Extended sources were identified and separated from point sources using a "decision tree" method. The method uses a variety of source attribute measures to separate resolved from the unresolved, including size, central surface brightness and color. In all, we use about 10 different measures (or dimensions), whose values are correlated to one degree or another. Read more about this parameter set in IV.5.a.vi.
Two different decision tree "scores" are generated: (1) G score and the (2) E score. The "G score" is geared towards recognizing galaxies. Not only does it include the standard star-galaxy parameters (e.g., shape, size & SB, ... etc) but it also includes the "color attribute". The "E score" is a simpler version of the G-score: it has only three parameters that are tested in the decision trees, including surface brightness, shape and the double/triple star discriminator. It does not include a color attribute. In this sense, it is less biased than the G-score. It was conceived to capture Milky Way objects that might have unusually blue colors, or shapes distinctly different from galaxies.
It should be understood that: (1) the E score is less reliable than the G score, and (2) although it may be less biased to color, the E score is not necessarily better at finding Milky Way objects (the G-score is more than happy to find HII regions, nebula, YSOs, etc).
These are employed as the final arbiter for star-galaxy separation. It turns out that a value of around 1.4 gives a satisfactory separation, while maintaining satisfactory completeness. Summary: extended objects have scores between 1 and 1.4, while point-like objects have values between 1.4 and 2.0.
iii. Lower Reliability for Sources at Low Galactic Latitudes
The parameters of the XSC were tuned to meet
as closely as possible
the
Level 1 Science requirements of
99% reliability for "G" sources above point source
densities representative of galactic latitude glat ~ 20° and
80% reliability for "G" sources at glat ~ 10°.
Reliability is defined here as the percentage of sources which are truly
extended (e.g., galaxies and nebulae), with multiple point sources,
artifacts around bright stars and meteor trails counted as "unreliable"
extended sources.
For higher source densities, the emphasis shifted to completeness, at the
expense of reliability. Hence the reliability of "G" sources may be as low
as 50% at the highest source densities allowed in the catalog
(cf. VI.4).
No requirements were placed on the reliability of Galactic Plane sources, and hence a
higher percentage of them are composed of multiple point sources.
iv. Incompleteness Due to Source Confusion
The presence of foreground Galactic stars limits galaxy detection both in the
sense that the background noise is enhanced (i.e.,
confusion noise) and the ability to
distinguish extended emission from the
stellar (point-like) emission is diminished. The rate at which these effects
become increasingly important is dictated by the
stellar number density,
roughly matching the exponential distribution of stars
along the the Galactic latitude axis.
The drop in completeness with the source confusion can be clearly seen in
the
allsky "movie" maps and more quantitatively in
the
XSC source count "movie" map. Note the large gaps in completeness near the Galactic Center,
and note the "reddening" of the XSC sources as seen within
(or projected against) the Milky Way.
v. Unreliability and Incompleteness Due to Bright Stars
The problems of bright stars mentioned above (primarily scattered light and
diffraction spikes) create vast numbers of spurious extended source detections.
Hence it was necessary to reject more area around bright stars for the XSC
in order to meet reliability requirements. About 1 to 2% of the entire sky
is excluded from the XSC due to bright star confusion (cf.
IV.5).
When in doubt about a given source, consulting the Atlas Image will usually
immediately reveal whether a source is an artifact or not.
The "visual classification" flag,
vc, is totally separate from
any aspect of the Catalog Generation,
and can be used as a means of quickly assessing the reliability of any
given selection criteria imposed on the catalog.
The visual classification flag was assigned on the basis of
examining the image of a source, by eye, comparing across bands and
(when needed) with the Digital Sky Survey.
The catalog contains
316107 sources with "galaxy" or "extended" visual classifications;
7383 with "artifact" or "unreliable" visual classifications;
and 35270 with "unknown" visual classifications (that is, the source
defies classification, usually because it is too faint in surface
brightnesss).
vi. Artifacts
Visual inspection of the XSC reveals a small fraction (less than 1% of the total) of
sources that may be classified as image artifact -- usually associated with
bright stars. These "sources" are often possessed with
unusual photometry or size attributes. Be wary of sources with
unusual colors or other "outlier" attributes. In addition to bright stars, other kinds of
artifact inducing phenomena include "airglow" (particularly at H-band), transients (e.g., meteor
streaks) and image edges. We have made every effort to minimize these contaminants to the XSC.
TABLE of Known or Suspected Artifacts in the XSC
vii. False Extended Objects
With projection effects, close groupings of stars come in every permutation imaginable.
These "double" and "triple" stars represent a small contaminant to the XSC,
mostly confined to the Plane of the Galaxy.
Since these sources are stellar in nature, they tend to have blue colors, J-Ks < 1 mag,
with respect to galaxies. Refer to the XSC FAQ:
Reliability: What is a Star Doing in the XSC?.
