stacking difference

[Anthonyexmouth]

Is there any difference in output for image stacking in batches?

say 1. 100 subs stacked together 

2. 2 batches of 50 or 10 of 10 that are then all stacked together. 

[vlaiv]

Depends on stacking method and number of subs in batches.

Look at this for example:

1,2,3,4,5 - average of that is 3 right (15/5)?

Take first two - average is 1.5, take second three - average is 4 - average those two and you get 5.5 / 2 = 2.75

2.75 != 3

Similar thing happens with noise and you get different SNR depending on the way you stack - not to mention sigma clip or quality weighing - that work differently depending on what is being stacked, so sub that is low quality might end up in partial stack (because "surrounding" subs are lesser quality as well so it does not look too poor) while it might get discarded if we stack all together (too low quality compared to average of all stacks).

 

[Anthonyexmouth]

3 minutes ago, vlaiv said:

Depends on stacking method and number of subs in batches.

Look at this for example:

1,2,3,4,5 - average of that is 3 right (15/5)?

Take first two - average is 1.5, take second three - average is 4 - average those two and you get 5.5 / 2 = 2.75

2.75 != 3

Similar thing happens with noise and you get different SNR depending on the way you stack - not to mention sigma clip or quality weighing - that work differently depending on what is being stacked, so sub that is low quality might end up in partial stack (because "surrounding" subs are lesser quality as well so it does not look too poor) while it might get discarded if we stack all together (too low quality compared to average of all stacks).

 

What i'm thinking of is my older data where I haven't saved the flats or can identify what flats go with the subs. I have the masters created in WBPP. So what I'm really asking is, does stacking the masters of historic data compare to stacking all the subs over multiple sessions produce the same super master. 

[vlaiv]

1 hour ago, Anthonyexmouth said:

What i'm thinking of is my older data where I haven't saved the flats or can identify what flats go with the subs. I have the masters created in WBPP. So what I'm really asking is, does stacking the masters of historic data compare to stacking all the subs over multiple sessions produce the same super master. 

I'm sort of confused about what you are actually asking.

Will throw out some thoughts that I think might be applicable, so let me know if I got it right:

- stacking flats from multiple sessions to produce super master does not make a sense since flats are per session calibration files. Unless you can guarantee that your setup did not change between sessions (like observatory pier without taking apart things and with precise repeatable electronic focuser) - you should not do it as dust particles and vignetting won't align.

- stacking darks from multiple sessions is totally doable and, while it will suffer from same things described above - in principle it should not matter as darks tend to have very uniform signal (if cooled to same temperature and having well behaved sensor). My only advice would be to weight final stack of stacks according to number of subs each have. Say that you want to stack one stack made out of 10 subs and another made out of 40 subs. Final stack should be (first * 10 + second * 30)/40. It's a bit like "undoing" initial averages, summing things up and doing global average.

- lights behave as described above and I would avoid making stack of stacks if at all possible.

[Anthonyexmouth]

1 minute ago, vlaiv said:

I'm sort of confused about what you are actually asking.

Will throw out some thoughts that I think might be applicable, so let me know if I got it right:

- stacking flats from multiple sessions to produce super master does not make a sense since flats are per session calibration files. Unless you can guarantee that your setup did not change between sessions (like observatory pier without taking apart things and with precise repeatable electronic focuser) - you should not do it as dust particles and vignetting won't align.

- stacking darks from multiple sessions is totally doable and, while it will suffer from same things described above - in principle it should not matter as darks tend to have very uniform signal (if cooled to same temperature and having well behaved sensor). My only advice would be to weight final stack of stacks according to number of subs each have. Say that you want to stack one stack made out of 10 subs and another made out of 40 subs. Final stack should be (first * 10 + second * 30)/40. It's a bit like "undoing" initial averages, summing things up and doing global average.

- lights behave as described above and I would avoid making stack of stacks if at all possible.

Ok, let me try and explain better. 

1. I'm rubbish at file organisation. 

2. I have lots of old historic data 4+ hour in each session thats been stacked but still have the subs but no linked flats.

Is it worth stacking those stacked masters?

going forward I'd like to think I'll be better with the flats but as I've got a permanent setup I don't do flats for every session, more like bi-monthly, maybe If I bought one of the deep sky dad flat panels to integrate into NINA I might but too expensive at present. 

 

 

[vlaiv]

4 minutes ago, Anthonyexmouth said:

Ok, let me try and explain better. 

1. I'm rubbish at file organisation. 

2. I have lots of old historic data 4+ hour in each session thats been stacked but still have the subs but no linked flats.

Is it worth stacking those stacked masters?

going forward I'd like to think I'll be better with the flats but as I've got a permanent setup I don't do flats for every session, more like bi-monthly, maybe If I bought one of the deep sky dad flat panels to integrate into NINA I might but too expensive at present. 

 

 

Ok, so if I understand correctly now - you have old sessions without flats but with stacked data and with separate subs (which are sort of useless now that you don't have flats, right?), and you wonder if it would be good idea to use those stacked masters with your current data.

Well, I'd say it is very easy to try out. It would be better to stack all the data in one stack - but since you are lacking flats - that is a no go.

Take your current stacked dataset, and perform following:

1. keep one copy of just new data

2. make mix of old and new data in 1:1 ratio

3. make mix of old and new data in several more varying ratios - based on say total imaging time or "perceived" quality - maybe even let software determine weights.

Then simply compare results and select one that looks the best?

[Anthonyexmouth]

2 minutes ago, vlaiv said:

Ok, so if I understand correctly now - you have old sessions without flats but with stacked data and with separate subs (which are sort of useless now that you don't have flats, right?), and you wonder if it would be good idea to use those stacked masters with your current data.

Well, I'd say it is very easy to try out. It would be better to stack all the data in one stack - but since you are lacking flats - that is a no go.

Take your current stacked dataset, and perform following:

1. keep one copy of just new data

2. make mix of old and new data in 1:1 ratio

3. make mix of old and new data in several more varying ratios - based on say total imaging time or "perceived" quality - maybe even let software determine weights.

Then simply compare results and select one that looks the best?

Next period of bad weather I'll sit down and do that. 

What I should do is get into the habit of copying the current flats into the session folder and then learning how the grouping works in Pixinsight so I can stack multi nights and it uses the correct flats for the correct night. I think thats possible anyway.