V. Catalog Generation


3. Source Selection

The source selection criteria for 2MASS Catalog Generation, described in the following sections, are designed to draw from the Working Databases lists of reliable sources with accurate photometry and positions. Because reliability and completeness trade directly, though, the criteria have been tuned so as not to compromise the completeness of the Catalogs.

a. Tile Edge Boundaries

Point sources are required to be >10´´ from Tile edges, and extended sources must be >15´´ from Tile edges to be candidates for passing from the Point and Extended Source Working Databases to the respective Catalogs. For this purpose, a Tile Edge is defined to fall along the great circles interpolated between the four Tile corners that are listed for each Tile in the Release in the Scan Information Table. These safety boundaries serve three purposes:

  1. The three arrays in each 2MASS Camera are aligned on the sky to within 1-2 pixels. Although small, the misalignment produces a 2´´-3´´ region of less than three-band coverage around the Tile borders within which a source may not be reliably detected in all survey bands. This can result in even a bright bandmerged source missing detections in one or more bands.


  2. The point and extended source photometry routines (cf. IV.4b and IV.5) may produce incorrect brightness estimates for sources that that fall close enough to a Tile edge that the source or sky apertures intersect the edges. Use of the safety boundaries omits these sources.


  3. The RA tracking stability of the 2MASS telescopes during the fast declination scanning (~1°/minute) is excellent. Figure 1 shows the RA difference (in arcseconds) between the astrometrically-reconstructed frame centers and the starting of-date RA for a typical 2MASS scan. The maximum deviation from the interpolation between the endpoints is < 2´´. Histograms of the maximum east and west RA deviation from a great circle path on the sky for all northern and southern observatory scans of Tiles included in final 2MASS processing are shown in Figures 2, 3, 4, 5, respectively. The most common maximum deviation is only 1.7´´ at Mt. Hopkins and 0.7´´ at Cerro Tololo. The small tails of the distribution indicate that rare excursions of up to 7´´-8´´ are possible, though, so the true Tile edges may lie slightly inside or outside the great circle interpolated edges. The safety boundaries will compensate for these cases.


Figure 1

Figure 2Figure 3Figure 4Figure 5

b. SNR and Detection Limits

Sources selected for the All-Sky Data Release source catalogs are required to satisfy the following band-detection and and signal to noise ratio (SNR) criteria:

For the PSC, the SNR threshold must be satisfied by either the SNR derived from the measurement uncertainty of the default magnitude (see IV.4d;   SNR = 1.0857 / [jhk]_cmsig) or the scan SNR ([jhk]_snr). The latter ensures that bright sources with poor quality photometric measurements are not inadvertently omitted from the Release.

For the XSC, the SNR threshold must be satisfied by either the 20 mag arcsec-2 isophotal fiducial elliptical aperture measurement uncertainty ([jhk]_msig_k20fe), or the 7´´ circular aperture measurement uncertainty ([jhk]_msig_7), where SNR = 1.0857 / mag. The circular aperture measurement SNR is included because isophotal magnitudes cannot always be reliably extracted for extended sources, because of contamination by foreground stars.

i. Reliability

The source detection thresholds used in 2MASS pipeline processing (cf. IV.4a) are set to SNR~3.5 to ensure that the completeness requirements of the Survey are met. The SNR thresholds given above are intended to limit the contents of the Catalogs to reliable sources, per the 2MASS reliability requirements. Reliability, in this context, refers to a detection that corresponds to a real source on the sky, but that may suffer from some flux-overestimation bias (see below).

The effect of the SNR thresholds are illustrated in Figures 6, 7 and 8. These Figures show J, H and Ks detection and measurements statistics, respectively, for all point sources in the WDB with b>+85°. The three panels in each figure show, counterclockwise starting from the upper left: a) the differential source count curves (dlogN/dM) curves; b) source photometric measurement uncertainty plotted versus the default magnitude; and c) histograms of sources as a function of photometric uncertainty. The horizontal green lines in the bottom two panels indicate the equivalent SNR=10, 7, and 3 levels. Two indicators of reliability for high latitude point sources are detection in more than one band (particularly for and H and Ks) and association with an optical source from the Tycho 2 or USNO-A catalog (cf. IV.4f). The red and blue curves in the upper left and lower right panels show the distributions for multi-band detections and sources with optical counterparts, respectively. The distribution of putatively reliable multi-band detected sources begins to fall-away from the distribution of all sources for SNR levels below 7; this is well above the nominal detection threshold of SNR~3.5 indicated by the peak in the uncertainty histogram near ~0.3 mag. The SNR~7 threshold occurs at a flux level brighter than the peak and roll-over in the dlogN/dM curves. Therefore, the SNR>7 thresholds do not compromise the completeness requirements for 2MASS. The vertical green lines in each figure show the magnitudes brighter than which the Level 1 Requirements specify the 2MASS Catalogs shall be >99% complete.