TABLE of Suspected Stars and Other Non-Extended Objects in the XSC
viii. Special Sources: the cc_flg flag
The cc_flag is used to highlight XSC entries that
have some probability of being artifacts or contaminated
by nearby large galaxies, or that are
Large Galaxy
entries that were processed specially.
Sources flagged as artifact (cc_flg="a")
include those corrupted by a bright star or those that are
outright false detections of filter glints or ghost
images produced by bright stars. These sources were identified
as such during
visual inspection operations.
The following table summarizes the possible values in the cc_flg, and
shows the number of sources in the All Sky
XSC having each cc_flg value.
ix. Duplicate Sources
Duplicate sources are very rare, but do live within the XSC. The (usual) phenomenon arises
from scan-to-scan overlap observations (some 10-15% of the scan), where the same source in two
different scans may have slightly (~few arcsec) different centroids. The differing centroids
are usually associated with low S/N sources (or nuclei), where noise fluctuations may induce
the difference. Another source of mis-measurement of the centroid is the scan edge itself,
which might (under certain circumstances) disrupt the source characterization process.
Be wary of any two XSC sources that are within a few arcsec of each other in coordinate space.
Refer to the XSC FAQ:
Duplicity: Why Do Some Galaxies Appear to be Duplicates".
x. Null Photometry and/or Attributes
"Null" values arise from either (1) corruptions in the source characterization process, or
(2) Signal-to-Noise limitations. The source characterization process is corrupted (or
interrupted) by a variety of circumstances, including the deleterious presence of bright star artifacts,
scan edges, high-frequency background gradients ("airglow"), and source confusion.
Faint sources may (for example) prove to be too formless to adequately measure the isophotal shape
or flux. Consequently you will often find "null" attributes for sources in
confused regions and for
faint (low S/N) sources. "Null" photometry (or corrupted photometry) may result
in bizarre, unphysical associations (e.g., strange colors).
Refer to the XSC FAQ:
Why Do Some Galaxies Appear to Have Unphysical Colors or Brightnesses?.
There is another class of XSC
object that is rife with "null" attributes. These are "sources" associated with
very large galaxies, and are believed to be (for the most part) unreliable "junk"
spawned by the parent galaxy. They may be real (e.g., H II regions), or they may be bogus
(e.g., noise bump on top of the large galaxy), but either way their source characterization
is seriously compromised by the confused environment. For this reason, these sources have
been identified and their photometry and source attributes artificially "nulled". The
cc_flg value is set to "z" (see table above).
The only
useful information in the XSC is the coordinate position of the object itself.
Read more about it in:
"Sources in close proximity to large galaxies".
xi. Mis-positioned Galaxies & Foreground Contamination
The close presence of foreground stars to an XSC galaxy (i.e., confusion) will on occasion
induce (1) poor central positions (i.e., astrometry) and/or (2) poor integrated fluxes
(e.g., see below "Wacky Measurements"). The user should beware of stellar confusion.
TABLE of source with poor positions due to stellar confusion
xii. Wacky Measurements
In the same vein as corrupted photometry and their resultant strange colors,
the occasional source characterization mis-measurement may result in totally unphysical
attributes, such as angular size. These measurement gaffes typically arise from
image problems (e.g., capricious background variations inherent to ground-based observations)
or disorder from source confusion. (See above discussion)
The 2MASS survey design did not cater to the needs of large angular extent objects.
Not only are these objects clipped by the edges of scans, but the background removal process
itself is compromised by the small size of the survey images with respect to nearby (zero redshift) galaxies.
The
Large Galaxy Atlas (LGA) was created to rectify this problem. The XSC now includes
the ~500 largest galaxies in sky. The LGA is a still a work in progress, however. A number of
moderate to large (non-LGA) galaxies in the XSC are clipped by scan edges, due to the unfortunate
proximity within a survey scan. These galaxies can be expected to have underestimated integrated
fluxes, as well as compromised large-scale attributes (e.g., isophotal radii). In time the
LGA will include these XSC galaxies, and the fluxes and source characterization will be recovered.
Here is an example of an XSC galaxy in close proximity to a scan edge:
xiv. Pieces of Large Galaxies or Clusters
A very small fraction of the XSC sources are pieces of larger objects, including galaxies and clusters.
xv. Little Big Man Galaxies
The
Large Galaxy Atlas is, as the name suggests, comprised of large angular-extent galaxies.
It was created to fill a large gap in the XSC where these sources were either unmeasureable
or mis-measured in the 2MASS pipeline. However, due to rather obtuse circumstances
(usually related to the author's interest in these special galaxies), a handful
of small galaxies have found their way into the LGA. They were created and processed using
the LGA pipeline. As such, their photometry and source characterization is perfectly
valid. Their only outstanding quality is that they are much smaller than nearby galaxies,
and in fact they are small in comparison to typical XSC galaxies. An example is
MRK 897.
