This is a composite view of 3 different methods for removing background sky gradients in an image stack.

The top left image shows the result of stacking 16 frames, each of which had a linear gradient removed before star alignment and stacking integration.

The top right image shows the result of not removing the sky background gradient from each individual image before alignment and stacking, then applying an iterated DBE to the final image. The iterates were finished after the removed background showed less than 1% peak-peak variation, relative to background levels.

The bottom image shows the result of removing an 8th order polynomial surface from each component image before alignment and stacking.

Both of the top images still show faint traces of a colored band near the bottom edge of each image. These bands are not always easy to see in individual finished images. But when several such finished images are combined into a mosaic, these artifacts stand out more clearly where multiple images join together.

I consider anything that shows a geometry related to the frame structure (edge bands, circular domes or troughs) to be residual artifacts in the image, not related to the actual cosmic image. Of course, this judgement must be tempered because it is often the case that actual brightness distributions mimic such artifacts.

I can only guess at this time that such image artifacts are indeed edge artifacts since they seem to appear repeatedly in other image sets. The cause is a puzzle to me. My flats show no such structure and they have been carefully and repeatedly crafted over months of effort.

Perhaps these image artifacts are the result of scattered light from nearby terrestrial origin (i.e. lights in the adjacent house where my wife sits reading). I do not have a light shade on the business end of my C8, and I frequently run into even greater difficulties with images taken at low elevation pointing over the nearby city of Tucson.

These particular frames were taken very near the west of zenith, proceeding down to an elevation of 50 degrees. There are distant street lights to the west. The artifacts are quite faint, and show up only after 180 sec of exposure. But perhaps they are the result of scattered light entering the broad open face of the C8 corrector plate. (??)

But, at least for this particular sky image, it appears that the use of an 8th order polynomial to remove individual frame backgrounds, works best. I can see the background estimate showing the band structure near the bottom edge of the images. And so that gets removed quite nicely as shown in the bottom image. No doubt the use of an 8th order surface would be contraindicated in some other celestial views.

We are dealing here with background structures that have amplitudes very near the noise floor limits of the imaging process. As such, everything with very low spatial frequency becomes suspect.

The only way to really know what the sky truth is, would be to make a wider field of view image that encompasses these edges in the interior of an image with higher signal to noise ratio.

--------------

Of course,... these background variations could also be viewed as arising from variations in the choice of histogram curve stretching. If I forego the final curve stretch after masked stretching to nonlinear, the resulting nebula is to unremarkable and faint.

But I could also perform a sigmoid stretch at the 25% and 75% levels to both diminish the background visibility and stretch the midrange levels to increase nebula contrast. In that case, almost all of the variation between images disappears into the near-black levels in the image, and the nebula and stars stand out more uniformly between the various methods used to remove background sky gradients.

Perhaps that is actually the smarter approach, since we are just basically bouncing noise at the background levels. The difference between the ruddy background level and black is only a few electrons.