Impacts of Large Satellite Constellations on Astronomy

[jg]

You also missed Tony's statement that the signal from the satellites is present in the data to a hundred to one signal to noise ratio. That's a huge problem wherever the light from the satellites are present. Say you can make things 10 times center which is great You're still going to have one hell of a single noise ratio contribution from the satellite.

It doesn't mean anything like what you just said. The only thing a Starlink having a significant SNR guarantees is that the satellite can be detected.

It says nothing directly about how much other signals could be detected. It has no effect on anything that is not the relatively small fraction of pixels directly affected. ChrisWilson68 actually left out part of the potential application of predicting the satellite trails. You don't have to just ignore the trails, you can subtract off the trails to the limits of your ability to predict the trail. We don't know what that would be yet, but the best case is to remove it down to the noise level, so there is no net effect on the science data.

It does make a difference. If you add N electrons to a pixel you are adding square root of N noise to that pixel. So if your signal is 100 times higher than the noise it is also adding a large amount of noise to the pixel. This is simple photon statistics.

Often signal of object might be even a fraction of a photon per pixel. Often you integrate for a very long time just to get the sky background level high enough that it's root and statistics allow you to detect the object.

[meberbs]

It does make a difference. If you add N electrons to a pixel you are adding square root of N noise to that pixel. So if your signal is 100 times higher than the noise it is also adding a large amount of noise to the pixel. This is simple photon statistics.

That is not even close to what SNR means. Adding more signal does not automatically increase the background noise. In the case I mentioned where the signal is treated as a known and removed the remaining noise is dependent on how well the signal is known, which is not directly related to the signal itself in any way that you can make broad statements about. It might be a fixed knowledge value, it in some cases could be proportional to some extent, but the current information tells us nothing about what that proportionality factor would be.

Often signal of object might be even a fraction of a photon per pixel. Often you integrate for a very long time just to get the sky background level high enough that it's root and statistics allow you to detect the object.

There are a number of integration techniques. If you integrate across multiple frames, the satellite would only be present in one of them, making the satellite even less relevant. In these cases this is where noise gets reduced by a factor of sqrt(n) where n is the number of frames. Even if it was just eliminating certain pixels from one frame entirely, sqrt(n-1) reduction in noise instead is not that much worse. (Whether including pixels corrected for satellite removal help or hurts depends on the residual error and the total number of frames.)

[Nomadd]

There are a number of integration techniques. If you integrate across multiple frames, the satellite would only be present in one of them, making the satellite even less relevant. In these cases this is where noise gets reduced by a factor of sqrt(n) where n is the number of frames. Even if it was just eliminating certain pixels from one frame entirely, sqrt(n-1) reduction in noise instead is not that much worse. (Whether including pixels corrected for satellite removal help or hurts depends on the residual error and the total number of frames.)

Our old 70s radars had false signal and noise reduction that worked something like that.

[Johnnyhinbos]

So tonight my wife and I are in Nantucket Harbor  aboard our Sabre 42’ Hardtop Express, Sequel. Beautiful night - finally clear sky’s and wind not blowing 30+ knots. We went to the bow to try to catch the meteor shower, traced the Big Dipper up to Polaris and, and - and we watched one satellite after another trace the exact same path through the sky in military precision. We saw maybe fifteen pass by in the same path, separated by about four seconds. Amazing. I can only assume they were part of the 58 newly launched Starlink satellites.

What a time...

