IV. 2MASS Data Processing

7. Artifact Identification

One of the primary sources of unreliability of 2MASS detections is confusion with artifacts produced by bright stars. These artifacts include familiar optical effects such as diffraction spikes, ghost images, and filter and dichroic glints, as well as features unique to the NICMOS3 detectors used by 2MASS such as latent images and "stripes" (see below).

The following J-Band Atlas Image from the northern night of 1997 Nov 16 UT (Figure 1), contains the bright star 2MASS J02450919+2851242 (= HD17054; J~5.8, H~5.1, K_s~5.0 mag). Figure 2 shows the same image with source detections in a "blinking" (or "animated") overlay. "Sources" associated with artifacts from this star and others are marked with squares. Note that these artifacts are not included in the Point Source Catalog. The color coding of the squares is as follows:

• Blue - Diffraction spike detections
• Red - Persistence (latent) image detections, extending to the north of bright stars
• Magenta - Confusion sources detected in the wings of HD17054
• Cyan - "Stripe" detections approximately 256´´ north and south of HD17054
• Brown - Filter and Dichroic glint detections

The PSC sources in the image are marked with circles. Green circles are "clean" detections, not influenced by any artifacts. Yellow circles are sources believed to be real, but which may have position and brightness measurements affected because of proximity to artifacts. The yellow circles also have artifact boxes within them to indicate the nature of the offending artifact.

Figure 1	Figure 2

During pipeline processing for each 2MASS Tile, artifacts are identified in the extracted source lists by searching for detections that have the correct positional and brightness relationship to bright stars. All sources, true and artifact, are output to the Working Source Databases, but artifacts are flagged therein. Detections that are highly probable artifacts are not included in the final release Catalogs for 2MASS. Sources which are believed to be real objects on the sky, but may have positional or brightness measurements affected by nearby artifacts are flagged in the appropriate bands with the "cc_flg" (confusion and contamination flag) in the PSC. Users are urged to pay attention to the values in this flag.

The possible values of this three character flag, one character per band [JHK_s], the conditions they indicate, and the number of sources in each band having those cc_flg values in the PSC are summarized in Table 1 below. Sources that are affected by more than one of the contamination conditions have cc_flg set in the priority shown by the order of the values in the table. For example, a source that is identified as both contaminated by proximity to a latent image and a diffraction spike from a nearby bright star will have cc_flg="p" in the affected bands.

The variable m_par in this subsection is defined as the parent source's magnitude in a specific band. The artifact identification algorithms and positional parameters work in the universal(U)-scan coordinate system, which is in arcsec and independent of band. In this system, (x_par ,y_par) is the position of the parent source in the U-scan coordinates for the scan. The y-axis of the U-scan system is parallel to the telescope scan direction, pointing in a northward direction. The relationship between the U-scan system and J2000 RA-Dec axes changes with Tile, but the x-axis of the U-scan system never gets more than 3-4 arcmin out of the J2000 equatorial plane. Thus, away from the equatorial poles, to an approximation, x or cross-scan is in the negative RA direction and y or in-scan is in the Dec direction. In this page, the cardinal directions north, south, east and west actually refer to these approximations, so that "north" is the positive in-scan direction and "east" is the positive RA direction. For more information on 2MASS coordinate systems and transformations, see IV.6.

Additionally, final calibrations and corrections were applied to source magnitudes near the end of pipeline processing, so the magnitudes used in the pipeline do not exactly match those found in the Catalog. Therefore, any calculation of artifact areas and flags using RA, Dec, PSC magnitudes, and the algorithms and parameters found herein is only a close approximation to those calculated during processing, especially near the equatorial poles.

a. Specific Artifacts

i. Diffraction Spikes [cc_flg="d"]

Diffraction spikes are linear features that extend in the north, west, south, and east directions from bright stars (in clockwise order on the images). Their widths gradually increase with distance from the parent source, and their lengths are dependent on the parent source's magnitude.

Since the diffraction spikes also vary in length, width, and brightness with facility (northern or southern), band, and direction, separate parameters are used to calculate the boundaries for each spike direction in each band and facility. The parameters were determined by the patterns of false sources created by the diffraction spikes and faint real sources pushed over the detection threshold by the extra flux, so the spike boundaries trace the areas in which the spikes cause false sources, not necessarily the full area of the visible spikes on the images. These parameter values are given in Table 2 below. The spike widths and lengths, as a function of band, are shown in Figures 3-14 below.

