previous page   back to table of contents   next page

IV. WISE Data Processing


6. Quality Assurance (QA)

a. Steps in the QA Process

iii. Multiframe QA

Automated QA consisted of a set of metrics judged against pre-assigned thresholds. The QA web interface contained other tabular and graphical info with which QA scientists could further judge the quality of the data. The automated metrics are described in the first subsection below, followed by the ancillary tabular and graphical data in the second subsection.

1. QA metrics

Factors qc, qa, and qi were determined per coadd and were the basis for the Coadd Quality (CQ) score. CQ was the minimum of the qc, qa, and qi factors multiplied by ten. CQ was an integer of value 0 (poor quality) or 10 (high quality). However, all Atlas images and sources lists for the All-Sky Data Release were found to have CQ values of 10, so all passed the QA metrics listed below.

qc: This was the factor gauging quality of completion. If the status value from the Multiframe pipeline was equal to 0 (indicating a problem-free completion), then qc=1.0. Otherwise, qc=0.0 because the coadd failed processing.

qa: This was the factor gauging astrometric quality. This factor checked the requirement that the one sigma error in WISE positions with respect to the 2MASS Point Source Catalog positions was < 0.5 arcsec on each axis (i.e., 0.71 arcsec RSS) for sources having SNR > 20 in at least one band. For the purposes of QA, we checked the SNR=20 bin for each of the bands W1, W2, and W3. If the standard deviation of the 2MASS-to-WISE radial distance was less than 0.71 arcsec, qa=1.0; otherwise, qa=0.0.

qi: This was the factor gauging quality of image shape. This was a check to watch for potential image smearing by using the axial ratio of the PSF. The value of qi was 1.0 if rbovera > 0.60 for W1, W2, and W3; otherwise it was 0.0.

2. QA checks

In the following discussion, the processing of the Atlas image with a coadd_ID of 2438m394_ab41 was used to illustrate the kinds of tabular and graphical material used by the QA scientists to assess the quality of the data.

Image quality: Just as with the scan-level data, image quality was checked in the Multiframe atlas images. The main difference between the Multiframe and Scanframe image quality checks arose from the fact that only those frames with passing image quality scores (qi) were used in Atlas image generation. Thus, individual frames were already known to have good image quality so the Multiframe checks tested the quality of position registration for frames going into the Atlas image.

There were two main image quality checks for Multiframe processing. The first was the coadd-averaged PSF, which is analogous to the scan-averaged PSF created for Scanframe processing. As in ScanFrame processing, the visual appearance of the PSF (Figure 1) and a table providing the actual PSF moments (Table 1) were monitored.

Figure 1a - W1 for Atlas image 2438m394_ab41Figure 1b - W2 for Atlas image 2438m394_ab41Figure 1c - W3 for Atlas image 2438m394_ab41Figure 1d - W4 for Atlas image 2438m394_ab41

Table 1 - PSF Moments for Atlas Image 2438m394_ab41
idbandr_major rboverarphir_05pr_01pboveraphinoisepixsignspix sigrmajorsigphisigbovera
2438m394_ab41W14.75700.9539164.604.53988.6056 0.8800161.0098.1690.0570.01080.140.0031
2438m394_ab41W24.70720.9545144.395.01389.5422 0.9000173.00116.4640.0470.01640.220.0048
2438m394_ab41W34.35750.9414157.015.77160.0000 0.00000.00133.3990.2070.05190.630.0164
2438m394_ab41W49.12790.90184.509.41850.0000 0.00000.00523.1281.6640.09820.160.0145

The columns in the table are

Note that all length units are in resampled pixels, where 3 pixels = 1.375".

The second method by which we checked the image quality in the Atlas images was to examine the differences between the profile-fit and aperture-derived magnitudes. For objects that are not excessively bright or faint, this difference should be near zero. Deviations from zero indicated likely problems with image quality because the model PSF did not provide a good fit for the observed PSF. Figure 2 shows plots of the difference in the aperture and profile-fit magnitudes as a function of profile-fit magnitude. No problems were seen for this Atlas image.