[su27k]

Rubin observatory (LSST) impact:

The paper linked in the video: Mitigation of LEO Satellite Brightness and Trail Effects on the Rubin Observatory LSST

We report studies on mitigation of optical effects of bright Low-Earth-Orbit (LEO) satellites on Vera C. Rubin Observatory and its Legacy Survey of Space and Time (LSST). These include options for pointing the telescope to avoid satellites, laboratory investigations of bright trails on the Rubin Observatory LSST Camera sensors, algorithms for correcting image artifacts caused by bright trails, experiments on darkening SpaceX Starlink satellites, and ground-based follow-up observations. Starlink satellites with no brightness mitigation are presently g ~ 5.1 mag, and an initial experiment "DarkSat" is g ~ 6.1 mag. Future Starlink darkening plans may reach g ~ 7 mag, a brightness level that enables non-linear image artifact correction to the same level as background noise. However, the satellite trails will still exist at S/N ~ 100, generating systematics that may impact data analysis and limiting some science. LEO satellite trails are wider than a point-spread function because satellites are slightly out of focus due to their finite distance; for Rubin Observatory's 8.4-m mirror and a satellite at 550 km, this widening effect is about 3 arcsec, which helps avoid saturation by decreasing the trail's peak surface brightness.

Also the SATCON1 conference mentioned in the video: Satellite Constellations 1 Workshop, the bottom of the page has links to most of the presentation slides in conference.

Also an email discussion about SATCON1 on the SeeSat-L mailing list: http://www.satobs.org/seesat/Jul-2020/0013.html

[Dizzy_RHESSI]

It does make a difference. If you add N electrons to a pixel you are adding square root of N noise to that pixel. So if your signal is 100 times higher than the noise it is also adding a large amount of noise to the pixel. This is simple photon statistics.

That is not even close to what SNR means. Adding more signal does not automatically increase the background noise. In the case I mentioned where the signal is treated as a known and removed the remaining noise is dependent on how well the signal is known, which is not directly related to the signal itself in any way that you can make broad statements about. It might be a fixed knowledge value, it in some cases could be proportional to some extent, but the current information tells us nothing about what that proportionality factor would be.

That is how SNR works. See eq 2.1 from the ESO exposure time calculator formula book. On the bottom you have the noise terms, where you see the first term is sqrt(signal), which is summed in quadrature with the other contributions.

https://www.eso.org/observing/etc/doc/formulabook/node6.html

Adding signal always increases the noise, even if it is marginal because other noise sources dominate (not the case here)."Background noise" is only one type of noise in an image, the total noise is whats important. A big bright satellite trail will certainly increase the noise dramatically.

When you talk about removing the signal you there is a limiting case. You can never remove a signal to better precision than the Poissonian noise. That is just physical fact. What jg says is true. This is the best case scenario. If you don't know the signal that well then what you end up with is residuals plus noise, that will be worse than the limiting case.

[meberbs]

It does make a difference. If you add N electrons to a pixel you are adding square root of N noise to that pixel. So if your signal is 100 times higher than the noise it is also adding a large amount of noise to the pixel. This is simple photon statistics.

That is not even close to what SNR means. Adding more signal does not automatically increase the background noise. In the case I mentioned where the signal is treated as a known and removed the remaining noise is dependent on how well the signal is known, which is not directly related to the signal itself in any way that you can make broad statements about. It might be a fixed knowledge value, it in some cases could be proportional to some extent, but the current information tells us nothing about what that proportionality factor would be.

That is how SNR works. See eq 2.1 from the ESO exposure time calculator formula book. On the bottom you have the noise terms, where you see the first term is sqrt(signal), which is summed in quadrature with the other contributions.

Please look at the post I was responding to. It incorrectly claims that the noise is simply sqrt(signal) Those other contributions simply make the result invalid. (And we have no idea based on what was presented what is dominating, because the video does not specify what any of those contributions are, it uses an adjusted scale with background signals subtracted off.)

Adding signal always increases the noise, even if it is marginal because other noise sources dominate (not the case here)."Background noise" is only one type of noise in an image, the total noise is whats important. A big bright satellite trail will certainly increase the noise dramatically.

You are making assertions without the full set of information required to back them up. If one term is negligible, than it in effect does not increase the noise, and certainly not at a fixed rate.

When you talk about removing the signal you there is a limiting case. You can never remove a signal to better precision than the Poissonian noise. That is just physical fact. What jg says is true. This is the best case scenario. If you don't know the signal that well then what you end up with is residuals plus noise, that will be worse than the limiting case.

