Repeatibility Tests

T. Jarrett, IPAC
(000517) revised: 010313

Galworks V3.0 Modification Tests

Go to summary report

Three major data sets are considered: repeats of the Abell 3558 galaxy cluster, repeats of the ChamI cloud complex and repeats of the Orion GMC core.

Repeat tests will include the following parameters:

We treat sources that are "contaminated" separately from sources that are "clean". Contamination means that the photometry flags are tripped (non-zero), indicating that another source is nearby and was masked accordingly (with isophotal substitution to recover the lost flux).

We apply the Gscore threshold = 1.4 as the primary star-galaxy discriminant.

Abell 3558 (typical stellar density = 3.1)

A number of repeat scans were acquired during the month of April 2000 to complement the set taken in March of 1998. Here is what I have compiled:

The total area is depicted with this Areal Coverage plot. The scan-scan overlaps of the Abell 3558 core were repeated over 30 times, while the outskirts and most of the cluster area were covered over 15 times; hence, there is more than enough repeats to provide excellent statistics.

Note: I have eliminated inscan duplicates (which are meaningless for these tests); but I have included scan-to-scan dupes (which are meaningful).

RESULTS

Summary Table of Nominal GALWORKS V2 Results

-- -- -- non-contaminated ... | contaminated ... |
fname mag- mag+ #r #p C #r #p C
J7_comp_1.tbl 14.00 15.00 2023 2180 0.93 1269 1536 0.826
H7_comp_1.tbl 13.70 14.20 1075 1193 0.90 658 795 0.828
K7_comp_1.tbl 13.00 13.50 1277 1276 1.00 449 514 0.874
fname val- val+ # valbar RMS # valbar RMS
J7_repeats_1.tbl 14.00 15.00 93 14.47 0.069 52 14.48 0.076
H7_repeats_1.tbl 13.70 14.20 51 13.97 0.105 25 13.94 0.110
K7_repeats_1.tbl 13.00 13.50 62 13.24 0.095 18 13.24 0.098
Jf20e_repeats_1.tbl 14.00 15.00 83 14.50 0.079 55 14.48 0.098
Hf20e_repeats_1.tbl 13.70 14.20 44 13.97 0.105 24 13.95 0.122
Kf20e_repeats_1.tbl 13.00 13.50 59 13.24 0.123 18 13.20 0.132
Jf20c_repeats_1.tbl 14.00 15.00 83 14.47 0.075 58 14.46 0.089
Hf20c_repeats_1.tbl 13.70 14.20 44 13.96 0.110 31 13.94 0.116
Kf20c_repeats_1.tbl 13.00 13.50 55 13.24 0.113 22 13.21 0.130
RKf0e_repeats_1.tbl 10.00 20.00 30 12.02 1.607 47 13.26 1.706
RKf0c_repeats_1.tbl 10.00 20.00 9 12.42 1.452 35 12.69 1.632
J_ba_repeats_1.tbl 0.20 0.80 28 0.49 0.061 75 0.55 0.075
K_ba_repeats_1.tbl 0.20 0.80 30 0.52 0.068 84 0.57 0.096
S_ba_repeats_1.tbl 0.20 0.80 27 0.47 0.054 77 0.55 0.061
J20e_Jext_repeats_1.0.01 0.80 89 0.32 0.115 104 0.35 0.114
H20e_Hext_repeats_1.0.01 0.80 76 0.31 0.119 83 0.32 0.119
K20e_Kext_repeats_1.0.01 0.80 49 0.29 0.116 52 0.27 0.106

TEST #1

Injecting the Calmon zero-point mags into GALWORKs.

TEST #2

Improved Large Galaxy Processing

TEST #3

Coverage Masks and Off-scan JCAT Improvement

For some examples of the new coverage masks, click here.

The JCAT improvement included less stringent masking from JCAT stars off-scan (adjacent scans). Only JCATS brighter than 4th mag (J) are considered. Results in better coverage. Note the increased number of sources (see table below).

TEST #4

Conversion to SNR Thresholds

GALWORKS V2 used mag thresholds to ignore faint sources. The modification is to use an SNR threshold instead, using a fixed circular aperture measurement. Since a 7" radius appears to be a natural size limit for most galaxies (i.e., faint ones), we compute the SNR from this aperture. The SNR threshold itself should approximately match the original mag thresholds. This turns out to be a value of 5:1.