General Parameters

Faint parent sources do not cause diffraction spikes bright enough to create false sources. Therefore, diffraction spike searches are conducted only if the parent source is brighter than the m_thrD parameter for that spike.

All sources found within the predicted boundaries of a spike are at least contaminated by the spike. Those fainter than m_par + _mD are considered spurious detections and are culled from the Catalog during the Catalog Generation process (see V.1). Sources brighter than the threshold are included in the release Catalogs, but are marked as contaminated by the spike. Such contaminated point sources have a value of "d" in the appropriate band in the "cc_flg" in the PSC record.

Spike Length

The calculated length l in arcsec of each spike area, from the center of the parent source, is:

l = l₀ * 10 ^{[a * (m₀´ - m_par) ]}

Source Density Adjustment

The initial value of the zero-point parameter m₀ for all diffraction spikes is 6.0. However, as source densities increase, the confusion noise raises the background noise levels such that the diffraction spike lengths decrease. For these purposes, the source density value src_dens is simply the number of source detections in the Tile for one band divided by the survey Tile area, 0.85 deg². If the Tile's source density is greater than 5883.0 per deg^-2 (i. e., the number of source detections in the Tile for that band is greater than 5000), the m₀ parameter is shifted using:

m₀´ = 6.0 - [ log₁₀(src_dens/5883.0) / d ]

Spike Width

The diffraction spike widths increase with distance from the parent source, making the spikes into wedges that fan out from the center. The equation to calculate the spike half-width is

w = w₀ + (e * z)

Table 2: Diffraction Spike Parameters
Northern Facility
	J							H							Ks
	mthrD	mD	l0	a	d	w0	e	mthrD	mD	l0	a	d	w0	e	mthrD	mD	l0	a	d	w0	e
N spike	9.75	6.5	91.0	0.127	0.3172	2.0	0.0167	9.25	6.0	76.0	0.132	0.2175	2.0	0.0167	9.00	6.0	62.8	0.139	0.2701	2.0	0.0167
W spike	9.00	6.5	65.4	0.121	0.3517	2.5	0.0278	8.50	6.0	55.0	0.125	0.4641	2.5	0.0300	8.00	6.0	54.2	0.141	0.3223	2.5	0.0300
S spike	9.00	6.5	79.4	0.111	0.3012	2.0	0.0167	8.50	6.0	61.6	0.129	0.4723	2.0	0.0200	8.25	6.0	59.6	0.140	0.5304	2.0	0.0233
E spike	9.75	6.5	86.4	0.114	0.2940	2.0	0.0250	9.75	6.0	65.4	0.126	0.3166	2.0	0.0296	8.75	6.0	75.0	0.112	0.2027	2.5	0.0250
Southern Facility
	J							H							Ks
	mthrD	mD	l0	a	d	w0	e	mthrD	mD	l0	a	d	w0	e	mthrD	mD	l0	a	d	w0	e
N spike	9.00	6.5	92.2	0.112	0.3620	2.0	0.0200	9.25	5.5	57.0	0.124	0.6244	2.0	0.0167	9.75	6.5	63.4	0.137	0.6943	2.5	0.0167
W spike	9.75	6.5	82.2	0.105	0.3831	2.0	0.0200	9.75	5.5	88.4	0.103	0.3845	2.0	0.0200	8.50	6.5	73.2	0.101	0.3893	2.0	0.0200
S spike	9.75	6.5	85.2	0.122	0.4871	2.0	0.0200	9.75	5.5	60.4	0.127	0.4871	2.0	0.0200	9.00	6.5	56.2	0.137	0.7818	2.0	0.0200
E spike	9.75	6.5	75.0	0.116	0.4258	2.0	0.0200	8.75	5.5	61.2	0.118	0.7503	2.0	0.0200	8.00	6.5	66.2	0.113	0.4923	2.0	0.0200