Except it is not limiting, you are making assumptions that are invalid. Please read the rest of my post where I address the integration part and how mentioning integration further weakens the argument. The link you provided nicely shows the sqrt(n) I talked about in eqn 2.2. For some reason it only applies it to infrared bands for whatever special case it is discussing.

[Steven Pietrobon]

If noise power N is proportional to the square root of signal power S, then the signal to noise ratio SNR (which is what ultimately defines performance) is proportional to S/sqrt(S) = sqrt(S). This means that as signal power increases so does the SNR.

[Dizzy_RHESSI]

That is how SNR works. See eq 2.1 from the ESO exposure time calculator formula book. On the bottom you have the noise terms, where you see the first term is sqrt(signal), which is summed in quadrature with the other contributions.

Please look at the post I was responding to. It incorrectly claims that the noise is simply sqrt(signal) Those other contributions simply make the result invalid. (And we have no idea based on what was presented what is dominating, because the video does not specify what any of those contributions are, it uses an adjusted scale with background signals subtracted off.)

No it does not. jg said "If you add N electrons to a pixel you are adding square root of N noise to that pixel.". Which is exactly correct. I don't see any claim that this is the only source of noise.

And we do know what is dominating, because there are lots of other instruments already in existence with similar parameters.  Subtracting the background does not change the noise.

Adding signal always increases the noise, even if it is marginal because other noise sources dominate (not the case here)."Background noise" is only one type of noise in an image, the total noise is whats important. A big bright satellite trail will certainly increase the noise dramatically.

You are making assertions without the full set of information required to back them up. If one term is negligible, than it in effect does not increase the noise, and certainly not at a fixed rate.

The ESO reference backs that up. You can see from the noise part of the equation that increasing the signal increases the noise. Trying to insert "in effect" is totally vague and moving the goalposts. The contribution to the noise from the signal is always the same (as jg claimed), the total noise depends on the other terms.

When you talk about removing the signal you there is a limiting case. You can never remove a signal to better precision than the Poissonian noise. That is just physical fact. What jg says is true. This is the best case scenario. If you don't know the signal that well then what you end up with is residuals plus noise, that will be worse than the limiting case.

Except it is not limiting, you are making assumptions that are invalid. Please read the rest of my post where I address the integration part and how mentioning integration further weakens the argument. The link you provided nicely shows the sqrt(n) I talked about in eqn 2.2. For some reason it only applies it to infrared bands for whatever special case it is discussing.

It is limiting. If you're talking about subtracting the trail then you will only ever do that on the single exposure. You correctly stated the "best case is to remove it down to the noise level", but because the trail is there the noise will be higher.

What you described in your other comment is about rejecting the data by masking it in the stack, at that point you're not beating the limiting case you've just thrown away the data. Masking is what LSST will use in their stacks, but they're not only interested in stacks. For the transients and asteroids for example they will be using individual exposures.

[meberbs]

No it does not. jg said "If you add N electrons to a pixel you are adding square root of N noise to that pixel.". Which is exactly correct. I don't see any claim that this is the only source of noise.

If it is not the only source of noise, the statement is plainly mathematically wrong. given X already exists, you get
sqrt(N + x) - sqrt(x
more noise. That is only equal to sqrt(N) if x is 0.

And we do know what is dominating, because there are lots of other instruments already in existence with similar parameters.  Subtracting the background does not change the noise.

What parameter on what instruments?  You are making assertions without providing evidence, there has been entirely too much nonsense thrown out there on this topic for me to trust assertions without scientific evidence. I didn't say that subtracting the background eliminates the fact that there is noise, I was pointing out that it eliminates information about the actual absolute value of noise present.

The ESO reference backs that up. You can see from the noise part of the equation that increasing the signal increases the noise. Trying to insert "in effect" is totally vague and moving the goalposts. The contribution to the noise from the signal is always the same (as jg claimed), the total noise depends on the other terms.

