169 replies to this topic

[t_image]

Aside from the annoyance of the ruined subs, I thought there was another phenomena of making space travel extremely dangerous due to the amount of material in these shells around earth?

 

https://youtu.be/yS1ibDImAYU

So I liked ~90% of the video. And I'm all about raising the alarms for avoiding the Kessler syndrome.

But I think it is important to get the facts correct so we can evangelize effectively.

The idea that even if a complete cascade of collisions happened in LEO,

that there would be some impenetrable wall where we couldn't do space travel is a little hollywood,

just like all the scenes where the heroes are flying through an asteroid belt in a spacecraft and has to dodge objects quickly every few seconds.

That's not how it works. That's not how any of this works....

 

There are consulting firms as we speak war-gaming this with brilliant math models....Consider the Kessler syndrome assumes innocent commercial objects. Nation states have modeled the weaponization of Earth orbits and all out destruction of assets since the 1980s.....

 

For more best info on the situation,

see here:

http://www.spacesafe...ssler-syndrome/

 

and here:

https://sma.nasa.gov.../orbital-debris

[t_image]

re:Frank

I don't think the odds are significantly impacted by Starlink in that case.

I thought the conversation is not just about one constellation but the momentum, and about all types of AP, however thoughts on specifics is fair....

My point with the video of the 3 crossing the two is about statistical chances.

Without the numbers and modeling the maths, do you think you could closely estimate the likelihood I could capture satellites in formation crossing the pass of two other objects in parallel flight. How many hours and how difficult do you think it would take, of tries before I captured something similar?

I can say anecdotally it was easy for me...

There are psychological phenomenons like "confirmation bias" we must guard against when we hypothesize likelihoods of events..

The issue with adding lots of new stuff up there has impact of statistical probabilities and that is my point, not on ordinary circumstances, but 'perfect storms." Fair enough to deconstruct my appeal to absurdity situations, but the concept stands. Sure as I said there are workarounds, but you can't tell me there won't be situations where photobombs won't make somebodies life more a challenge.

And without crunching numbers of statistical chances (beyond my scope), it is problematic to speculate on probabilities...

 

However I do appreciate the positive trend of this conversation to highlight problem solves to anticipated issues.

It is toxic to only complain, but conversing on strategies to deal are a great add, thank you!

 

And I've already made the point but will repeat (for all instances of AP), one way to avoid the photobombs is to plan ahead, since especially these communication mega-constellations will allow easy anticipation of their passes, whether or not they will be illuminated, and the precise timing of their locations in the sky during the pass.....

[David_Ritter]

I for one will not be planning ahead to avoid these things.

 

Most of the time I simply cannot plan ahead. I can only image when the conditions permit. Which is not very much of the time.

 

Also, this is a hobby, not something I need to do to survive. So if I do keep imaging and find that a high percentage of my frames have dozens or hundreds of trails in them, well forget it then. I have better things to do with the years I have left than look at piles of ruined pictures.

 

And finally, I am simply unable to collect the large numbers of frames I believe that I would need to find the good pixels left among the millions (or billions as the case may be) of bad pixels. Maybe people living further south in an arid or desert place with long nights will be able to take enough frames to work around this problem. But I do not live in such a place. Which means that I would be among those bearing the brunt of this problem.

 

I think the impact deniers should take a hard look at the few images we already have of these things. Then do the logical thing and then honestly extrapolate it to what our frames will look like once there are tens of thousands or millions of these things zipping all over the sky.

[freestar8n]

You really need more than just two images. You need to be able to robustly locate a mean in order to reject outliers. And to call the image vdb shared a non-issue, consider when there are not just 60, but 12,000 of these things up there, on top of the thousands from OneWeb, and eventually thousands to tens of thousands more other satellite swarms on top of that. You honestly do not think that multiple frames will be criss-crossed with bright satellite trails all the time, often overlapping at various locations in the frame? 