	Northern Facility
	J	H	Ks
Spike Lengths
	Figure 3	Figure 4	Figure 5
Spike Widths
	Figure 6	Figure 7	Figure 8
Source Density-Adjusted m0
	Figure 9	Figure 10	Figure 11
Source Density-Adjusted Spike Lengths for mpar = 3.0
	Figure 12	Figure 13	Figure 14
Source Density-Adjusted Spike Lengths for mpar = 5.0
	Figure 15	Figure 16	Figure 17
Source Density-Adjusted Spike Lengths for mpar = 7.0
	Figure 18	Figure 19	Figure 20
	Southern Facility
	J	H	Ks
Spike Lengths
	Figure 21	Figure 22	Figure 23
Spike Widths
	Figure 24	Figure 25	Figure 26
Source Density-Adjusted m0
	Figure 27	Figure 28	Figure 29
Source Density-Adjusted Spike Lengths for mpar = 3.0
	Figure 30	Figure 31	Figure 32
Source Density-Adjusted Spike Lengths for mpar = 5.0
	Figure 33	Figure 34	Figure 35
Source Density-Adjusted Spike Lengths for mpar = 7.0
	Figure 36	Figure 37	Figure 38

ii. Persistence [cc_flg="p"]

When a very bright source illuminates pixels on a NICMOS3 array, a latent signal that decays with time will persist in subsequent reads of those pixels. Because of the regular stepping in declination of 2MASS scanning (III.2a), bright stars will leave a "trail" of false stars of diminishing brightness, spaced at regular intervals in the direction opposite to the scanning. The brightness and duration of the persistence "trail" is a repeatable function of the parent source brightness. Note the prominent persistence trail extending north of HD17054 in the image above, as well as the fainter persistence artifacts just north of even much fainter stars. Because these artifacts appear stellar, they are particularly insidious especially when inspecting 2MASS images. The location of known persistence image artifacts are shown as overlays on the 2MASS Atlas Images that are retrieved using the IRSA 2MASS Image Services.

Algorithm

Because the scanning step size is known very accurately, and because the decay in brightness of the latent images is well understood, spurious detections of persistence artifacts can be identified in the source lists during pipeline processing for each scan using simple geometric and brightness relationships for parent sources brighter than 11.5 mag in the J band, 10.0 mag in H, or 9.5 mag in K_s. Very bright parents, especially those saturated in the Read1 images, have larger position errors than most 2MASS sources, which in turn cause larger errors in the predicted persistence positions. These persistence sources are also much less point-like than normal sources, because the saturated core of the parent is so large, and thus the persistence detections are also more spread out. Therefore, the persistence search radii and predicted persistence position errors for parents brighter than 5.0, 4.5, or 4.0 mag in J, H, or K_s, respectively, are larger than those for fainter parents. For each detection that is within 6´´ of an expected persistence artifact, or within 14´´ if the parent is brighter than 5.0, 4.5, or 4.0 mag in the J, H, or K_s bands, respectively, the probability that it is an artifact is evaluated based on a positional and brightness ² function, which depends on the expected position and magnitude of the persistence artifact and the difference between the actual detection and the expected values. Using Numerical Recipes routines (Press, W., Flannery, B.P., Teukolksy, S.A., & Vetterling, W.T. 1988, Numerical Recipes in C [Cambridge: CUP], Ch. 6), the probability is normalized such that values > 0.5 are likely persistence artifacts, and are culled from the final Catalog source lists. Objects with persistence probabilities between 0.1-0.5 are included in the Catalog lists, but are flagged as having measurements that are likely influenced by nearby persistence artifacts. The "cc_flg" value corresponding to persistence source contamination is "p."

The expected position of a persistence artifact found at the location of the i^th frame offset following the parent source is approximately:

x_iP = x_par + (i * xoffs)
y_iP = y_par + (i * yoffs)

xoffs ~ 0.5´´
yoffs ~ 83´´

and for south-going scans of Tiles:

xoffs ~ -0.5´´
yoffs ~ -83´´

The predicted magnitude of the i^th persistence artifact is approximately given by:

m_iP = m_par + _miP

The ² function is thus computed as:

² = [(x_s - x_iP)² / _x² ] + [(y_s - y_iP)² / _y² ] + [(m_s - m_iP)² / _m² ]

_x², _y²,

The persistence searches end when the value of m_iP, within its uncertainty, is fainter than the approximate source detection limits of the survey. Since the predicted magnitude uncertainty is approximately 1.0 mag, the search is ended when m_iP > 18.7 mag in J, 18.2 mag in H, or 17.7 mag in K_s for the northern facility, or 18.7 mag in J, 18.0 mag in H, or 17.6 mag in K_s for the southern facility.