The concept of something being a negligible contribution factor is not all that vague. And again please look at the simplified equations I wrote above. Things are neither linear nor constant. The effect of adding more noise of any type has an effect that depends on how much noise was already there.

It is limiting. If you're talking about subtracting the trail then you will only ever do that on the single exposure. You correctly stated the "best case is to remove it down to the noise level", but because the trail is there the noise will be higher.

It depends on the details, but it possibly helps if integrating a small number of frames (and some other variables). The more frames you have the less effect removing 1 frame has on the noise reduction.

What you described in your other comment is about rejecting the data by masking it in the stack, at that point you're not beating the limiting case you've just thrown away the data. Masking is what LSST will use in their stacks, but they're not only interested in stacks. For the transients and asteroids for example they will be using individual exposures.

Jg is the one that brought up the stacked situation, I simply pointed out that in general removing the satellite trail from the data could have a negligible effect on the output for this case, and you don't seem to be arguing against that. When stacking frames you are never limited by the noise in any single frame.

For the single frame case, that is where the above discussion comes back into play, which includes a possibility of removing the trail to see things that would be hidden by it. The video did not state how effective this could be, because they do not know yet. They would need more data before even considering that, and they have more information than is available to us here. Trying to argue about how much reduction this would be is fairly pointless. The range of possible values includes good enough to eliminate most or all of the effects on any science being done.

[Dizzy_RHESSI]

No it does not. jg said "If you add N electrons to a pixel you are adding square root of N noise to that pixel.". Which is exactly correct. I don't see any claim that this is the only source of noise.

If it is not the only source of noise, the statement is plainly mathematically wrong. given X already exists, you get
sqrt(N + x) - sqrt(x
more noise. That is only equal to sqrt(N) if x is 0.

Did jg say that the total noise would be sqrt(N) higher? No. I understand that noise terms sum in quadrature, so I do not have a problem with that statement. I think we agree on the outcome so I don't see any point in getting bogged down arguing semantics.

What parameter on what instruments?  You are making assertions without providing evidence, there has been entirely too much nonsense thrown out there on this topic for me to trust assertions without scientific evidence. I didn't say that subtracting the background eliminates the fact that there is noise, I was pointing out that it eliminates information about the actual absolute value of noise present.

The comparison I looked at was VLT/FORS2. It has a similar pixel scale, site quality and filter set, but lower read noise (which I accounted for). I don't know why this is interesting.

The ESO reference backs that up. You can see from the noise part of the equation that increasing the signal increases the noise. Trying to insert "in effect" is totally vague and moving the goalposts. The contribution to the noise from the signal is always the same (as jg claimed), the total noise depends on the other terms.

The concept of something being a negligible contribution factor is not all that vague. And again please look at the simplified equations I wrote above. Things are neither linear nor constant. The effect of adding more noise of any type has an effect that depends on how much noise was already there.

Did I say it was linear? No, in fact I made it quite clear that it wasn't. My statement was "Adding signal always increases the noise, even if it is marginal because other noise sources dominate (not the case here)." Which is true by the equation you have written, but for some reason you objected to the first part.

It is limiting. If you're talking about subtracting the trail then you will only ever do that on the single exposure. You correctly stated the "best case is to remove it down to the noise level", but because the trail is there the noise will be higher.

It depends on the details, but it possibly helps if integrating a small number of frames (and some other variables). The more frames you have the less effect removing 1 frame has on the noise reduction.

I'm not talking about removing frames, I specifically said subtracting the trail (and so did you in the post I quoted). The trail will only be in one image. If you stack other images of the same field you will decrease the SNR of the trail, because it's not in any of the other images. You want to subtract a trail in the image where it has the highest SNR, which is a single exposure.

[meberbs]

No it does not. jg said "If you add N electrons to a pixel you are adding square root of N noise to that pixel.". Which is exactly correct. I don't see any claim that this is the only source of noise.

