Comparison of Second V3 Ops Test to V2 Processing for 2MASS Galaxies
T. Chester and T. Jarrett Conclusions as of 9/10/01:
BOTTOM LINE: we have found no problems that impede declaring galworks ready to go, and using the current v3 ops db as the true beginning of the reprocessing.
Summary of analysis results:
- V3 produces about 1.4% more sources in low density areas that are brighter than k20fe = 13.5 mag. The new sources are largely good sources, as verified by visual examination by TJ. These extra sources are preferentially due to large galaxies, due to improvements in the V3 position for such galaxies. Some of the new sources result from the V3 implementation of star subtraction to improve extended source extraction.
- V3 sources have substantially the same position and aperture flux as in V2. The k20fe mags are systematically fainter by 3-4.5% as a function of the size of the galaxy, due to incorporating the actual calibration, which results from slightly smaller, but more accurate, radii to that isophote. These are all good things.
- V3 goes deeper into the galactic plane, again a good thing, as intended, since all the sources produced by GALWORKS in the plane can be checked manually.
- There is a "dupe source bug" in galworks that is of no consequence, being fixed by the dupe removal processor.
We intend to complete the comparison of v3 and v2, dotting all the i's and crossing the T's above, as well as do an analysis of how many v2 sources are still found in v3. (A database bug prevented an accurate comparison of the entire V3 ops db with the V2 processing until 8/28.) This analysis can proceed during the ops processing.
The above does not include the analysis of the Abell 3558 repeats, done by T. Jarrett, which also shows no problems with v3.
Change in Number of Sources from V2 We have begun our comparison of the v3 processing to the v2 processing by analysis of the second v3 ops test, processed in August 2001. See Preliminary unfinished analysis of the first v3 ops test for some analysis that has not been repeated.
For this analysis, we have analyzed all sources, galaxies plus other extended objects, that have k_m_20kfe < 13.5 mag. In order to unambiguously match sources with their v2 counterparts, we selected sources away from any coadd edge via:
50 < x_coadd < 460 and 50 < y_coadd < 974. And we also demanded that the sources be clean sources, with
j_flg_k20fe = h_flg_k20fe = k_flg_k20fe = 0
A total of 24,776 sources satisfied these constraints. Of those, 24,145 found matches in the v2 wsdb from the same hemisphere, date and scan with positions within 5" of the v3 position. Thus 97.5% of the sources are unambiguously matched in the v2 processing.
A plot of the percentage of unmatched sources vs. night shows that the previous problem of having some nights without matches in the v2 WSDB has gone away. Apparently, this was due to a non-current file in IRSA that was used for table compares. The night with the highest percentage of new sources is a night where nearly all the scans are at very high densities.
A plot of the number of unmatched sources vs. density shows a smooth distribution. Low-density scans, density < 3.0, have 1.4% new sources. The percentage of unmatched sources increases for scans with higher densities above 3.5, becoming over 50% at the highest densities. This is an expected result of the V3 processing, which was designed to go a bit deeper into high source density regions than was done in V2.
Selecting only the sources from scans with densities less than 3.3 shows that the number of unmatched sources depends on magnitude. This was expected as a result of a change in how GALWORKS treats galaxies with large optical diameters. Previously, GALWORKS used an optically-derived position, which in many cases was off significantly from the true galaxy position, resulting in a flux that was too low or even below the threshold to enter the V2 WSDB.
T. Jarrett has examined some of the new sources, and in nearly all cases they are true extended sources. The new sources fall into three main categories: some are now present due to the change for large galaxies, some appear due to the lowered thresholds at high source density, and some of them result from the new process of star subtraction. This analysis shows that all three software changes appear to be working as intended.
In previous analysis, we found that two sources each found two identical matches in the V2 WSDB. Tracey Evans tracked these down for us, and found that these sources were output by GALWORKS from the same coadd in the same scan, with identical positions. Similar sources have been found by Tracey in the current processing. T. Jarrett found that they result from a close pair of detections presented to GALWORKS, each of which produces an extended source too faint to be masked in further processing. Thus this close pair of detections generally produces two identical sources in the Extended Source WSDB. Fortunately, the in-scan dupe processor kills one of the pair, so this bug is harmless.
(add analysis of v2 sources in v3; specifically, if any are missing.)
Photometry Of the matched sources, we discarded sources with null fluxes from the v2 processing (19 sources), and selected only sources with matches within 1" (22,515 sources), necessary to provide good photometric comparisons. Thus 93% of the sources are used in the following photometric comparisons.
For the V3 processing, four changes were made that affect photometry directly:
- The maximum radius for the k_20 fiducial isophotal radius was no longer limited to 80".
- The minimum radius for the k_20 fiducial isophotal radius was decreased from 7" to 5". This will result in fainter v3 magnitudes for the smallest sources.
- V3 GALWORKS now subtracts stars, whereas in V2 GALWORKS only masked stars. This should result in improved results for galaxies that were severely compromised in V2.
- V3 GALWORKS processing uses an improved calibration to define the 20th magnitude isophote. V2 was forced to use an isophote defined using a fixed calibration which did not take into account nightly changes in the calibration, because the nightly calibration was not available at the time the processor was run. For V3, GALWORKS uses the actual calibration derived from the V2 processing.
We therefore first analyze the change in the k_20 fiducial isophotal radius. A plot of the v2 radius vs. the v3 radius shows that, as expected, several sources now have a radius larger than 80" and many sources now have a radius of 5-7". In addition, a few negative and zero values for the V2 radius are reasonable values in V3, and some V2 minimum radii are now much larger radii in V3. This is presumably an effect of the star subtraction in V3, which produces reasonable values of the radii now compared to minimum, zero or negative values in V2.
Surprisingly, a plot of the change in radius from v2 to v3 vs. the v3 radius shows that there is a slight mean change in the radius for most sources, with the V3 radius being smaller. This turns out to be a result of the incorporation of the improved calibration, since there is a slight non-zero bias in the average nightly calibration change from the pipeline assumed default calibration.
A plot of the median change in radius shows that the change in radius is largest for very large sources, and becomes smaller as the radius of the source decreases, until a radius of 7" where it increases again due to the changed minimum value. This is caused by the mean calibration shift interacting with the steepness of the source profile.
This change in radius leads to the V3 isophotal magnitudes being fainter by about 3%, with the smallest sources being fainter by about 4.5% due to the decreased minimum radius.
In contrast, the radius = 7" aperture magnitudes are essentially unchanged, with a median difference of only 0.001-0.002 mag.
(add back-of-the envelope calculation of the k20fe photometric effect.)
One change was made in the computed photometric errors for isophotal magnitudes in the V3 processing: incorporating the additional uncertainty due to the error in the radius determination. Aperture magnitude uncertainties are unchanged.
Both theoretical analysis and testing had shown the need for the inclusion of this error source, which was implemented in V3.
As expected, analysis of the ops test shows that the photometric errors for the 7" aperture magnitudes are essentially unchanged, but the errors for the 20th fiducial isophotal magnitudes have increased:
Selection J H K 7" Aperture - V2 J H K 7" Aperture - V3 J H K 20th isophotal fiducial elliptic - V2 J H K 20th isophotal fiducial elliptic - V3 J H K The Abell 3558 repeat analysis shows that the error estimates now track the observed repeatability.
Positions (to be redone with complete v2 wsdb; Preliminary unfinished analysis of the first v3 ops test found no problems with position.)
