UNDER CONSTRUCTION

Introduction

The highly reduandant observations of the 2MASS Calibration Fields provide a means to assess reliability of point source extractions in the various working databases from the Survey. In this memo, we use the confirmation statistics for point sources extracted from four calibration fields to form a truth table against which the the reliability of point sources drawn from survey scans can be compared as functions of various parameters. These comparisons may be helpful in assigning the reliability score to point source WDB entries.

This analysis employs a method similar to Mike Skrutskie's analysis of completeness for the All-Sky PSC.

Method

• Four high galactic and ecliptic latitude calibration fields that do not contain any very bright stars or galaxies are selected for this analysis: fields 90266 (2776 scans), 90272 (1973 scans), 90301 (3430 scans) and 90330 (1192 scans).



• The RA and Dec corners that define the regions fully covered by all scans of each field are determined.



• We select all point source extractions in the Survey point source WDB contained within the full-coverage regions for each field, excluding any extraction marked as an artifact (cc_flg NOT MATCHES '*[A-Z]*'), or that does not have a clean measurement in at least one band (rd_flg MATCHES '*[123]*').



• The Calibration point source WDB is then searched to find all extractions within 1.5" of the position of each survey WDB extraction.



• For each source in the Survey WDB, the mean and rms (root variance) position and magnitudes of all matching apparitions in the calibration WDB are computed. The skew and kurtosis of the magnitude distributions are also computed. Band-by-band detection statistics are compiled for each source.



• We use the calibration scan confirmation statistics for each source in the survey WDB to determine if the source is real or not, and compile reliability statistics as a function of source properties in the survey scans. Sources detected in a band in at least 50-100 calibration scans are considered reliable.

Calibration WDB Data as Truth Tables

The detection frequency of a source in the calibration WDB can be used as a reliability indicator. For example, Figures 1 and 2 show histograms of the number of times sources were detected in the calibration field 90301 (3430 total scans). J, H and K_s detections curves are shown in blue, green and red, respectively. The histograms show strong peaks near ~10 and ~3400 detections. Figure 2 zooms in on these two peaks. The ~10 detection peak corresponds to the chance corelation of random noise extractions, while the peak near the total number of detections represents reliable, high SNR sources. The relative paucity of sources with 50 to 3400 detections is a result of the steep decline in completeness with decreasing SNR, as is illustrate in Figure 3 which shows the fraction of cal scans in which each source was detected (N_det(J,H,K_s)/3430) plotted versus its average magnitude.

The fall-off in the low count peak in Figures 1 and 2 suggests that sources detected on ~50 more scans are real. To confirm this, the position of sources with J-band detections in 50-200 scans were visually examined on a deep image of the 90301 field made by stacking images of 616 north-going scans. 100% of objects with 100-200 detections have obvious sources at their position (usually faint galaxies), and 78% of sources with 50-100 detections are real. These reliability thresholds appear to be consistent in all four of the selected calibration fields.

The differential reliability within a range of a specific source parameter, such as brightness, SNR, etc., is given by the ratio of the number of reliable extractions to the total number of extractions in that range:

For the number of reliable extractions, N_rel, we use optimistic and conservative thresholds of detections in >50 and >100 calibration scans, respectively.

Figure 1 - Detection frequency histogram	Figure 2 - Detection frequency histogram detail	Figure 3 - Detection fraction versus magnitude

Reliability vs. Magnitude

Figure 4 shows the frequency distribution of reliable and unreliable sources from the four calibration fields as a function of survey scan J, H and K_s magnitudes. The distribution of reliable sources is shown in black and unreliable sources in red. For this illustration, reliable sources are defined to have detections in >50 scans and unreliable sources have <50 detections.

The resulting differential reliability of J, H and K_s extractions in the survey WDB falling in the full-coverage regions of the four calibration fields is shown in Figure 5. Curves for both the N>50 and N>100 thresholds are shown, and there is no appreciable difference. There is no drop-off in reliability until J>16.6, H>15.6 and K_s>15.0 mag.

Figure 4 - Frequency distribution of reliable and unreliable sources vs. magnitude	Figure 5 - Differential reliability vs. magnitude

Reliability vs. Sigma

Figure 6 shows the combined frequency distribution of reliable and unreliable sources from the four calibration fields as a function of the J, H and K_s photometric uncertainty (j_msig, h_msig, k_msig). The distribution of reliable sources is shown in black and unreliable sources in red. For this illustration, reliable sources are defined to have detections in >50 scans and unreliable sources have <50 detections.

Figure 7 shows the resulting combined differential reliability curves for survey WDB source in the four cal fields as a function of photometric uncertainty. In this figure, the reliability curves using N>50 and N>100 thresholds are both shown, and there is no difference until very low SNR levels. Each panel contains vertical lines corresponding to [jhk]_msig=0.1086, 0.1551, 0.2172 mag, the sigma values that define the primary boundaries between ph_qual=A,B,C,D values. The reliability is essentially perfect for ph_qual=A and B extractions.

Figure 6 - Frequency distribution for reliable and unreliable sources vs. sigma	Figure 7 - Differential reliability vs. sigma

Reliability vs. Band Detection

Figures 8,10,12,14 and 16 show the frequency distribution of reliable and unreliable extractions from the four calibration fields as a function of the J, H and K_s photometric uncertainty, for different combinations of bands-detected. As above, reliable sources are defined to have detections in >50 scans and unreliable sources have <50 detections.

Figures 9,11,13,15 and 17 show the differential reliability curves for survey WDB source in the four cal fields as a function of photometric uncertainty for different combinations of detected bands. Again, the vertical lines correspond to [jhk]_msig=0.1086, 0.1551, 0.2172 mag, the sigma values that define the primary boundaries between ph_qual=A,B,C,D values.

Figure 9 shows that 3-band detections have nearly 100% reliability independent of source SNR. This most likely results from the fact that low SNR extractions in one band are usually accompanied by a higher SNR extraction in another band, which raises the effective sensitivity. Figure 11 shows, somewhat surprisingly, that any 2-band detection combination are very reliable out to [jhk]_msig~0.2 (SNR~5.4). For single-band extractions, the reliability begins to drop off in the J and H-bands for msig>0.15 (SNR>7.2). For single-band K_s detections, the reliability is ~40% or lower. The lack of brighter single-band K_s-only sources makes it impossible to assess the reliability trend for SNR>7. Of course the lack of such detections is a good thing.

See Figures 8-17 full-size on one page here.

Distribution of reliable and unreliable sources vs. sigma	Differential reliability vs. sigma
Figure 8 - 3 band detections	Figure 9 - 3 band detections
Figure 10 - JH-only detections	Figure 11 - JH-only detections
Figure 12 - JKs-only detections	Figure 13 - JKs-only detections
Figure 14 - HKs-only detections	Figure 15 - HKs-only detections
Figure 16 - Single-band detections	Figure 17 - Single-band detections

Reliability vs. Chi-Squared

Reliability vs. ndet

Last Updated: 23 April 2004
R. Cutri - IPAC