If it is not the only source of noise, the statement is plainly mathematically wrong. given X already exists, you get
sqrt(N + x) - sqrt(x
more noise. That is only equal to sqrt(N) if x is 0.

Did jg say that the total noise would be sqrt(N) higher? No. I understand that noise terms sum in quadrature, so I do not have a problem with that statement. I think we agree on the outcome so I don't see any point in getting bogged down arguing semantics.

It is directly in the quote. "adding square root of N noise." You seem to be agreeing that this is wrong while denying that it was said.

The comparison I looked at was VLT/FORS2. It has a similar pixel scale, site quality and filter set, but lower read noise (which I accounted for). I don't know why this is interesting.

You made baseless assertions about what is dominating that are now only slightly less baseless. Science is based on evidence, and this forum should be better than the spreading of baseless FUD which has happened way to often on this issue. Your response here doesn't really give me confidence that you actually have a good idea of LSST's relevant parameters, to make solid claims about the threshold for what is dominating.

The ESO reference backs that up. You can see from the noise part of the equation that increasing the signal increases the noise. Trying to insert "in effect" is totally vague and moving the goalposts. The contribution to the noise from the signal is always the same (as jg claimed), the total noise depends on the other terms.

The concept of something being a negligible contribution factor is not all that vague. And again please look at the simplified equations I wrote above. Things are neither linear nor constant. The effect of adding more noise of any type has an effect that depends on how much noise was already there.

Did I say it was linear? No, in fact I made it quite clear that it wasn't. My statement was "Adding signal always increases the noise, even if it is marginal because other noise sources dominate (not the case here)." Which is true by the equation you have written, but for some reason you objected to the first part.

See the bolded statement. I responded to it because you said it and it is simply mathematically false. You directly said that the increase was constant, (and agreed with jg who made the same claim.) But the increase is simply not constant. The additional statement you quoted here does not change the meaning of your other statements, if you add 10 to 100, the total depends on the 100, and a constant but small increase happens from 10+100 = 110, but in reality, you would get sqrt(10^2 + 100^2) = 100.5. The second statement you quoted is compatible with both equations, but the statement I bolded is only compatible with the incorrect version. Both cases show the final result as dependent on the other sources of noise, but only in the first one is the increase constant as you stated. You are the one ignoring part of what you said, not me.

It is limiting. If you're talking about subtracting the trail then you will only ever do that on the single exposure. You correctly stated the "best case is to remove it down to the noise level", but because the trail is there the noise will be higher.

It depends on the details, but it possibly helps if integrating a small number of frames (and some other variables). The more frames you have the less effect removing 1 frame has on the noise reduction.

I'm not talking about removing frames, I specifically said subtracting the trail (and so did you in the post I quoted). The trail will only be in one image. If you stack other images of the same field you will decrease the SNR of the trail, because it's not in any of the other images.

So you agree that one way or another the satellite trails do not have any significant effect on images of very dim objects, a case that jg is the one who brought up in an attempt to claim that satellite trails cause significant unavoidable problems?

You want to subtract a trail in the image where it has the highest SNR, which is a single exposure.

Yes, that is the case where subtracting the satellite trail has the most direct effects, since the stacking usually dominates anyway for the case of dimmest objects. However, don't claim that it is what I want to do, it is something that was stated in the video as a potential future path. Currently no one has the information required to know exactly how effective it will be, but it at least leaves open the possibility of making the impact on the final science data negligible, even if the satellites are still visible in the frame (which they would be in any realistic scenario.)  You have presented no argument against this point, instead defending incorrect points jg made, even though you seem to be acknowledging the math that says something distinctly different than the original claims.

If you agree that stacking makes arguments about very dim objects irrelevant, and that the increase in noise is not in fact constant, but highly dependent on the existing noise level, then maybe we can stop wasting our time on this.

