Don't throw out those "bad" subs!!

[Rodd]

A lot has been said about weeding out bad subs from a stack.   I have done tests like thi son a number of occasions, always with the same result....I end up using the stack with all subs.  Here is an example.  These were taken with the TOA 130 with the ASI 1600.  This is the blue channel.  The first stack has 70 subs and the second stack has 93 subs.  I removed all subs with FWHM over 3.0, eccentricity values that were high, high median values that represented a passing cloud.   The removed subs amount to about 25%, so not a trivial number.  Here are the relavent statistics.  The stack with all subs has a FWHM of 2.78 while the "good stack" has a fWHM of 2.68---only a .1arcsec/pixel difference.  The stack with all subs has a SNR of 18 while the "bad stack" has a SNR of only 14.  Median values are also in favor of th bad stack--though only by .3 (61.7 vs 61.5.  Noise was also lower in the stack will all subs as expected 0.85 DN vs 0.74 DN.

No processing at all has been done to these stacks--no cropping, gradient removal, nothing but an auto stetch after integration.  I will collect another 150 subs or so to complete the stack--but I thought this was a good example, and the intention is to pass on information that may be useful to some.  If not--let it pass by like teh breeze.  I find it very interesting and counter intuitive.  Higher SNR, about the same FWHM, lower median and lower noise--I think the assessment is a "no brainer"

My advise to tose just strting out, or have not thought about things in this fashion before is not to be so hastey to eliminate subs that do not stand up to your preconceived standards.  SNR is very important and a tenth or a few tenths even of FWHM is not very detectable.  Anyway--I hope this proves useful to somebody.  At the very least, I hope it makes you take a second look and enters your decision making process.  Obviously I will toss out a FWHM of 8.5, or a sub that is obscurrred by clouds.  I am just saying iy is not necessary to be a perfectionist--it may very well have the opposite effect you desire.

Blue-70 120 sec subs

 

Blue 93 120 sec subs

Edited September 28, 2019 by Rodd

[Pompey Monkey]

Rodd - I completely agree.

But it all depends on the rejection criteria that you set. As the bulk of your subs are already at a high level of quality, you are still adding useful information to the image.

I do like the comparison between the stacks - it shows a lot. Is it possible to do a GIF or animation of one over the other?

[Rodd]

11 minutes ago, Pompey Monkey said:

Rodd - I completely agree.

But it all depends on the rejection criteria that you set. As the bulk of your subs are already at a high level of quality, you are still adding useful information to the image.

I do like the comparison between the stacks - it shows a lot. Is it possible to do a GIF or animation of one over the other?

That’s true.  But the dubs I tossed had Fwhm values between 3 and 6 with the majority being between 3.5 and 4. When shooting at s resolution of .78 arcsec/pix, that is pretty bad. But yes, I agree.   It depends on many things, such as the amount of data, and the difference between one’s idea of good and bad.  I am more suggesting a process, or at least don’t throw out data as a matter course because it has measurements above your model. I have seen folks decimate a stack because they believe removing many subs will improve things. 

Regarding an animation.  Yes it’s possible but I have no idea how to do it.  Feel free to download the images and throw them into pi’s blink tool, or another software

Rodd