A powerful 2MASS source reliability indicator is multi-band detection. Most spurious detection triggers such as the bright star halos and diffraction spikes, meteor trails, cosmic rays, hot pixels, etc. produce "sources" that do not bandmerge (see IV.4e). Bright star persistence and dichroic glint detections do tend to bandmerge, but are also the most reliably identified in scan processing (see IV.7). From a purely statistical standpoint, three-band sources are generally highly reliable, two-band sources are slightly less reliable, and single-band sources are the least reliable. Therefore, three-band-detection is used, together with a lower SNR threshold to augment the PSC with fainter sources that are still highly reliable. As an example of this, Figure 9 shows the J-H vs. H-Ks color-color diagram for a sample of low SNR sources in the PSC in the north and south Galactic caps. Black points represent sources with 5 < scan SNR < 7 in at least one band, yellow points have 5 < Scan SNR < 7 in 3 bands, and the blue points have 6 < scan SNR < 7 in 3 bands. The green, red and magenta overlays show the loci of colors expected for dwarf and giant stars, and S0 galaxies ranging from 0 < z < 1.0. The colors of the high latitude, low SNR PSC sources are consistent with those of unresolved galaxies with redshifts of a few tenths.

The 2MASS PSC and XSC contain sources with with SNR < 5 in some bands, because a higher-SNR measurement in one or two bands will "pull along" fainter detections in the other band(s).

Figure 6Figure 7Figure 8Figure 9

ii. Flux Overestimation Bias

The second purpose of the SNR thresholds for the All-Sky Data Release is to filter out sources from the PSC and XSC which are most affected by flux overestimation bias. This bias is a natural consequence of selecting a flux-limited sample of sources with non-zero measurement uncertainties. A source with an intrinsic brightness near the sensitivity limit of a measurement is more likely to be detected if noise drives up the measured brightness, as opposed to driving it down. Therefore, sources detected near the sensitivity limit will have, on average, a measured brightness higher than their true brightness, or equivalently a higher SNR than their true value. Such sources will also have measurement errors that do not accurately represent their true SNR. The closer a measured brightness is to the detection limit, the larger the amplitude of the statistical overestimation. A simulation based on pure Gaussian noise statistics (see V.3a) shows that the flux overestimation is ~10% at the SNR~5 level, and still 5% at the SNR~7 level, but diminishes rapidly for brighter sources. Because the noise is rarely Gaussian, this is a lower limit for the expected flux overestimation.

The flux overestimation bias is especially troublesome for distributions in which the number of sources rises rapidly with decreasing brightness, such as most astronomical source distributions. There are more sources fainter than a detection threshold than brighter, so more sources are scattered above the threshold than will be scattered below it. Therefore, at low SNR, sources will "pile up" in the faintest magnitude bins, causing an apparent excess in source count curves. This effect is clearly seen in the dlogN/dM curves for all WDB point sources in Figures 6, 7 and 8. Because of its statistical nature, it is impossible to avoid this bias entirely, but limiting the Catalogs to higher SNR sources minimizes its impact.

c. PSC - Frame-Detection Limits

Because they do not persist on the sky, transient events, such as cosmic ray strikes, residual meteor trails and hot pixel events, cause spurious detections in only one out of the six (and occasionally seven) frames covering a particular spot on the sky. Most isolated single-frame "events" and meteor trails are filtered out during production of the Atlas Images (IV.3), and this avoids spurious detections from them. However, cosmic ray strikes near true sources, those affecting many pixels (i.e., grazing hits), and a few residual meteor trails do persist into the coadded Atlas Images, and will therefore trigger spurious detections. These spurious sources can have any brightness from the saturation limit down to the faint detection limit.

To filter out these spurious detections, candidate PSC sources that are brighter than J < 14.5, H < 14.0, or Ks < 13.5 mag (SNR > ~25) and that are not saturated on the 51 ms Read_1 exposures are required to satisfy the following frame detection criterion:

The frame detection statistics for point sources are compiled in the ndet parameter. Ndet is a six character flag (NJMJNHMHNKMK) that gives for each band the number of frames on which the source was detected at >3 in aperture photometry (IV.4c), Nb, and the number of frames which were available for measurement, Mb. Thus, the frame detection thresholds are also referred to as the "N-out-of-M" criteria.

i. Low Frame Coverage

The number of frame coverages of a source position, Mb, can be less than six or seven for several reasons, including:

Figure 10 shows the Ks band differential count curves for all point sources in the WDB with |b|>70° that satisfy the edge distance and SNR requirements, given above. The distributions for sources with MK>6, MK=5, MK=4, MK=3, MK=2 and MK=1 are shown in different colors, as specified on the figure. The great majority of sources are measurable in >6 frames. The distribution of sources measurable in MK=5, 4 and 3 frames are well behaved. The distribution changes significantly at MK=2. Most of the brighter sources (Ks<10.5 mag) in the one and two frame coverage curves are spurious detections around bright stars.