For the lsst, it is already set to take two exposures at each location and then deal with objects that will end up in the frames.  They wouldn't use normal stacking rejection and instead will need to identify trails and cosmic rays by morphology - and then mask off those pixels.  The field is 3.5 degrees and the exposures are 15 seconds - so I don't think the result will be anything like the long exposure images shown with multiple parallel lines going across the field.  There would mostly be partial trails, depending on how high up in the sky it is looking.  If any pixels are hit in both of the two images, then those pixels would have no info.  But that would still be a small fraction of the 3.2 billion pixels in each target image.

 

And keep in mind the images being shown are of a train of satellites that has just been released - and not the final mesh after they have all been spaced out.

 

Frank

[freestar8n]

re:Frank

I thought the conversation is not just about one constellation but the momentum, and about all types of AP, however thoughts on specifics is fair....

My point with the video of the 3 crossing the two is about statistical chances.

Without the numbers and modeling the maths, do you think you could closely estimate the likelihood I could capture satellites in formation crossing the pass of two other objects in parallel flight. How many hours and how difficult do you think it would take, of tries before I captured something similar?

I can say anecdotally it was easy for me...

There are psychological phenomenons like "confirmation bias" we must guard against when we hypothesize likelihoods of events..

The issue with adding lots of new stuff up there has impact of statistical probabilities and that is my point, not on ordinary circumstances, but 'perfect storms." Fair enough to deconstruct my appeal to absurdity situations, but the concept stands. Sure as I said there are workarounds, but you can't tell me there won't be situations where photobombs won't make somebodies life more a challenge.

And without crunching numbers of statistical chances (beyond my scope), it is problematic to speculate on probabilities...

 

However I do appreciate the positive trend of this conversation to highlight problem solves to anticipated issues.

It is toxic to only complain, but conversing on strategies to deal are a great add, thank you!

 

And I've already made the point but will repeat (for all instances of AP), one way to avoid the photobombs is to plan ahead, since especially these communication mega-constellations will allow easy anticipation of their passes, whether or not they will be illuminated, and the precise timing of their locations in the sky during the pass.....

Well - one way to answer is to find an image - any image - that shows three satellite trails crossing at the same point.  You can't calculate meaningful odds without knowing how the different orbits are distributed - as an ensemble.  But if you have an image with 10 million pixels and three trails in it - and you assume two of the trails cross somewhere and you want the odds that the third one crosses that particular pixel also - I think the odds would be on the order of less than one in sqrt(10M), or one in 3000.  So it will definitely happen at some rare rate - but even when it does happen, you have a small region of pixels that has three hits in it - which isn't a big deal. 

 

For the lsst that would mean no data in that small region, but for a stack of many images - it just means that small region of overlap pixels will lose three exposures of info - assuming the trails happen in separate exposures.  If they all happen in the *same* exposure - that's great because it only means one frame is rejected - for those particular pixels in the stack.

 

Frank

[chadrian84]

Spitballing here, but there are 41,253 square degrees in the sky (360 sphere).  That's 2.5 satellites per square degree on average if there are 100k in the sky - 10 per frame if your FOV is 2x2 degrees.  If it gets to 1 million it'd be 100 per frame at 2x2.  I have no idea how many of them will actually show up in images though.

Edited by chadrian84, 30 May 2019 - 08:49 PM.

[t_image]

I posted in the Light Pollution forum's Starlink convo a simulation I made of just the first orbital shell (66sats in 24 planes at 550km)=1,584 satellites and their illumination by the Sun over a summer night.

 

The issue with AP is images go a lot deeper than the visual experience of the satellites. So to avoid cross-posting I'll just reference you there.

 

https://www.cloudyni...-sky/?p=9400774

[t_image]

Just as an example of what to expect from just the 60 up and when they are finally place at 550km with an equal separation,

I used a simulation to find a random convergence of a lit pass and a familiar target from a random location in the U.S.

Here is a 5 minute "exposure" where consecutive satellites in the 53 degree inclined orbit will pass through the same area,

thus creating 4 streaks in the same image.

Note this is to predict the occurrence with just the 60 up right now.

The point to note is the satellites will be evenly spaced in orbital rings (planes).

Because of the movement of the Earth/thus stars under the orbit,

the same orbital plane will slowly drift across the same starfield, thus allowing streaks from consecutive satellites to be parallel but spaced accordingly to time interval separation....