[Dizzy_RHESSI]

You made baseless assertions about what is dominating that are now only slightly less baseless. Science is based on evidence, and this forum should be better than the spreading of baseless FUD which has happened way to often on this issue. Your response here doesn't really give me confidence that you actually have a good idea of LSST's relevant parameters, to make solid claims about the threshold for what is dominating.

It wasn't a baseless assertion, I did check. You asked questions about how I established that and I answered. You asked "What parameter on what instruments?", which I told you. It's quite something to stress the need for "evidence" and then dismiss a claim based on your feelings about my answer.

The contribution to the noise from the signal is always the same (as jg claimed)

You directly said that the increase was constant

No I didn't. Read what is says carefully. I said the contribution was always the same, which is true, the term going into the equation is always sqrt(N_obj). My statement says nothing about how that combines with the other noise terms or how the total noise changes. Also note that I had already said in the post before that one that the increase was not linear, it was perfectly clear. It's completely disingenuous to take things out of the context of the thread like this, just so you have something to complain about.

So you agree that one way or another the satellite trails do not have any significant effect on images of very dim objects, a case that jg is the one who brought up in an attempt to claim that satellite trails cause significant unavoidable problems?

No I don't agree with that as a general statement, and you haven't even bothered to explain in what context or what "very dim" even means.

Yes, that is the case where subtracting the satellite trail has the most direct effects, since the stacking usually dominates anyway for the case of dimmest objects. However, don't claim that it is what I want to do, it is something that was stated in the video as a potential future path. Currently no one has the information required to know exactly how effective it will be, but it at least leaves open the possibility of making the impact on the final science data negligible, even if the satellites are still visible in the frame (which they would be in any realistic scenario.)  You have presented no argument against this point, instead defending incorrect points jg made, even though you seem to be acknowledging the math that says something distinctly different than the original claims.

It was a generic "you". And you're moving the goalposts again. The point I contested was your claim that the best case scenario after subtraction could be "no net effect on the science data", which is impossible. I have described why, because adding signal always increases the noise. You eventually seemed to agreed with this point (" Things are neither linear nor constant. ") and hence, the noise has to increase with the presence of a trail. But instead of staying on point you've diverted the discussion onto semantics.

[meberbs]

You made baseless assertions about what is dominating that are now only slightly less baseless. Science is based on evidence, and this forum should be better than the spreading of baseless FUD which has happened way to often on this issue. Your response here doesn't really give me confidence that you actually have a good idea of LSST's relevant parameters, to make solid claims about the threshold for what is dominating.

It wasn't a baseless assertion, I did check. You asked questions about how I established that and I answered. You asked "What parameter on what instruments?", which I gave you. It's quite something to stress the need for "evidence" and then dismiss a claim based on your feelings about my answer.

You still provided no numbers, references, or calculations. I am not going to take the word of someone who has trouble with the definition of someone who has trouble with the definition of a simple word like "constant." (see below.) Your answer boils down to "because you said so."

The contribution to the noise from the signal is always the same (as jg claimed)

You directly said that the increase was constant

No I didn't. Read what is says carefully. I said the contribution was always the same,

The definition of the word constant (according to google) is "remaining the same over a period of time," so yes you did say that it was constant, something completely false.

which is true, the term going into the equation is always sqrt(N_obj). My statement says nothing about how that combines with the other noise terms or how the total noise changes.

Do you speak some other language as a primary language? (Serious question, this is an international forum) Because your statement in English plainly says that the noise increase is always the same (constant.) There is no sqrt(N) appearing by itself, instead it appears in the form sqrt(N + x + y...) That is a very different term.

Also note that I had already said in the post before that one that the increase was not linear, it was perfectly clear.

No, your post that I was responding to  (8/22/2020 at 6:15 pm UTC) was the first time you even used the word linear, and before that you were clearly defending jg's posts which stated a simple (implicitly linear) addition of noise equal to sqrt(n)

It's completely disingenuous to take things out of the context of the thread like this, just so you have something to complain about.

