SubframeSelector and light pollution gradient

[Tromat]

Hi folks,

I'm facing a philosophy problem with the weighting of my subframes with the said script (or process now). I'm using a full frame DSLR + 200 f/5 newtonian from my backyard in a not too polluted countryside so I have a rather wide field. When I follow the same target along the night the light gradient from the cities around my place slowy diminishes as the target reach the zenith then slowly raise as it passes the meridian.
When I'm then processing my subs, after calibration and debayer I have to weight the subs and this is where I'm facing the problem. My best frames are supposed to be right above my head and not around the horizon, but ,assuming FHWM and eccentricity doesn't change too much, the SubframeSelector will give more weight to the wrong images as it considers that light pollution is useful signal (quick answer: it's not).
I have the same issue with small clouds that will reflect light pollution and will give more weight to a poor quality subframe.

So, how do you deal with it ? I thought removing gradient using DBE or ABE but it can be complicated with lots of subs and potentially add artifacts. I was also thinking weighting only using FHWM and eccentricity but then again, a small cloud passing by will make the subs have smaller stars and then tricking the selector.

Thanks for your time

[ngc1535]

Hi folks,

I'm facing a philosophy problem with the weighting of my subframes with the said script (or process now). I'm using a full frame DSLR + 200 f/5 newtonian from my backyard in a not too polluted countryside so I have a rather wide field. When I follow the same target along the night the light gradient from the cities around my place slowy diminishes as the target reach the zenith then slowly raise as it passes the meridian.
When I'm then processing my subs, after calibration and debayer I have to weight the subs and this is where I'm facing the problem. My best frames are supposed to be right above my head and not around the horizon, but ,assuming FHWM and eccentricity doesn't change too much, the SubframeSelector will give more weight to the wrong images as it considers that light pollution is useful signal (quick answer: it's not).
I have the same issue with small clouds that will reflect light pollution and will give more weight to a poor quality subframe.

So, how do you deal with it ? I thought removing gradient using DBE or ABE but it can be complicated with lots of subs and potentially add artifacts. I was also thinking weighting only using FHWM and eccentricity but then again, a small cloud passing by will make the subs have smaller stars and then tricking the selector.

Thanks for your time

I can verify that SNRWeight will indeed have this behavior with variation in sky brightness. I had a similar issue with passing thin clouds. The brighter sky with the thin clouds was evaluated as a greater weight. The expectation for a weighting scheme of course is that the noisier and lesser signal should have less weight.

This is why I started down the road of investigating noise evaluation through ImageIntegration without passing SNRWeight to it...but then my questions are summarized in the recent post before this one. (The weights there are not what I am expecting..but I may not be interpreting them correctly.)

-adam

[bulrichl]

There has been a discussion about this topic: https://pixinsight.com/forum/index.php?topic=12131.msg74489#msg74489

Bernd

[ngc1535]

Hi Bernd,

Yes, I saw the discussion- but it doesn't seem to result in a method that takes care of the issue.

I propose:

1. You cannot look at a frame visually and know there are thin clouds (transparency) affecting it. (Or said another way, we shouldn't have to make this arbitrary decision).

These frames will have higher background values- but lower SNR.
 
What method do we use to give us lower weights in the above case?

-----
As an aside,  a compromise is to not use our brains to decide about clouds...but instead use our brains to decide what the background is and what real signal is.
So... if we create a preview that we *know* is only the sky and another preview that we *know* is only the object signal... something much brighter than the sky
so we know the sky contribution is negligible...then we do the analysis (subtracting the sky value and scaling for signal strength... or a noise evaluation version of this
) to get the proper weights? Would it make sense to have this in ImageIntegration?

[bulrichl]

Hi Adam,

I guess if you have a situation like in the plot that I showed (reply #17), where the median is continuously decreasing AND SNR is continuously increasing at the beginnig of an imaging session, you can be pretty shure that no thin clouds are involved. This is exactly the effect of what you described: higher altitude means lower LP. If thin clouds were a problem, the plot would exhibit strongly variable values.

Bernd

[ngc1535]

Hi Bernd,

I am not certain I agree with you. I can think of cases where this doesn't occur. Imaging near the celestial pole (constant airmass) and also (for mid-northern latitudes) tracking a target to the south through the meridian... this would be nearly constant airmass as well.

But I think this is besides the point. Images are generally taken across many nights. One night of relatively constant thin clouds and another photometric night will not reveals patterns that can be easily discerned even graphically.

However, if we choose a fiducial in the image for which we claim- this bit is brighter than the sky (significantly) and then also find a bit of the image with a typical sky value- the noise analysis can move forward with greater certainty. Is there is a method or technique  to do this and properly weight changes in attenuated signal that generally go along with increases in background/sky brightness? Both clouds and light pollution (even the moon can be an issue) are almost always part of the problem.

-adam

[bulrichl]

Hi Adam,

usually I can discern thin clouds by the halo that is produced around stars. I sort out affected frames by visual inspection. Maybe that this approach is not always feasible with larger focal length (I use a scope with fl of only 530 mm).

> I am not certain I agree with you. I can think of cases where this doesn't occur. Imaging near the celestial pole (constant airmass) and also (for mid-northern latitudes) tracking a target to the south through the meridian... this would be nearly constant airmass as well.

Even with constant airmass, fluctuations of median and SNR in opposite directions would be clearly visible.

> But I think this is besides the point. Images are generally taken across many nights. One night of relatively constant thin clouds and another photometric night will not reveals patterns that can be easily discerned even graphically.

I agree.

> However, if we choose a fiducial in the image for which we claim- this bit is brighter than the sky (significantly) and then also find a bit of the image with a typical sky value- the noise analysis can move forward with greater certainty. Is there is a method or technique  to do this and properly weight changes in attenuated signal that generally go along with increases in background/sky brightness? Both clouds and light pollution (even the moon can be an issue) are almost always part of the problem.

I like your idea with user-chosen previews, one for "pure background" and one for a part of the frame with "real signal". Perhaps, evaluating the same previews in registered frames of a series could yield a more reliable trend of SNR values. However, I admit that I don't understand Subframe Selector's current evaluation of the SNRWeights.

Perhaps Mike Schuster could comment on that?

Bernd