Table 3: Persistence Parameters
	Northern Facility										Southern Facility
i	1	2	3	4	5	6	7	8	9	10	1	2	3	4	5	6	7	8	9	10
J miP	5.8	7.6	8.4	8.8	9.1	9.4	9.8	10.5	12.0	13.0	6.2	8.1	8.9	9.5	10.0	10.4	10.8	11.5	12.0	13.0
H miP	6.3	7.8	8.4	8.8	9.1	9.2	9.4	11.0	12.0	13.0	5.8	7.1	7.6	8.3	8.6	8.9	9.3	10.2	12.0	13.0
Ks miP	6.3	7.8	8.4	8.8	9.1	9.4	9.6	10.5	12.0	13.0	6.3	8.0	8.8	9.3	9.4	9.5	9.8	11.0	12.0	13.0

iii. Filter and Dichroic Glints

Glints are point-like images that result from internal reflections of bright stars within the filter and dichroics of the 2MASS cameras (III.1b). They are found at well-defined positional offsets relative to the bright source, and have magnitudes that are approximately constant offsets from the bright source's magnitude. For both the northern and southern 2MASS cameras, the glints lie within 34´´ of the parent object. Like persistence artifacts, the glints are stellar in appearance, so they can be easily confused with real objects on the sky in the images. The filter and dichroic glints have the added feature that they do not appear in all bands, so they can have extreme apparent colors. For example, because the K_s light passes through two dichroics, there is a K_s-only dichroic that produces red "companions" to all bright stars in 3-color composite images produced from the 2MASS Atlas Images. An example of these can be seen in the images of Gliese 445 and Kapteyn's Star in Section II.4b. There are 2, 1, and 2 glints in the J, H, and K_s bands in the northern camera and 4, 2, and 2 glints in the southern camera, respectively.

Glint artifacts are found in the detection lists during scan pipeline processing by searching for objects near the predicted position offset from a bright star that also have magnitudes within the uncertainty bounds of the expected magnitude. Such objects are culled from the detection lists and not included in the released 2MASS Catalogs during the Catalog generation process (V.1). As with persistence artifacts, the position of filter and dichroic glints are indicated on overlays to the Atlas Images retrieved using the IRSA 2MASS Image Services.

Algorithm

A source is considered a glint artifact if it is found within r_gl arcsec of the glint's expected position (x_gl ,y_gl) and has a magnitude within ±_mgl mag of m_par +_mG, where r_gl, _mgl, and _mG depend on the glint number, band, and facility. The expected position of the glint is:

x_gl = x_par+f_gl
y_gl = y_par+h_gl

Table 4: Glint Parameters
	Northern Facility					Southern Facility
	J		H	Ks		J				H		Ks
Param	glint #1	glint #2	glint #1	glint #1	glint #2	glint #1	glint #2	glint #3	glint #4	glint #1	glint #2	glint #1	glint #2
f	1.5	3.0	1.5	1.5	-4.0	-2.5	1.0	-6.5	-10.0	-2.5	1.0	-2.5	1.0
h	-14.0	-28.5	-14.0	-14.0	16.5	11.0	-13.5	22.0	33.5	11.5	-13.5	11.0	-13.5
rG	2.0	3.0	2.0	2.5	2.5	3.0	1.5	3.5	5.0	3.0	1.5	3.0	2.0
mG	6.7	9.0	7.9	7.8	7.2	6.5	9.0	9.0	12.0	8.0	8.2	7.3	7.5
mG	0.9	2.0	1.6	1.8	1.3	1.0	2.0	2.5	3.0	2.0	2.0	2.4	1.5

iv. Stripes [cc_flg="s"]

Horizontal "stripe" artifacts extend the full east-west width of the scan at the declination of very bright stars, as well as at 256´´ north and south of the star's position. They are 8´´ in full-width north to south, and often have electronic echoes of the star at the locations (x_par ±0, 256´´, y_par ±0, 256´´). These stripes are an electronic feedback associated with the quad readout of the detectors, and are much fainter than diffraction spikes; only parent sources brighter than 6.7 mag in J, 6.2 mag in H, or 5.7 mag in K_s in the north or 6.7 mag in J, 6.0 mag in H, or 5.6 mag in K_s in the south are bright enough to create these lines of heightened charge. An example can be seen in the images of tau Ceti in Section II.4b.