SNR Threshold: 5:1

Detection Repeatibility
white points are uncontaminated; green points are contaminated; red points are false extended sources (usually double stars)

TEST #5

Removing Known (Large) Galaxy Bias

GALWORKS V2 used a "seed" catalog of known, large (diam>1') galaxies to assist the processor. These sources were processed and passed into the DB regardless of their flux and scores. The new modification removes this bias. These galaxies are now processed the same as non-catalogued galaxies. They are subject to the same star-galaxy discrimination scores.

Note: With this modification, the problem of galaxy blending will only get worse since we now do not force processing on known galaxies. V3, by design, should now include large galaxies that have eaten their neighboring companions.

Big BIG Galaxies

Galaxies are a little bigger to 2MASS now. Previously we resricted processing to galaxies with sizes < 5' or so. Now we work on everything. As discussed earlier, what makes this possible is by using the "scan" background (fit over the background) instead of the coadd backgroud for those coadds affected by the big galaxies. In addition, the working "area" has been increased from 80" radius to 120" radius. Correspondingly, the postage stamps images are now larger for the larger objects, with a maximum size now set at 5'.

Some examples are given below, including M31, M32, M33, M51 and a sextet group of galaxies. The downside of including these galaxies is that there is a lot of "debris" surrounding the galaxies (some examples given below). The reliability is in question near the largest galaxies (see M31).

Improved Star Subtraction

Goal: by improving the subtraction of neighboring stars, the photometry will be more repeatible (and hence more reliable). The modification is subtle, yet complex to carry out robustly. See Improved Ellipse Fitting and Isophotal Photometry in Crowded Fields for a description of the algorithm.

The following examples show the resultant stellar subtraction around galaxies using the new algorithm. The images come from the extracted postage stamp images, where a new modification is to add three new planes showing the field with stars subtracted.

Note that stars nicely subtract from the images. Very bright stars subtract reasonably well, with some residual left over. Neighboring galaxy subtraction is not as good due to the design: the objects are *assumed* to be point sources that follow the PSF (or generalized exp function). Subtraction is usually very good for stars well separated from the nuclear regions. Nuclear contamination (star within a few pixels of the nucleus) represents the most destructive contamination and for which our correction is shakey -- fortunately these cases are rare. Finally, note that for the larger galaxies, there are in some cases 'false' sources that are being subtracted (this is due to the fact that the PIXPHOT star detector is finding chaff within galaxies, resulting in unwanted chaos).

Examples of Sources in the Galactic Plane (glon = 40, glat = 0) Notes: Stellar contamination is brutal for these examples (sources density > 4.1). The method is failing for close nuclear contamination. Some adjustment is needed to minimize destruction of the nucleus.

Examples of Large Galaxies with star subtraction
(warning: these images are BIG!) Notes: The second image represents the star-subtracted image. Since the maximum working area is 120" in radius, the image itself may be larger than the working area. Hence, I have masked away any parts of the image that were not under the star-subtraction jurisdiction. Note also that the subtraction is carried out assuming a point-source profile -- that is, we assume the sources are stars. That is why the subtraction is lousy for actual galaxies (see for example, Group).

Discussion: Overall the star subtraction about large galaxies is working fine. But there are some obvious problems: pieces of big galaxies are chopped to bits by nearby star (and galaxy) subtraction/masking. These "sources" are clearly not being characterized correctly (and there is probably no way to do it right in any case). The second obvious problem is that too many "stars" are being subtracted -- that is, pieces of the galaxy itself. How do we avoid subtracting H II regions and noise bumps on top of big galaxies? One idea: use the PSF-fit CHI^2 to eliminate these sources (which presumably have a large CHI^2) from the subtraction list. When do we apply this criterion? Only with big galaxies? Uniformly throughout the survey? Not at all? In any event, extended sources around bright galaxies should probably be treated as chaff and filtered accordingly from catalogs.

Note: I made some improvements that greatly reduced the number of sources subtracted from large galaxies. Improvs include: apply PSFCHI and local-max criteria for coadds with galaxies > 4', and subtraction of only stars with mags within 2.5 mag of the primary (thus for big bright galaxies, we leave alone the faint stars).

Examples of Galaxies in Abell 3627 (Great Attractor), glon = 320, glat = -5 Notes: The images were generated using the latest version of GALWORKS (V3) that incorporates the new method for star subtraction.

