Does NSG benefit from using an integrated reference? Or does it work just as well with (or even require) a single best sub?
Is it worth averaging several good images together to make a 'super' reference?
(1) Lets look at the accuracy of the (brightness) scale measurement first:
This is a useful test data set because it covers a wide range of conditions. The imaging run started in twilight, which adds a large light pollution gradient. Between image 9 to 15 the sky was as dark as it was going to get. After image 15, the sky started to brighten. After image 17, the images were taken through thin cloud. Of course twilight does not change the sky transparency, so we only expect it to increase the noise. The scale should not be affected. The thin cloud after image 17 does affect sky transparency, so we expect the measured scale to go decrease.
The blue line (NSGS0) is the scale calculated by NSG. I checked the accuracy by using the photometry program APT (used by professional astronomers) on the unregistered data. See the red line. The APT standard deviation was an impressive +/- 0.003 The NSG accuracy was +/- 0.005 The NSG scale measurement is far more accurate than is required, so averaging together images is unlikely to have any benefit.
(2) Gradient subtraction
It is likely that any single reference image will contain a gradient. Provided that this gradient is smooth, it will not usually cause a problem. Even if a smooth gradient is large, a smooth gradient can easily be removed using DBE after stacking. Remember that normalization is only designed to remove the relative gradient between the reference image and the target images.
If all images contain a very lumpy gradient, then it might be worth being a bit more creative. There are several options:
- The best option is to take a single image on a particularly dark, moonless, clear night when the object is high in the sky. Use this as the reference.
- Alternatively, apply more smoothing to the gradient correction. The gradient trend will still be corrected, and the 'lumpiness' might average out.
I would not recommend averaging several images together to create a super reference, because if the images were dithered, you may end up with 'staircase' artifacts at the edges of the averaged image. This will adversely affect the gradient model.
DO NOT apply DBE to the individual subs. The black areas around the registered images may become non zero, which will also affect the gradient calculation.
So, in conclusion, you should use a single sub as the reference image.
Hope this was useful,
John Murphy