Any detections that fall within the area covered by a stripe may have photometry affected by the artifact. Such objects are flagged with an "s" in the "cc_flg".

v. Bright Star Confusion Artifacts

Sources that are saturated in the "Read 2-Read 1" (R2-R1) and/or "Read 1" exposures usually cause a flood of false detections on those images, in and around their cores. The radius in which these false detections are found is termed the "artifact confusion" radius, and is dependent on the brightness of the parent object, facility, and read type. For very bright objects, such as Pegasi (K_s~-1.9) in the Second Incremental Data Release, the confusion radius can be many arcminutes. "rd_flg"="2" sources found within the R2-R1 exposure artifact confusion radius and "rd_flg"="1" sources within the R1 exposure artifact confusion radius around a very bright star are identified as a spurious detections and not selected for inclusion in the release Catalogs.

Algorithm

Confusion artifacts can only be caused by parents that are saturated in the R2-R1 and/or R1 exposures. However, for sources saturated in the R2-R1 exposure, there often are no good data from that exposure and thus no explicit indication that the parent was saturated. To deal with this situation, magnitude thresholds m_sat are also used to determine if the parent should have been saturated in the R2-R1 exposure. These thresholds are found in Table 5 below. If the parent is brighter than the threshold in that band, saturation is assumed for that band. (See also Cautionary Notes below for more information.) The artifact confusion radii for the northern facility is shown in Figure 39, and for the southern facility in Figure 40.

The equation to calculate the artifact confusion radius for a read type (rd_flg = "1" or "2") is

r_iAC = r_0AC * 10^{[ b * ( m_i0AC - m_par ) ]}

Table 5: Artifact Confusion Parameters
	Northern Facility			Southern Facility
	J	H	Ks	J	H	Ks
msat	8.0	7.7	7.1	8.0	7.6	7.1
r0AC	12.90	10.20	8.50	15.06	9.46	7.74
b	0.202	0.203	0.171	0.207	0.201	0.194
rd_flg="1" m0AC	2.5	2.5	2.5	2.5	2.5	2.5
rd_flg="2" m0AC	6.0	6.0	6.0	6.0	6.0	6.0

Figure 39	Figure 40

vi. Photometric Confusion [cc_flg="c"]

While saturated sources can cause false detections in and around their cores, any real source located near another real source may have magnitudes that are affected by the neighbor's light. When the photometry is affected at the 5% level or greater, the source is flagged as having "photometric confusion" with a "c" in the "cc_flg". As with artifact confusion, the photometric confusion radius can be very large for very bright parents.

Algorithm

All sources near other sources brighter than the approximate source detection limits are checked for photometric confusion. The photometric confusion radii depends on both the magnitudes of the parent and the fainter neighboring source and thus is calculated separately for each possible confusion pair. The equation to calculate the radius is:

r_PC = r_0PC * 10 ^{[c * (m_fs - m_par) ]}

Table 6: Photometric Confusion Parameters
	J	H	Ks
r0PC	4.678	4.380	3.980
c	0.1165	0.1148	0.1243

Figures 41 and 42 show the maximum photometric confusion radius for a certain parent magnitude in each band and facility, where the second fainter source has a magnitude equal to the approximate source detection limits in that band and facility.

Figure 41	Figure 42

The following discusses known problems in the application of the artifact identification algorithms described above. Confirmed unreliable sources in the PSC are maintained in an Anomaly File that will be updated periodically.

i. Residual Diffraction Spikes from Very Bright Parents

A detailed analysis has been made of residual diffraction spike artifacts in the PSC from very bright stars. These residual artifacts are most obvious around parents brighter than ~ -2 mag, although some faint remnants can also be seen around parents between about 0 mag and -2 mag in data from the southern facility. There are a total of 5 stars creating the most obvious residual spike artifacts for the northern facility, and 7 stars doing the same for the southern facility; all of these stars are known variables. Since these bright stars can cast diffraction spikes across scan boundaries, into adjacent scans that may have been observed at very different times, the artifact search parameters used in these adjacent scans may be based on the incorrect brightness of the parent at the time of the actual scan observation.

The sources located in the areas covered by the residual spikes were also examined. Most are single-band sources; almost all of the rest are two-band sources with ph_qual lower than "A" in both detected bands; and only a few in the K_s band from the southern facility are triple-band sources which have low photometric quality, with ph_qual = "C" or "D" in each band. The profile-fit photometry ² values for the sources in the residual spike areas tend to be larger than 2, but the extended high ² tail for both the populations of real sources and possible artifacts in these regions makes it impossible to use this quantity alone to separate the populations. Table 7 below gives the total numbers of single-band and double-band sources found in the rectangular areas approximately covered by the residual spikes for each band and facility; these numbers would be the maximum number of unreliable sources left in the PSC from these residual spikes.