Figure 2 - Aperture minus profile-fit magnitude vs. profile-fit mag for each of the four bands. From left to right are shown W1, W2, W3, and W4.

Astrometric calibration: In order to check astrometric quality, WISE sources were searched for positional matches to 2MASS stars using a 3-arcsec matching radius. Then the mean, median, and standard deviation of radial distance and position angle of these were computed. These values are shown in Table 2 below.

Table 2 - Astrometry Checks for Coadd 2438m394_ab41
r_dist_meanr_dist_medr_dist_stddevr_posang_meanr_posang_medr_posang_stddev
0.514
0.325
0.384
-0.600
0.232
89.524

The columns in the table are:

Next, the radial distances were binned by the per-band signal-to-noise ratio (SNR) of the profile-fit photometry (WPRO). The SNR bins were set at 3, 5, and 20. These bins were cumulative, meaning that each bin contains all stars with SNR greater than the given SNR floor. For each bin, the mean, median, and standard deviation of radial distances were computed, as shown in Table 3 below.

Table 3 - Astrometry Checks Binned by WPRO SNR for Coadd 2438m394_ab41
nameW1W2W3W4
SNR3_mean_r_dist0.5110.4980.4260.427
SNR3_med_r_dist0.3240.3110.2460.250
SNR3_stddev_r_dist0.3810.3750.3400.337
SNR5_mean_r_dist0.5090.4840.3860.343
SNR5_med_r_dist0.3230.2960.2110.188
SNR5_stddev_r_dist0.3790.3680.3170.281
SNR20_mean_r_dist0.4380.2830.1410.204
SNR20_med_r_dist0.2650.1740.0990.143
SNR20_stddev_r_dist0.3330.2120.0910.134

The entries in the table are:

The quality of astrometry of the Atlas image was judged by the standard deviation of the radial distance of matches in the cumulative SNR = 20 bin. The threshold of 0.71 arcsec RSS is equivalent to a threshold per axis of 0.5 arcsec.

The plots of coadd astrometric quality shown in Figure 3 were produced by first binning the radial distances of matches separately by profile-fit and aperture photometry magnitude. The bins were each of size 0.2 mag. For each bin, the mean, median, and standard deviation of radial distances were computed. Then plots of bin mean (red) and median (green) radial distance vs. bin magnitude were made, with error bars centered on the mean, of +/- 1 σ. There is one set of plots for bins by profile-fit magnitude, and one set for bins by aperture magnitude.

Figure 3 - The mean radial offset between WISE computed astrometry and the 2MASS positions of sources detected by 2MASS. Radial offsets are computed for each magnitude bin, as described in the text. Profile-fit photometry is shown in the top row of plots; aperture photometry is shown in the bottom row.

Visual checks: As a final check of data integrity, QA scientists visually inspected the images in all bands. The QA interface presented jpeg thumbnails of each frame along with the coverage maps in all bands (Figure 4). The visual inspection was done in an effort to reveal any problems not captured by the QA metrics.

Figure 4 - In the first and third rows are shown the full 1.56x1.56 deg Atlas images for (from left to right) W1, W2, W3, and W4. The first row shows the images at our canonical stretch and the third row shows the same images with a higher contrast stretch. Rows two and four show three-color images at these same stretches. The first of the two images is comprised of W1(blue) + W2(green) + W3(red) and the second is comprised of W2(blue) + W3(green) + W4(red). The fifth row shows coverage maps for W1, W2, W3, and W4. The greyscale stretch in the coverage maps goes from black for a coverage of four or fewer frames up to white for the maximum coverage in the image, which can vary depending on factors such as ecliptic latitude and quality score.


Last update: 2012 January 24


previous page   back to table of contents   next page