Updated Oblique Decision Trees

The training sets used for the OBDT classifier have tripled in number since the last time I build the trees. It is high time to revisit the classification scheme. New trees will be constructed using training sets with some 160K sources. A new method for combining the parameter space and subsquently derived trees will be carried out and compared with current (V2) OBDT-derived Gscore.

Nominal Results, using the GALWORKS V.2 classifier.
Gscore limit = 1.4.

Abell 3558 Repeat Scans: 2.0 < Density < 3.6
Jlow Jhigh nGT nG C nB R nS nD nT
9.00 12.00 17 17 1.00 0 1.00 0 0 0
12.00 13.00 398 377 0.95 0 1.00 0 0 0
13.00 14.00 1371 1328 0.97 1 1.00 0 1 0
14.00 14.50 1775 1678 0.95 1 1.00 0 1 0
14.50 14.75 937 798 0.85 7 0.99 4 1 2
14.75 15.00 993 824 0.83 20 0.98 9 9 2
15.00 15.50 4311 3388 0.79 321 0.91 237 74 10
---- ---- 9802 8410 ---- 350 ---- ---- ---- ----
Hlow Hhigh nGT nG C nB R nS nD nT
8.50 11.50 14 14 1.00 0 1.00 0 0 0
11.50 12.50 303 303 1.00 0 1.00 0 0 0
12.50 13.50 1954 1890 0.97 1 1.00 1 0 0
13.50 14.00 1734 1582 0.91 8 0.99 5 2 1
14.00 14.25 1010 851 0.84 16 0.98 12 3 1
14.25 14.50 1229 979 0.80 30 0.97 21 5 4
14.50 15.00 3540 2775 0.78 294 0.90 210 76 8
---- ---- 9802 8410 ---- 350 ---- ---- ---- ----
Klow Khigh nGT nG C nB R nS nD nT
8.00 11.00 20 20 1.00 0 1.00 0 0 0
11.00 12.00 442 419 0.95 0 1.00 0 0 0
12.00 13.00 1551 1525 0.98 1 1.00 0 1 0
13.00 13.50 1842 1746 0.95 3 1.00 2 1 0
13.50 13.75 1009 939 0.93 16 0.98 12 3 1
13.75 14.00 1428 1257 0.88 41 0.97 28 11 2
14.00 14.50 3477 2475 0.71 288 0.90 207 70 11
---- ---- 9802 8410 ---- 350 ---- ---- ---- ----

Big Galaxies (M31, M32, M33, M51, M66, M101, Maffei 2) fields
Jlow Jhigh nGT nG C nB R nS nD nT
9.00 12.00 9 9 1.00 5 0.64 5 0 0
12.00 13.00 10 10 1.00 8 0.56 5 2 1
13.00 14.00 51 51 1.00 17 0.75 11 4 2
14.00 14.50 86 84 0.98 23 0.79 18 2 3
14.50 14.75 89 82 0.92 8 0.91 3 4 1
14.75 15.00 95 91 0.96 9 0.91 6 2 1
15.00 15.50 472 440 0.93 70 0.86 48 21 1
---- ---- 812 767 ---- 140 ---- ---- ---- ----
Hlow Hhigh nGT nG C nB R nS nD nT
8.50 11.50 12 12 1.00 6 0.67 6 0 0
11.50 12.50 9 9 1.00 7 0.56 6 1 0
12.50 13.50 82 80 0.98 32 0.71 25 6 1
13.50 14.00 140 134 0.96 24 0.85 15 5 4
14.00 14.25 102 97 0.95 10 0.91 10 0 0
14.25 14.50 108 100 0.93 11 0.90 3 7 1
14.50 15.00 357 333 0.93 50 0.87 31 16 3
---- ---- 812 767 ---- 140 ---- ---- ---- ----
Klow Khigh nGT nG C nB R nS nD nT
8.00 11.00 9 9 1.00 4 0.69 4 0 0
11.00 12.00 11 11 1.00 8 0.58 6 1 1
12.00 13.00 69 67 0.97 24 0.74 15 6 3
13.00 13.50 135 130 0.96 18 0.88 11 4 3
13.50 13.75 110 108 0.98 16 0.87 13 3 0
13.75 14.00 153 147 0.96 18 0.89 11 6 1
14.00 14.50 323 293 0.91