You are the one taking what you said out of context, and claiming that what you said is completely different than what you actually did. You even deny that you said something, when you said practically the word for word definition of it. And then you try to call me disingenuous?

So you agree that one way or another the satellite trails do not have any significant effect on images of very dim objects, a case that jg is the one who brought up in an attempt to claim that satellite trails cause significant unavoidable problems?

No I don't agree with that as a general statement, and you haven't even bothered to explain in what context.

What context is there to explain? Jg tried to claim that satellite trails would be a problem for LSST detecting objects at a fraction of a photon per pixel level, which obviously needs stacking to detect at all. I already explained how stacking makes throwing out pixels from a single frame have marginal impact in the case of any reasonable number of frames.

It was a generic "you". And you're moving the goalposts again.

I am me, you were addressing me, there is no "you" which includes both me and the scientists working LSST. I have done no goalpost moving, you are the one who keeps conflating different things and trying to rewrite old posts.

The point I contested was your claim that it could have "no net effect on the science data", which is impossible.

This statement is pure FUD. You have provided exactly 0 evidence to support it.

I have described why, because adding signal always increases the noise. You eventually seemed to agreed with this point (" Things are neither linear nor constant. ") and hence, the noise has to increase with the presence of a trail. But instead of staying on point you've diverted the discussion onto semantics.

The semantics is because you are straight up lying about things that were previously said by both me and you. When you deny stating something that you said a word for word definition of, I am going to pull out a dictionary. (and please don't try to say it wasn't word for word, "always" falls into the category "a period of time")
I gave the mathematical example where the noise started at 100 and "10" more noise actually meant 0.1 more noise. A change like this cannot be meaningfully called an increase or realistically claimed to have any impact on the science data. You have provided no counterargument to this.

[FutureSpaceTourist]

https://twitter.com/planet4589/status/1298321196879151105

The SATCON1 report is now available at https://aas.org/satellite-constellations-1-workshop-report

https://twitter.com/thesheetztweetz/status/1298325220332376067

The "Impact of Satellite Constellations" report's 2 findings and 10 recommendations:

[abaddon]

It's a little bit disappointing that their first listed mitigation is basically "don't do it".  That seems unrealistic and not particularly helpful.

As for #2-4, with respect to Starlink:

#2 - Deploy satellites at 600km or lower - SpaceX is doing this currently and wants to do this in the future, pending FCC approval
#3 - Darken satellites by lowering albedo, shading reflected sunlight, or some combination - SpaceX tried the former, tried the latter, and has settled on doing that moving forward.
#4 - Control satellites attitude in orbit so it reflects sunlight less on Earth - SpaceX does this on orbit raise, does not do this on-orbit (but see #3).

#5-6 are aimed at the image acquisition/processing side.

The report notes explicitly that constellations above 600km "are an even greater concern to astronomers as they include all of the impacts [of <=600km satellites] but can also be illuminated all night long.

In addition, for Constellation Operators, there are several recommendations I am too lazy to transcribe.

[eeergo]

It's a little bit disappointing that their first listed mitigation is basically "don't do it".  That seems unrealistic and not particularly helpful.

I'm at a loss why it should be disappointing to anyone:
- the conclusion is accurate (no way to have zero impact with megaconstellations),
- it's in this -very moderate and accommodating- lobby's interest,
- it *should* be their official ideal position considering their objectives, and
- the general damage-to-benefit assessment for megaconstellations is currently, at best, highly debatable (except of course regarding the very narrow topic of the financial benefit to the operating corporation, which anyway is also somewhat debatable as already empirically observed).

Some quotes from the study, directly contradicting misguided, wishful-thinking or directly malicious comments right upthread:

On wide-brush impacts and possible mitigations:

Nighttime images without the passage of a Sun-illuminated satellite will no longer be the norm. If the 100,000 or more LEOsats proposed by many companies and many governments are deployed, no combination of mitigations can fully avoid the impacts of the satellite trails on the science programs of current and planned ground-based optical-NIR astronomy facilities. [...]  Astrophotography, amateur astronomy, and the human experience of the stars and the Milky Way are already affected.

