The Sky section of the pipeline allows settings to make cuts on the properties of the sky. We investigate a couple of possible sets of settings to see if tighter settings will improve the quality of the data.
At the Batavia, IL location, the sky brightness on a fair night might be 2000 counts in the V filter and 3500 count in the I filter. That is, a 90 second exposure will produce an image where the median value of the image is 2000 counts above a dark for the V filter. As clouds move in, this number gets larger. Clouds light up the sky from reflected city lights far more than they obscure any light coming from the sky. In fact, full cloud cover saturates the images in both filters. Thus the median sky background is a measure of the cloud/haze/fog cover.
It is not obvious that thin cloud cover harms the photometric computation for a particular star. It clearly increases the background noise, so any cloud cover tends to make the dimmer stars hide in the background noise.
The sky program allows making a fit to the sky background for a dark subtracted and flat fielded image. What remains after dark subtraction and flat fielding can have a gradient due to the city lights in the sky, and other problems due to cloud cover. The Sky program allows an attempt to flatten what remains. See the write up in the pipeline documentation for the procedure. We always use the first order fit. That is a plane is subtracted from the dark subtracted and flat fielded image to attempt to produce a flat image. We find that this works quite well, and the result is flat well within the noise for a reasonable sky. In order to make this fit, the image is sampled in a number of places. Then a plane is fit to the sample points. Several points can be set to allow a good fit. min_accept an max_accept help prevent problems from saturated stars. We set these: min_accept -26000 max_accept -10000
We have chosen 5 as the number of points in the fit. Larger numbers seem to increase the probability that the solution will blow up in some way. Since we are only interested in a planar fit, a large number of points is not needed.
There remain two parameters to set to control the cut.
For this test, we have chosen two sets of parameters.
The Loose parameters were selected on the basis that broader parameters would not result in a significant number of additional stars being measured.
The Tight parameters were selected on the basis that tighter parameters would eliminate almost all star measurements.
To study the effect of these parameters, we processed the data taken during March of 2003 on TOM1 with both sets of parameters. This consisted of about 1600 image pairs acquired on 12 different nights. While it is possible that a star would be measured 48 times, only a few were measured 10 or more times. TOM1 scanned the sky taking 4 x 4 degree images which overlapped slightly. Thus it was possible for a star to be in a four way overlap on each night. This was not a particularly good month for data, and most of the runs had some clouds. There were also some clear sections.
As max_sky is increased, we accept more and images with clouds. Stars can still be seen through the clouds, but as the sky gets brighter and brighter, the dynamic range remaining is reduced. The noise (at least) increases as the sqrt of the sky value. So the difference between an allowed value of 2000 and 8000 for the sky might double the noise. This generally means we detect fewer dim stars.
As max_skysig is increased, we accept more cloud variation across the image. Here the Tight cut corresponds to about the image noise level. The loose cut is several times the noise level. Since only stars 3 sigma above the noise are measured, both criteria are relatively tight.
The Loose cut produced 574150 measurements of 285844 stars.
The Tight cut produced 96788 measurements of 71469 stars.
Figure 1 shows Loose cut coverage for stars measured at least once.
Figure 2 shows the Loose cut coverage for stars measured at least 3 times. One can notice tha boundaries of the 4 x 4 degree frames. It is always more likely to get multiple hits where the frames overlap.
Figure 3 shows the Tight cut coverage for stars measured at least 3 times. Now it is seen that most of the multiple star measurements are on the boundaries.
Figure 4 (Loose) and Figure 5 (Tight) compare the V measurements where each star was measured 3 or more times.
Figure 6 (Loose) and Figure 7 (Tight) compare the I measurements where each star was measured 3 or more times.
It was speculated that there would be some improvement in the Tight data. I can see no obvious difference in the quality of the data. This is good since it allows us to keep more measurements. However, it is hard to make a valid comparison. The Tight data tends to be dominated by the overlap stars. So it is really measuring the ability to get the same measurement using opposite sides of a frame. The Loose data has more stars taken in the centers of the frames in the result. The general shape of the distribution does not seem to change no matter how we select the measurements. An earlier test done where the centers of the images were isolated did not show any improvement.
I can make this data available to anyone that might want to analyze it by other means than looking at the plot.
While this result is somewhat disappointing, it appears to be what we can do in this suburban location. The ASAS 2002 paper gives a sigma of 0.038 for overlap stars. This is similar to the measurement shown here. We hoped to do somewhat better due to our superior optics. Perhaps our terrible location is compensated by the better optics (Our better optics means that a smaller flat field correction is needed compared to the camera lenses used by ASAS).
I can see no reason not to continue processing all the data using the Loose cut definition.