Now when a few orbital planes are populated, and a target happens to fall within the intersection of two orbital planes for the location/date/time,

there will be a criss-cross set of streaks in a long exposure.

Depending on alt/az location in the sky (timing interval between each satellite will shrink closer to horizon across a FOV).

 

I guess AP'ers could intentionally plan this out for a "fun" effect. Ready, set, go!

 

[BenKolt]

I didn't see this thread before.  I don't want to post the picture again, but here's one of my frames from last night through which some of the swarm passed.  It wasn't as bad as the Lowell Obs shot when the satellites were closer packed together, but I still caught several together during a 120s exposure.

 

https://www.cloudyni...hood/?p=9404488

 

Ben

 

 

 

 

[Hubbs]

When I heard about this, I was also concerned about the impact on astrophotography.  So far I have been fortunate enough that it has been rare to get a satellite trail in a light frame, to the point that if I get one, I have just not used the frame.

 

I went back over some old data that I had, and tried including light frames that had satellite trails using Kappa-Sigma stacking mode in DSS.  Given the possible impact of Starlink, I tried creating 2 stacked images, one using 5 frames (one with a satellite trail and 4 with no trail) and again with 10 frames (1 with the trail and 9 without).  The trail still appeared in both my final stacked images. 

 

Are there other settings I would need to change in DSS to have it remove the trail?  I left my Kappa value at the default (2.0) with 5 iterations...  I really hope so, because from the sounds of how many of these satellites will be going up, I was thinking that 10% was a conservative estimate of how many frames might have trails (especially for relative newbies like me who are using a short refractor with a wide field of view)...

[chadrian84]

I’m surprised stacking programs don’t have methods to detect and reject sat/airplane trails in single frames. I’m confident this is possible, not too difficult, and will be coming soon. OCR (optical character recognition) programs do amazing things such as converting handwriting to text and deciphering CAPTCHA security boxes. Detecting straight lines should be a piece of cake.

Edited by chadrian84, 02 June 2019 - 10:51 AM.

[Jon Rista]

Programs DO have rejection algorithms that can reject satellite trails. You just need enough data for these statistical processes to be effective. And the more trails you have, especially if there is the potential for them to stack up on top of each other or overlap, the more subs you will need for the algorithms to be effective. With a trail or two every few frames out of a stack of dozens of frames, it is no problem. If you have trails in most frames, or all frames, especially if they are overlapping each other in ways, then it becomes a more complex problem. Time will tell, I guess, how these satellites actually affect our subs and how effective existing rejection algorithms are.

[chadrian84]

Programs DO have rejection algorithms that can reject satellite trails. You just need enough data for these statistical processes to be effective.

Why should it require more than two frames though?  If a trail is detected in one frame and not the next, don't include pixels from the trail in the calculation.  It's just a matter of creating an algorithm to detect trails and reject pixels affected by them.  Using statistical methods to find outliers works but only when there are several subs and rejection settings are properly applied - and those rejection settings might not be ideal for parts of the image not affected by trails.  In any case, I agree with those who've said even large numbers of trails will easily be removed with software.

 

@Hubbs:  I don't know what settings in DSS might get rid of your trails.  Hopefully someone can answer your question.

Edited by chadrian84, 02 June 2019 - 01:38 PM.

[Jon Rista]

Why should it require more than two frames though?  If a trail is detected in one frame and not the next, don't include pixels from the trail in the calculation.  It's just a matter of creating an algorithm to detect trails and reject pixels affected by them.  Using statistical methods to find outliers works but only when there are several subs and rejection settings are properly applied - and those rejection settings might not be ideal for parts of the image not affected by trails.  In any case, I agree with those who've said even large numbers of trails will easily be removed with software.

 

@Hubbs:  I don't know what settings in DSS might get rid of your trails.  Hopefully someone can answer your question.

If you have only two frames, which frame has the valid pixel? How do you determine that without an intellect? You need to use statistics, and statistics requires data, enough data that you can determine a best guess at what the mean should be, so you can then identify what pixels are likely outliers. 

