RealSky/DSS Copyright

Hi, folks,

Here is some interesting news on RealSky, the 100X compressed version of the Digitized Sky Survey available through ASP. For the past few months, I've been measuring galaxy positions in the north equatorial polar cap. Finding lists include all CGCG, MCG, and UGC galaxies, plus a few random objects from other sources (Markarian, Lick, UGC Notes, Zwicky's compact galaxy catalogue, etc.). Altogether, I've measured positions for just over 1000 galaxies in the 27 POSS1 fields in the +90, +84, and +78 zones. I was curious about the accuracy of the positions, of course. RealSky has a minimum pixel size of about 1.7 arc sec, it uses an adaptive compression algorithm (the more information in a given area of the plate, the less the compression), the astrometric solutions are not the same as were used for GSC (the reference catalogue, the AGK3, is the same, however, at least north of the equator), and the software I used is that included in the package with no special modifications (the Win95 version of RealSkyView from Software Bisque ).

How do all of these factors affect the positional accuracy? Going through the Lick Northern Proper Motion reference galaxy list, I found that 323 of the galaxies there are also included in the list that I measured. The differences in arc seconds were easy to find, of course, as were the mean differences, the errors in the means, and the standard deviations. Here are the results: Mean del RA = -0.20 +- 0.06, std dev = +- 1.06 arc sec. Mean del Dec = -2.44 +- 0.06, std dev = +- 1.10 arcsec. Since the known std dev in the Lick positions is 0.3 arc sec, 1.0 arc sec is essentially the std dev of the RealSky positions. Formally, if we take sigma_(RS-Lick) = 1.08 arcsec, then sigma_RS = 1.08^2 - 0.30^2 = 1.04 arc  sec. All this assumes a Gaussian distribution of the differences, of course.

The big surprise to me was the small standard deviation. Since the minimum pixel size in RealSky is about 1.7 arcsec, I was expecting a std dev of around 2-3 arcsec. On reflection, I should not have been surprised for at least two reasons:

1) Since a RealSky field can be magnified to the point where an individual pixel can take up a significant part of the display window (perhaps even the whole window; I've not tried), the center of that pixel can be seen very easily. The RA and Dec of the pixel then depend almost entirely on the astrometric calibration of the field, and not on pointing error. The displayed position is almost always the same no matter where in the pixel the cursor points.

2) Assume a Gaussian probability that the center of any pixel is actually at the equatorial position assigned to that pixel. Then, the expected standard deviation is about 2/3 the pixel size. In this case, that is 2/3 of 1.7 arcsec, or about 1.1 arcsec -- assuming a good astrometric calibration, of course. Lots of assumptions, but the numbers agree nicely. A couple of final comments about the standard deviations. The value I found is probably close to the minimum that anyone can find for RealSky positions, even for faint stars. Here's another way to see that: the Lick galaxies were specifically chosen to have nuclei that appeared essentially stellar on the 20-inch astrograph plates. These same galaxies still appear almost stellar in RealSky, and have nuclear regions contained within just a few pixels at most. This is just another way of saying that the measured standard deviation should agree with the expected value. However, for larger, brighter galaxies with bright central regions, the nuclei are not seen, being lost in the clearly saturated cores. For those brighter objects, the accuracy will drop considerably. Finally, there is a mean offset between the Lick and RealSky positions of nearly 2.5 arc sec in declination. I do not know yet where this comes from, nor do I know if it is confined to the polar regions. There is no indication that it is any smaller at +75 deg than it is at +88 or +89 deg, however, so I'll have to do some more tests further south to see if it remains at the same value.

Whatever happens there, though, the difference is statistically significant for these northern-most fields, so the declinations in RealSky need to be increased by 2 arcsec to bring them into agreement with the system of the Lick reference galaxies. In summary, it looks like RealSky can be used to find positions that are fully adequate for identification of anything that can be seen in it. I am not going to suggest that these positions represent "real" astrometry, but they are clearly the result of careful work by the folks at STScI and at Software Bisque, who respectively provided the astrometric calibrations, and the Windows 95 software used to retrieve the images and measure the positions. All in all, I'm quite pleased by the results, and will not hesitate to use RealSky for my work with NED as well as for the NGC/IC project.

Best to all,

Harold