Refer to the XSC FAQ:
Why Are Some "Large" Galaxies in Fact Barely Resolved?"
xvi. Half-Light Radii and Diminishing Returns Near the Resolving Limit of 2MASS
The shape and light concentration are difficult to determine for small galaxies,
whose profiles are significantly distorted by the point spread function (PSF). For galaxies
smaller than 10 arcsec (representing most of the XSC), the half-light radii
and concentration indexes will probably be overestimated due to the rounding and
extending effects of the PSF. These attributes should be used with caution.
See also
Half-Light "Effective" Aperture and
Concentration Index (IV.5.e) and
Axis Ratio and Small Radii.
xvii. Other Miscellaneous Notes
The only way to properly recover the flux of these complex objects is to treat
them with an optimized pipeline (such as the Large Galaxy Atlas pipeline). This in fact
has been carried out for the
Tarantula Nebula in the LMC.
Atmospheric seeing variations cause the observed PSF to change,
and the scan processing pipeline
attempts to follow such PSF changes in order to properly discriminate
extended sources from point sources. However, it is not possible to follow
rapid seeing changes, which usually results in some point sources falsely
identified as extended sources.
Extended sources are sensitive to a wider spectrum of noise sources than
point sources, which are affected only by high spatial frequency noise.
Some of the known noise sources are mentioned below.
The extended source background-removal algorithm removes
any background variation at J and Ks such that the residual noise
in the Images is usually consistent with the measurement error. However,
atmospheric OH airglow emission variations
contribute extra noise in the H-band roughly equal to the measurement
error.
The H-band photometric error due to airglow noise varies strongly with time
and spatial position and with the total brightness and size of an extended
source. A statistical analysis of galaxies with H = 13.8 mag shows that
about 25% of all sources have a measurable increased uncertainty which is
correlated with the measured
In the First Incremental 2MASS Data Release, scans that have large
measured background sigmas were not
included. For this All Sky Release, we have refined our airglow
diagnostic to only reject scans that have measured noise that is
significantly greater than that predicted by the measured H background.
This should allow better rejection of scans truly contaminated by
background structure due to airglow.
Some Tiles undoubtedly still contain airglow structure that is not
extensive enough to trigger our thresholds for Tile rejection. Therefore,
users should still be aware that H band fluxes for a significant number of
sources will have a higher photometric error than the quoted error, which
reflects only the Poisson noise in the background.
The best estimate we can make at this time for the magnitude of the excess
noise comes from a statistical analysis of sources from Atlas Images with a
residual background No correction of the photometric uncertainties has been made for this
statistical analysis result.
Electronic noise with spatial periods of 50-75´´ is sometimes
present in the
Atlas Images. Preliminary analysis shows that the noise can sometimes
resemble a square-wave distribution in the Images, producing a bias in
extended source photometry that is either full-amplitude positive or
negative. These biases can be as large as 17%, 7% and 11% at J, H and
Ks, respectively, for
galaxies with Poisson errors of less than 10%, for perhaps ~1% of all
galaxies.
TABLE of interesting XSC galaxies and other notable sources
EXAMPLEs
2MASX01581860-0704402: this is an "edge flare" created by a moderately bright star located just off the eastern
edge of the scan
EXAMPLEs
EXAMPLEs
EXAMPLE
EXAMPLE
EXAMPLEs
They also arise from
software design flaws. For example, the early 2MASS pipeline data (see the incremental
releases) featured a set of objects, thought to be associated with large galaxies, but instead
were either off-center pieces of large galaxies or nearby stars:
"Special processing was used to select sources spatially coincident with large galaxies (but not too large; see I.6.c.ix above) and pass them into the Extended Source Database.
Usually sources so identified capture at least part of the flux of the large galaxies. However, some large galaxies have poorly determined positions, and as a result sources that
are not in fact part of these large galaxies were picked up by the same processing."
Now this problem has been, for the most part, fully rectified in the All Sky XSC.
In addition,
the largest galaxies are treated separately, the
Large Galaxy Atlas (LGA),
fully rectifying any problems associated with scan-to-scan
edge problems (see below for the exceptions to this rule).
There is still the occasional XSC source that has an anomalously
large image size (mimicing a large galaxy), but in fact is a small extended object. These sources
tend to be associated with NED galaxies that are "known" to be large in angular extent
(but not necessarily in the near-infrared!).
Here is an example:
EXAMPLE
xiii. Large Galaxies and Survey Scan Edges
EXAMPLEs
TABLE of Known Large Galaxy/Cluster Pieces
[Last Update: 2007 January 3; T. Jarrett, T. Chester, R. Cutri]
of the background-removed pixel
intensity distribution.
of just under 1.20 DN.
These sources have an
extra H photometric error equal to their Poisson uncertainty of ~0.10 mag,
making the total photometric error ~0.15 mag.
Previous page. Next page.
Return to Explanatory Supplement TOC Page.
