Results from Tests using Tycho-2 in Pipeline
H.L. McCallon 04-25-00
Using Tycho-2 as the reference star set in PosFrm has been tested in the 2MAPPS pipeline for two nights, one northern (980304n) and one southern (980727s). These nights were selected because they had an unusual number of problem scans. Problem scans generally result from a lack of ACT stars in terms of count and/or distribution within a scan. They have been previously discussed with regard to 2MASS/Tycho-2 differences, as well as 2MASS/UCAC differences. The night of 980727s was run twice, once with Tycho-2 uncertainties as stated in the release and then again with uncertainties of 50 mas and 150 mas for stars originally in Tycho-1 and new to Tycho-2, respectively.
Based on overlap differences, "problem scan" improvements with the use of Tycho-2 are comparable for the northern and southern nights. Given that the newly released UCAC is available for the southern night, differences w.r.t. that catalog can be presented. Results for each scan are shown as a four-panel plot. In each case mean 2MASS-UCAC differences are plotted as a function of Declination in the left panels. Original differences "using the ACT" are plotted with dashed red lines and differences "using the Tycho-2" are plotted with solid black lines. Histograms of position differences are presented in panels to the right with the same color coding. Cross-scan differences are plotted in the upper panels and in-scan differences in the lower panels. Comparisons for the 980727s111, 980727s119, 980727s120, 980727s132 and 980727s137 problem scans show strong improvement.
It should be pointed out that the expression "using the ACT" here refers to having the ACT as the high-accuracy (low-density) reference catalog in PosFrm. When the high-accuracy catalog stars do not meet the minimum number/distribution requirements, PosFrm instead reconstructs with the higher density (lessor-accuracy) USNOA catalog. Three of the scans listed above (980727s111, 980727s119, 980727s137) fell into this category when the ACT was the high-accuracy catalog. All scans from both nights tested with Tycho-2 as the high-accuracy catalog more than met the minimum requirements. The remaining problem scan 980727s098, which also did not meet the requirements for an ACT reconstruction but did for Tycho-2, shows improvement in-scan, w.r.t. the original USNOA reconstruction, but walks off somewhat in x-scan at one end.
It is also of interest to consider 980727s080 which was sub-par, but not bad enough to make the problem scan list. This scan suffers from a fortunately infrequent anomaly which allows the the reconstruction to systematically move away from the reference stars for a period as demonstrated by the standard reference star residual plot for the scan. Star symbols plot Tycho-2 (also in Tycho-1) residuals; triangles plot Tycho-2 (not in Tycho-1) residuals and the tiny points represent USNOA residuals. Note that despite a good density of Tycho-2 stars, the solution deviates somewhat in in-scan between -34 and -35 degrees declination. A fair amount of effort has yet to reveal the source of the problem. It must be some kind of PosFrm bug, anomalous data coming into PosFrm or some combination thereof. Although increasing the number of reference stars doesn't help, it's encouraging to note that the Martinizing planned for version 3 should remove it's effects.
Overall statistics confirm that using Tycho-2 provides a marked improvement in reconstruction accuracy as demonstrated by the pre/post overlap difference histograms for SNR>20 presented for the night of 980304n and the night of 980727s. These are presented as log histograms to emphasize the improvement in the tails. The new Tycho-2 results are plotted in solid black lines and the old ACT results in dashed red. Pre/Post differences w.r.t. the UCAC for the southern night are also histogramed. The improvement seen here, while not quite as dramatic as with the overlaps, is still quite obvious. It's possible that applying an SNR=20 cutoff to the UCAC difference data would show further improvement. Comparisons between the two methods of modeling Tycho-2 uncertainties showed negligible differences in the 2MASS reconstruction accuracy.
Based on these results, I recommend we switch production to the Tycho-2
version of PosMan previously delivered for test.
http://spider.ipac.caltech.edu/staff/hlm/2mass/tyc2test/tyc2test.html
Comments to: Howard McCallon
Last update: 28 April 1999