 

Now, if every frame is riddled with outliers, and many of those outliers cross or overlay each other, if the loci and paths at which they cross become common across frames...then being able to statistically identify and reject outliers becomes more of a problem. If you have ever imaged around M42, you will understand the issue. M42 is right near a geostationary satellite highway and rejecting satellite trails there, even though they do not always occur on EXACTLY the same pixels, is very problematic and fully removing the artifacts those trails leave from the image is sometimes a challenge (depends on just how many you pick up in your data).

[chadrian84]

If you have only two frames, which frame has the valid pixel? How do you determine that without an intellect?

A trail identification algorithm would determine if a pixel value is valid -- reject it if it falls within a trail.

[Jon Rista]

A trail identification algorithm would determine if a pixel value is valid -- reject it if it falls within a trail.

How do you determine what a trail is, though? Does such an algorithm exist? And further, during integration, you operate on pixel stacks. You are not operating on "images"...so, there is no spatial information to work with. You operate on a "vertical" series of pixels that all correlate to the same exact location, but acquired at different points in time. So, there is no "trail" to be identified when integrating an image.

 

Currently, image integration programs utilize simple statistical algorithms to identify and reject outliers. They locate the mean, determine the (optimized, possibly) standard deviation, then allow you to reject anything some number of sigma from that mean. Sigma, Kappa-Sigma, Winsorized Sigma. There is also Linear Fit rejection if you use PixInsight. These are not identifying trails, though. They are working with the statistical distribution of samples for a single pixel at a time...the locus around which these samples are distributed, their dispersion, and they identify those that are more widely dispersed than the majority. With two samples, you cannot do this. With three, you might be able to do something, but the results are likely to be skewed. With many samples, you can more accurately locate the mean, more accurately compute sigma, and more accurately identify which pixels actually fall well enough outside of that dispersion to actually qualify as an outlier...in that particular vertical pixel stack.

 

You then repeat this process for each and every pixel location. There are no "trails" to be identified and removed. You might be able to do some kind of large scale rejection on each individual frame before integration. PixInsight supports this with Local Normalization. Local normalization is not really intended to remove trails, though...it is a normalization process that is intended to perform normalization at a scale less than the whole entire field (which is what image integration already does), but it can be very difficult to apply properly, and without proper application it can introduce artifacts rather than eliminate them, or trade one set of artifacts for another. 

 

Further, these are all tradeoffs we have to make to get a good image. Local normalization in PixInsight is NOT a fast process by any means, and if you aim to do it right you are usually stuck running multiple passes until you find the settings that work. If we have to do some kind of large scale pre-processing to clean up our frames, especially something that could add a significant amount of time to our processing, BEFORE integration, that right there indicates how severe the problem is of having so many satellites in LEO. I have done hours-long pre-processing in the past. I've done local normalization. I've integrated hundreds to thousands of frames with lots of trails in them. I don't want to have to spend that kind of time processing my images...I hardly have the time to process the data I have the way I process now, and I've spent quite a lot of effort optimizing my routines to give me clean results at a minimal time investment.

 

These satellites will be a problem. Again, it isn't just about StarLink...StarLink is just the herald of what is to come. There are going to be tens of thousands of satellites in LEO. Right now, there are 4900 satellites in ALL of earths orbital bands...which means there are fewer than that in LEO. We are adding tens of times the number of satellites we have now up there. It is going to be a problem... I just hope it doesn't end up a disaster...as at least as long  as they can control these things, they could eventually guide them to burn up in the atmosphere (and I can only hope that at some point we realize our folly here and try to correct it...) Once we experience Kessler syndrome, though...tens of thousands of satellites become millions upon millions of pieces of debris, some of it small, some larger chunks, all uncontrolled, and a lot of which we will be able to see in our photos. Kessler syndrome is really what I fear...I see it as an inevitability if we put so many satellites up there (especially since a lot of this is mostly billionaires playing popularity games here), and once it happens... Thousands of satellites will be the least of our problems...