Table 7: Residual Spike Source Counts
		# of 1-band srcs in spike rectangles	# of 2-band srcs in spike rectangles
Northern Facility	J	591	231
	H	872	333
	Ks	804	252
Southern Facility	J	2915	1289
	H	3044	1570
	Ks	5029	1694

ii. Artifacts from Skipped Single-Band Parents

Since very bright stars can create artifacts across scan boundaries, the processing searched for artifacts in Tiles near stars brighter than or equal to 5 mag in any band. However, the pipeline processing has trouble cleanly detecting and band-merging each component of close visual multiple stars, sometimes resulting in single-band or double-band detections of one or more component stars and occasionally a set of false "detections" between the components.

For the very bright parents with double-band detections, the processing interpolates a parent source magnitude in the missing band as an improvement over the band-filled magnitude found in the source data, which is optimized for faint sources not detected on the R2-R1 exposures. The interpolation equations are:

m_j´ = 2.695 * m_h - 1.695 * m_k
m_h´ = 0.370 * m_j - 0.630 * m_k
m_k´ = 1.598 * m_h - 0.587 * m_j

However, when a parent source was not detected in at least two bands, it was not used to find artifacts in nearby Tiles. There are 12 of these single-band parents brighter than 5 mag that have no other "good" duplicate sources in the database; all but one are members of close visual multiples. ("Good" parents are found in the set of released scans of Tiles, not flagged as artifacts, and have detections in two or three bands.)

None of these parents are very close to Tile edges, so they did not create any confusion artifacts in nearby Tiles. For each of the 11 skipped parents that are members of close visual multiples, there was at least one two- or three-band extraction of another component of the group or in the middle of the components, and artifacts that happened to fall in the areas masked by those other extractions were properly flagged. Also, the processing did search for artifacts found in the same Tile as their parents. Therefore, only a few of the diffraction spike artifacts located in nearby Tiles remain unflagged; 28 of these are found in the PSC, and are found in the PSC Anomaly File. These artifacts were caused by nine of the 12 skipped parents; the other three did not cause any unflagged artifacts.

iii. Confusion Artifacts from the Read1 Images

As mentioned above in section IV.7.a.v, confusion artifacts can only be caused by saturated parents. However, saturated sources are sometimes not detected in the saturated exposure; in these cases there are no explicit data indicating saturation. To deal with this situation, magnitude thresholds m_sat are also used to determine if the parent should have been saturated in the exposure. If the parent is brighter than the threshold in that band, saturation is assumed for that band.

Unfortunately, the same magnitude threshold values used for R2-R1 exposures were also used for R1 exposures, instead of those values minus ~ 3 mag: 5.0 mag in J, 4.7 mag in H, and 4.1 mag in K_s for the northern facility, and 5.0 mag in J, 4.6 mag in H, and 4.1 mag in K_s for the southern facility. Therefore, for the confusion artifact searches, parents between 8.0 and 5.0 mag in J, 7.7 and 4.7 mag in J, and 7.1 and 4.1 mag in K_s for the northern facility and 8.0 and 5.0 mag in J, 7.6 and 4.6 mag in J, and 7.1 and 4.1 mag in K_s for the southern facility were treated as saturated in the R1 exposure even if they were not actually saturated.

The Read1 exposures' artifact confusion radii, however, are very small for parents with magnitudes between the correct thresholds and those actually used: the radii at the correct threshold magnitudes are 4.0´´ in J, 3.6´´ in H, and 4.5´´ in K_s for the northern facility, and 4.6´´ in J, 3.6´´ in H, and 3.8´´ in K_s for the southern facility, and the radii at the incorrect threshold magnitudes are, respectively, 1.0´´, 1.0´´, 1.4´´ 1.1´´, 1.0´´, and 1.0´´. For comparison, source deblending will only work on source pairs separated by ~ 4´´ or less, and the minimum source resolution of the survey is ~ 1.5´´. Therefore, most of the rd_flg="1" sources incorrectly flagged as confusion artifacts are probably caused by missed frame merges from the Read1 images.

iv. Unflagged Persistence Sources

As described above in section IV.7.a.ii, bright stars are followed by a "trail" of false persistence sources of diminishing brightness, spaced at scan step intervals in the direction opposite to the scanning. In addition to searching for persistence from parents found in the Tile, the processing also looks for bright stars in the first five frames at the beginning of each scan used to build up to the six-frame-deep coverage of points in the Tile, and searches for persistence and other artifacts inside the Tile area from these "low-coverage area" parents.