For reliable measurements in the PSC, we require that a source should have at least one detected band for which there are at least three usable coverages of a source (Mb>2). Sources with at most two usable coverages in any band are selected only if they are detected and measured cleanly in all three bands on non-saturated in the 1.3 s R_2 exposures (rd_flg="222").

ii. Minimum Number of Frame Detections

The minimum frame detection threshold required for PSC sources is based on the premise that one- or two-frame detections of high SNR sources are highly improbable, and are most likely spurious detections of transient events. A source can fail to have a >3- detection in the aperture measurements on individual frames for the following reasons:

The most common of these occurrences is the first one, and it is why frame-detection thresholds are applied only to sources brighter than J < 14.5, H < 14.0, or Ks < 13.5 mag. This is also the reason why reliable, but faint, sources can have Nb=0. The second occurrence listed above is why the frame-detection thresholds are not applied to any source that is saturated on the 51 ms R_1 frames.

Figure 11 shows the fraction of sources with six frame coverages having different numbers of >3 frame detections, plotted as a function of Ks magnitude for the same |b|>70° sample shown in Figure 10. Sources with missing frame detections are clustered around the R_1 and R_2-R_1 saturation levels, and the fraction of missed detections increases towards fainter magnitudes. The "plateau" of bright NK=5 detections (shown by the blue curve) for 5 mag < Ks < 8.5 mag is not caused by large number of missing 51 ms R_1 frame detections, but is due to an accounting error in how the number of measurable frames was tabulated when masked pixels were encountered for those sources.

Figure 10Figure 11

d. XSC - Extended Source Classification, Untracked Seeing and the Galactic Center

i. The e_score and g_score Limits

Candidate sources for the All-Sky Data Release XSC are required to have:

The Extended Source WDB contains candidate objects selected to be extended with respect to the observed point-spread function within a Tile. These candidates are a combination of resolved objects, galaxies and nebulae, and "false" extended objects, primarily close multiple stars. As discussed in Section IV.5, further classification of objects in the Working Database using an "Oblique Decision Tree" (IV.5) which incorporates three radial extent attributes, three symmetry attributes and four photometric attributes, assigns a confidence score to each band for a source. The final scores are SNR-weighted average of the three wavelength band scores, the e_score and g_score. The e_score is tuned to finding any resolved sources, and the g_score is optimized to identify galaxies among the candidates, using color information. A point source has e_score=2.0, while a clearly resolved source has e_score=1.0.

The upper panel of Figure 12 shows the e_score for extended source candidates from the Working Database plotted as a function of the integrated J magnitude The classification of each source in this figure was determined using either visual inspection of optical or non-2MASS infrared imaging data, or with spectroscopy. Verified galaxies are denoted by filled white circles, double stars by red triangles and higher multiples of stars by cyan cross symbols. Galaxies cluster in two places, either at an e_score of 1.0 or around 1.1 - 1.4 at the faint end (J~14, Ks~13 mag). The galaxy clustering is due to the weighted averaging; the score value jumps from 1 if all three bands agree, to an intermediate value of about 1.3, if the source failed in one band. "False" galaxies are predominantly located at 2.0, with clustering around 1.5 to 1.8. The lower panel of Figure 12 shows the g_score for extended source candidates from the Working Database, plotted as a function of the integrated Ks magnitude. Verified galaxies tend to have a g_score that ranges from 1.0 to 1.4. This score does a slightly better job of discriminating galaxies from non-galaxies than does the e_score.

Figure 12

e. Bright Star Artifacts

As discussed in Sections IV.5 and IV.7, spurious detections of and real source detections affected by optical artifacts of bright stars are identified during scan pipeline processing and in the Catalog Generation phase (V.5). For inclusion in the PSC and XSC, sources must not be identified as artifact detections in that processing. Sources which are believed to be real astrophysical objects, but which may have positions and/or photometry affected by nearby artifacts, are passed to the Catalogs and are flagged as possibly corrupted using the "cc_flg" (cf. II.2a and II.3a).

[Last Update: 2003 March 8, by R. Cutri, T. Jarrett and T. Chester]


Previous page. Next page.
Return to Explanatory Supplement TOC Page.