[chadrian84]

No, trail identification algorithms haven't been created by anyone as far as I know - that's been the point of my posts... it's possible and shouldn't be difficult.  As mentioned - for well over a decade software has been able to convert images of handwriting to digital text and even crack CAPTCHAs.  It would be easy for software to identify which pixels fall in the path of straight lines within an image of deep space.  It would not be CPU intensive relative to the other tasks of PI. 

 

I'm sure incorporating the rejected pixels into the stacking process can be done as well.  For example, it might store references for the frame number and pixel location of the rejected pixels - that info can then be referred to and excluded from the final pixel value calculation.

[funkysandman]

No, trail identification algorithms haven't been created by anyone as far as I know - that's been the point of my posts... it's possible and shouldn't be difficult.  As mentioned - for well over a decade software has been able to convert images of handwriting to digital text and even crack CAPTCHAs.  It would be easy for software to identify which pixels fall in the path of straight lines within an image of deep space.  It would not be CPU intensive relative to the other tasks of PI. 

 

I'm sure incorporating the rejected pixels into the stacking process can be done as well.  For example, it might store references for the frame number and pixel location of the rejected pixels - that info can then be referred to and excluded from the final pixel value calculation.

Interesting, I implemented a machine learning algorithm using tensorflow to detect meteors on my skycams and scans 10s of thousands of 5 second exposure images each night  - it is also very good at detecting satellites..takes less than a second to evaluate an image.  I've actually had to train the model to differentiate between meteors and satellites as well as planes...maybe there is an application of this for astrophotography.

[Lucullus]

1)

Just as an example of what to expect from just the 60 up and when they are finally place at 550km with an equal separation,

I used a simulation to find a random convergence of a lit pass and a familiar target from a random location in the U.S.

Here is a 5 minute "exposure" where consecutive satellites in the 53 degree inclined orbit will pass through the same area,

thus creating 4 streaks in the same image.

Note this is to predict the occurrence with just the 60 up right now.

The point to note is the satellites will be evenly spaced in orbital rings (planes).

Because of the movement of the Earth/thus stars under the orbit,

the same orbital plane will slowly drift across the same starfield, thus allowing streaks from consecutive satellites to be parallel but spaced accordingly to time interval separation....

Now when a few orbital planes are populated, and a target happens to fall within the intersection of two orbital planes for the location/date/time,

there will be a criss-cross set of streaks in a long exposure.

Depending on alt/az location in the sky (timing interval between each satellite will shrink closer to horizon across a FOV).

5minsim.jpg

 

I guess AP'ers could intentionally plan this out for a "fun" effect. Ready, set, go!

As freestar8n states, I think astrophotography outlier rejection algorithms can reliably get rid of occasional satellite path intersections if the affected pixels remain a minority in all taken frames. Even if the orbits are stable and satellites precisely follow behind each other in minutes intervalls, as long as the celestial sky in the background changes - which it does due to Earths rotation. But of course it will be an increasing nuissance to optimise the (pre-)processing time consumption.

 

I am wondering about the following. Bear with me as I no nothing about orbital dynamics and controls. Let's take the above image of M42 with the evenly separated satellite orbit streaks. If I am right, on the one hand, Earth rotates, but on the other hand, an orbital plane also precesses. What if we have a bright satellite burning a trail through one frame and the next bright satellite following behind on the orbital plane burns a trail right next to the first, so that both trails contact each other in a two-image-stack-without-outlier-rejection, and the same for the following sats throughout the imaging session. We would have like a satellite haze over our final image. Of course rejection-algorithms can solve occasional orbital intersections as mentioned above. But what about such a bad orbital scenario as outlined, and in addition with other crossing orbital planes so that we have a wandering orbital mesh? In a simple rejection stack, I could imagine that everywhere where orbital planes slowly cross each other there are faint orbit intersection streaks in the final stack.

How realistic is this fear?

 

2)

The LSST was mentioned several times above and that the satellite constellations will only have a negligible impact on it's scientific output due to the rejection algorithms already in place. While there is an impact on the LSST's science it will only be in the form of rejecting a handful of frames. So no worries here it seems. Although I don't know anything about the LSST team's confidence in this regard, the following link https://www.iau.org/...ann19035/#notes relativises this somewhat partially concluding in not too rough translation that these constellations 'can be detrimental to [...] large ground-based astronomical telescopes, including [see the footnote] the LSST'.

