Appendix 5. Working Database Source Reliability Estimation

1. Introduction

The 2MASS Survey, 6x and Calibration point and extended source Working Databases (WDBs) contain all "source" extractions made from the each of the respective imaging data sets. These extractions include detections of real astronomical sources as well as a large number of spurious detections of faint noise excursions, image artifacts from bright stars and transient events such as cosmic rays, residual meteor trails and hot pixels. This "detect-and-save-everything" strategy was designed to ensure the completeness of the 2MASS All-Sky and 6x Catalogs. The strict reliability requirements of the Catalogs were achieved by drawing from the WDBs the subset of extractions that satisfied criteria designed to filter out the spurious extractions (see V.1 and A3.6).

Because of the large number of spurious extractions in the 2MASS WDBs and Reject Tables, the tables must be used with caution. To assist in navigating the potentially treacherous waters of unreliable extractions, each entry has been assigned a reliability score (rel) that is related to the probability that it is a valid detection of a real astrophysical source at the time of the 2MASS observation. For extended sources, the reliability score also refers to the probability that the source is extended with respect to the instantaneous point spread function and not a single or multiple star.

By selecting on reliability score values, subsets of the WDBs and Reject Tables appropriate for different applications may be constructed. For example, "blind" searches for objects with unusual color or flux properties should be limited to extractions with high probabilities of reliability. Searches for near-infrared counterparts to sources from other catalogs, on the other hand, can be extended to lower reliability extractions because there is a priori knowledge of the existence of a source at a certain position. However, the latter must be done carefully because in regions observed many times by 2MASS, such as the equatorial poles or the calibration fields, the surface density of spurious noise extractions may be large. The probability of matching to the position of a random noise detection is correspondingly high in such areas. When in doubt about the reliability of any 2MASS source table entry, we recommend that you visually examine the image of the source.

a. The Reliability Score

The reliability score assigned to each extraction in the 2MASS Survey, 6x and Calibration point and extended source WDBs is a single character flag with values ranging between "A" and "F". "A" indicates extractions with the highest reliability and "F" the lowest. The range of probabilities corresponding to each score value is given in Table 1. The reliability score is found in the rel column of each of the WDB tables.

As discussed below, the reliability score is derived from the fractional reliability of large ensembles of extractions. It is a statistical estimator that may not be accurate for individual sources. For example, even among a sample of extractions with the highest estimated reliability (rel="A"), up to 10% may be spurious. Thus, taken individually, any extraction may be unreliable despite its reliability score. This is somewhat tempered for extended sources where extensive visual classification of images has been used to identify spurious detections of artifacts with high confidence.

The reliability score does not encode the accuracy of the measured position or brightness of a WDB extraction. Users should refer to the various uncertainties and measurements quality indicators in the respective databases for this.

Table 1 - Estimated Range of Probabilities for Each Value of the Reliability Score, rel
relProbability Range
A P>90%
B 80<P<90%
C 70<P<80%
D 50<P<70%
E 20<P<50%
F P<20%

b. Estimating Source Reliability

The same principle was used to estimate the reliability of both point and extended source extractions. In both cases, large "truth tables" were constructed that contain both reliable and unreliable extractions that were externally validated. The fractional reliability in the truth sets (i.e. the ratio of the numbers of reliable to total extractions) were then measured as a function of a variety of parameters such as SNR, band-detection combinations, extended source e_score and g_score, etc. Relationships between the fractional reliability and the various parameters were derived, usually in the form of look-up tables. These generalized relationships were then applied to the individual extractions in each database to assign a reliability score. In addition, some of the source quality criteria developed for the All-Sky PSC and XSC generation that were not incorporated into the fractional reliability relationships were used to refine reliability scores.

For point sources, the "truth set" was provided by the highly repetitive calibration scan observations (A4.1). Reliability of individual extractions was established using confirmation statistics derived from detection repeatability in the Calibration point source WDB (i.e. how many times a source was detected out of how many times it was observed). The derivation of the point source reliability scoring algorithm from the repeatability statistics is described in A5.2.

The extended source "truth set" was generated by visually examining very large numbers of images of entries in the Survey, 6x and Calibration Scan extended source WDBs. Confirmation statistics from the redundant calibration scans could not be used for extended sources because of the relatively small number of galaxies in the limited area of the fields. Derivation of the algorithm for estimating extended source reliability is described in detail in A5.3.

c. Other Reliability Indicators

In addition to the reliability score provided in the 2MASS WDBs, there are several other factors that can be used to assess the reliability of individual extractions, or to select preferentially real source detections among the chaff.

Visual Examination

Perhaps the very best way to verify the reliability of any 2MASS source extraction is to look at its image. Valid source detections are always visible on one or more of the J, H or Ks images. Spurious detections are usually blank in all bands, or some obvious image artifact such as a diffraction spike or scattered light ghost. Artifacts that can mimic real sources that were identified during 2MASS data processing, such as latent images or dichroic glints (IV.7), are conveniently marked on the images displayed using the IRSA Interactive 2MASS Image Service. When reviewing images of locations that may have been observed more than once by 2MASS, it is important to request the image from the epoch corresponding to the WDB extraction in question. This is easily done by specifying the date, observatory hemisphere and scan number, or the coadd_key in the interactive viewer form page. All of these are carried in the Catalog, Reject Table or WDB entry.

Repeat Detections

The confirmation statistics derived during the point and extended source WDB merging process (A6.1) provide reliability information for 2MASS extractions from regions that were observed multiple times. The spos and sdet columns in the Reject Tables and WDBs give the number of times an extraction was observed and the number of times it was detected in one or more bands. Extractions detected more than once, and for which sdet/spos is near unity are more likely to be real sources. Confirmation statistics were not incorporated into the reliability scoring of the WDBs because their availability is very non-uniform, being limited by the distribution of multi-epoch sky coverage during 2MASS observations. Where it is available, though, source repeatability is a useful reliability indicator. The Calibration Scan WDBs are particularly well-suited for this, hence their use to define the general reliability scoring algorithms for point sources.

Associations with External Catalogs

Positional association between 2MASS WDB entries and objects in reliable external catalogs with good astrometry is another good indicator of 2MASS extraction reliability. For example, reliable 2MASS detections are more likely to have Tycho or USNOA-2.0 associations listed in the various point source WDBs (IV.4.f). However, absence of an optical counterpart does not necessarily indicate unreliability because of proper motions, foreground obscuration, intrinsically red objects, and because of the optical association error present in the Survey data processing (I.6.b.xvi). Matching with large, deep optical surveys closer to the 2MASS epoch, such as SDSS, avoid much of the proper motion limitation. Longer wavelength surveys, such as FIRST and NVSS, can provide good confirmation for sources in regions were foreground extinction may mask distant optical counterparts to 2MASS detections. However, positional corelation between external catalogs and the 2MASS Reject Tables and WDBs must be done carefully because the probability of a chance assocation with an artifact or low SNR noise detection is non-negligible, and can in fact be quite high in regions 2MASS observed many times such as the equatorial poles and calibration fields.

[Last Update: 2006 September 29, by R. Cutri]

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