However, there was a bug in the search for persistence sources on the Read2-Read1 images from some of these parents in the low-coverage areas at the beginning of a south-going scan, when the predicted position of the first persistence source was north of, or before, the first rd_flg="2" source in the scan. In these cases, an incorrect flag setting would prevent further persistence searches on the Read2-Read1 images at later -- i.e. more southern -- frame positions. This bug does not affect persistence searches with parents in, or less than 1 frame preceding, the R2-R1 exposures' six-frame-coverage area, or in north-going scans. It also does not affect affect searches for rd_flg="1" persistence sources, nor any other type of artifact searches (e.g., glints, diffraction spikes, etc.).

Because of the particular circumstances that cause this problem, only one of these unflagged persistence sources has been found so far during analysis. The only way to estimate how many other unflagged persistence sources might remain in the working database and PSC is to count the flagged persistence sources caused by "low-coverage area" parents. If we assume that the north- and south-going scans of Tiles are distributed randomly over the sky, half of the sources should be found in north-going scans and the other half in south-going scans. This means that approximately half of this type of persistence sources were flagged correctly, and half were not.

The numbers of persistence artifacts, with persistence probabilities > 0.5, and persistence-affected sources, with persistence probabilities 0.5, in the PSC and working database from parents in the first five frames of Tiles are found in the first two lines of Table 8 below. Only 20-25% of these sources in the working database with persistence probabilities 0.5 are also found in the PSC, probably because most of them are, as expected, real sources that happen to fall on a persistence position and thus have duplicate extractions in neighboring Tiles. These duplicates are likely to be farther from the Tile edges and therefore would be selected in the duplicate-resolution processing, and they will either have full-coverage persistence parents if their scans run in the same direction as the scans with low-coverage parents, or will not fall on a persistence position if their scans run in the opposite direction.

However, not all low-coverage parents will be able to create more than one persistence source, as required for this bug to occur. The numbers of low-coverage parents creating more than one persistence source in the working database and PSC are found in the third and fourth lines of the table, and the numbers of persistence sources created or affected by these parents are found in the fifth and sixth lines. Unfortunately, we can only get these counts one band at a time, so the "any band" category is left blank. Since approximately 76% of the sources in the PSC are 3-band sources, these counts will certainly overestimate the numbers of persistence sources.

Assuming that approximately half of this type of persistence sources were flagged correctly, and half were not, and ignoring the fact that most of the sources are 3-band, not single-band, these last two lines of the table are a reasonable estimate of the maximum number of missed persistence sources. A maximum of 94364 working database sources should have had persistence probability > 0.5; if about 17%, 24%, and 19% of them in the J, H, and K_s bands respectively are in the PSC, the catalog contains approximately 5621, 9965, and 3758 missed persistence artifacts in J, H, and K_s, or 19344 total. This is 0.0041% of the total number of sources in the PSC (470992970). The maximum number of missed working database sources that should have had persistence probability 0.5 is approximately 117092; the PSC contains approximately 4130, 10492, and 9273 of these sources in J, H, and K_s, or 23895 total. This is 0.0051% of the total sources in the PSC.

Table 8: Low-Coverage Parents' Persistence Source Counts
	# of sources in working database				# of sources in PSC and (# in PSC / # in WDB)
	J	H	Ks	any band	J	H	Ks	any band
pers.prob > 0.5	53629	53643	30954	83226	--	--	--	--
pers.prob 0.5	36162	60593	67009	106636	7336 (0.20)	15364 (0.25)	14550 (0.22)	24084 (0.23)
parents w/ > 1 src having pers.prob > 0.5	9692	10723	6208	--	--	--	--	--
parents w/ > 1 src having pers.prob 0.5	7930	12861	15556	--	1392 (0.18)	3159 (0.25)	3139 (0.20)	--
srcs from parents w/ > 1 src having pers.prob > 0.5	33063	41521	19780	--	--	--	--	--
srcs from parents w/ > 1 src having pers.prob 0.5	24362	44618	48112	--	4130 (0.17)	10492 (0.24)	9273 (0.19)	--

[Last updated: 2007 December 28, by R. Cutri and T. Evans]