are assigned
spectroscopic fibers when fibers are available, following a cleanly
defined set of priorities. See the discussion of target priorities in the EDR
paper. Among these are:
Optical counterparts to ROSAT sources
These include quasars and AGN
(including some BL Lacs), some cataclysmic variables, and other
unusual classes of objects.
Optical counterparts to FIRST radio
sources
These go fainter than the quasar sample and do not have the
restriction that the object be unresolved.
Stars chosen by their unusual colors
These usually extend to magnitude limits fainter than the quasar
sample. Among these are objects with colors of blue horizontal branch
stars and white dwarfs.
Again, these classes of objects are observed spectroscopically
only when fibers are available, and therefore are not complete on the
sky in any sense of the term.
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: Tue Mar 9 11:37:05 CST 2004
|