The ESO also published this announcement https://www.eso.org/...ments/ann19029/ and speaks of impacts that need to be studied and mitigated. So if even the credible ESO is attentively observing the developments the problem seems to be greater than if it were only hot-pixel-rejection algorithms. But what do I know.

 

3)

[...]

 

These satellites will be a problem. Again, it isn't just about StarLink...StarLink is just the herald of what is to come. There are going to be tens of thousands of satellites in LEO. Right now, there are 4900 satellites in ALL of earths orbital bands...which means there are fewer than that in LEO. We are adding tens of times the number of satellites we have now up there. It is going to be a problem... I just hope it doesn't end up a disaster...as at least as long  as they can control these things, they could eventually guide them to burn up in the atmosphere (and I can only hope that at some point we realize our folly here and try to correct it...) Once we experience Kessler syndrome, though...tens of thousands of satellites become millions upon millions of pieces of debris, some of it small, some larger chunks, all uncontrolled, and a lot of which we will be able to see in our photos. Kessler syndrome is really what I fear...I see it as an inevitability if we put so many satellites up there (especially since a lot of this is mostly billionaires playing popularity games here), and once it happens... Thousands of satellites will be the least of our problems...

As we now know about the Kessler syndrome for several years thanks to space situational awareness scientists, aren't these billionaires and international companies held responsible for taking care and get rid of the (their!) trash they'll obviously create and pass on to humanities next generations?

In a business which basically trades huge private profits for future dead satellite trash affecting the whole community for the sake of mostly blabla-communication - "you wouldn't believe what happened to me today..."!

 

4)

No, trail identification algorithms haven't been created by anyone as far as I know - that's been the point of my posts... it's possible and shouldn't be difficult.  As mentioned - for well over a decade software has been able to convert images of handwriting to digital text and even crack CAPTCHAs.  It would be easy for software to identify which pixels fall in the path of straight lines within an image of deep space.  It would not be CPU intensive relative to the other tasks of PI. 

 

I'm sure incorporating the rejected pixels into the stacking process can be done as well.  For example, it might store references for the frame number and pixel location of the rejected pixels - that info can then be referred to and excluded from the final pixel value calculation.

My university's astro institute had a group being strong in space situational awareness and hanged out topics where students could write theses in. One was about the refinement of their existing trail identification algorithm in individual frames. So scientists involved in space situational awareness have such algorithms at their disposal for several years now. But they are searching for space debris and sats to catalogue them and not to get rid of the trails in the frames.

[Der_Pit]

No, trail identification algorithms haven't been created by anyone as far as I know - that's been the point of my posts... 

Not really true though.  AFAIR the Hough Transform has been used extensively to locate particle tracks in bubble chamber photos.  A quite similar task.

[OrionSword]

Wait until the subscribers see how much this will cost.  I used to do weather LEO satellite imagery using a steerable antenna array as well as geostationary.  This article goes into that as well.  Won't be the first or last failure in the satellite industry.

 

 

https://www.wired.co...llite-internet/

[chadrian84]

I guess I should have clarified “trail identification algorithms haven’t been created yet for any of the commonly used astrophotography stacking programs.” Could probably be written in a couple hours by a good programmer though.

[Der_Pit]

I think that one fits in here:

 

I've created a small movie from the (319) subimages of an M10 session I did July 27.  Seems from my latitude that looks directly to their path

 

Messier 10 under attack

 

Fortunately with that many subs winsorized sigma clipping has no problem removing it.  Still not nice 

Edited by Der_Pit, 01 August 2019 - 08:30 AM.

[InFINNity]

The effect of satellite and air plane trails can be seen in this recent APOD...

 

A mouse-over will remove the trails, the trail-less image was made using Astro Pixel Processor.

Nicolàs

[Der_Pit]

Lets see if that still works if SpaceX also starts the just requested additional 30000 starlink satellites.....

(see, e.g., here