On why the fuss:

Many astronomical investigations collect data with the requirement of observing any part of the sky needed to achieve the research objective with uniform quality over the field of view. These include studies that are among the highest priorities in the discipline: stellar populations in the Milky Way and neighboring galaxies; searches for potentially hazardous near-Earth objects; identification of gravitational wave sources such as neutron star mergers; and wide-area searches for transiting exoplanets.

At a minimum, a fraction of the area being imaged is lost to the trails or significantly reduced in S/N (signal-to-noise ratio). However, many of these areas of research also include a time-critical aspect and/or a rare, scientifically critical target. Such a missed target, even with low probability, will significantly diminish the scientific impact of the project. For example, if a near-Earth object is not recovered, its orbital parameters are lost. If the transit of a promising super-Earth exoplanet candidate is missed, the orbital timing may not be recovered. If the optical counterpart of a gravitational wave source is lost in the few percent of pixels in satellite trails, its rapid fading may preclude subsequent identification.

And an illustrative example of what is being given away for -comparatively- nothing:

In 1995, the Hubble Space Telescope used its valuable observing time to do something seemingly frivolous: stare over ten days at a blank patch of sky. The target area was minuscule — just twice the apparent size of Venus in its crescent phase — but the resulting image, the Hubble Deep Field, revealed thousands of galaxies from the earliest history of the Universe. Subsequent deep field images revealed tens of thousands more galaxies in equally tiny patches of sky. Dark skies hold many secrets, and the flagship ground-based facilities of today are steadily revealing them. Vera C. Rubin Observatory will be online in the next decade. In upcoming decades a set of thirty-meter facilities will come online, expanding substantially our view to our origins. For numerous reasons of expense, maintenance, and instrumentation, these facilities cannot be operated from space. Ground-based astronomy is, and will remain, vital and relevant.

Technical takes on added systematics which can NOT and will NOT easily and summarily be waved away:

- Satellites can ruin detections [of fast transient events] as well as spectroscopic follow-up and, as the events fade very rapidly, the ability to  re-acquire the data is lost.
- Simultaneous data from optical/IR observatories and other detectors, such as neutrinos or gravitational waves (GW), represent a unique multi-messenger science opportunity in the next decade. As frequently as once a week, the network of GW detectors will detect events at very high S/N and within minutes will announce 90% likelihood areas on the sky. The first job is to detect any electromagnetic counterpart. [...] Owing to the time-critical nature, some of this search will occur during twilight. Satellite trails interfere with algorithms developed to distinguish real transient events from false detections.
- Measuring the physics of dark matter and dark energy requires billions of extremely faint (26th magnitude) galaxies for which the shape must be accurately known to one part in 10,000. Science discoveries from these measurements will be more affected by systematics than by sample size. [...]  The masked long satellite trails present a low surface brightness systematic error at the edge of the mask, generating a line of correlated noise — potentially producing a cosmic shear bias. Each satellite trail will have its own low surface brightness systematic error extending 30–60 arcseconds from the trail, depending on the satellite brightness. These residual errors and correlated linear noise features scale with trail brightness. l. Processing and artifact removal [...] is algorithmically feasible for satellite trails that are fainter than magnitude 7–8, but it represents additional work beyond the original scope of the project [...] , the sheer task of tackling these new systematic errors in the released data products is likely beyond the capability of many.
- [NEOs inform] the formation and evolution of the Solar System, but the most direct motivation for discovering and characterizing NEOs is their potential to collide with the Earth and cause catastrophic damage to ecosystems and human civilization[/u]. NEO detection and characterization has a US congressional mandate and is also supported by the United Nations Office for Outer Space Affairs. These surveys operate in the twilight hours when their targets are visible but also when satellite interference is worst. Pairs, triples, or quads of observations must be made within a short time in order to form a moving object “tracklet,” and the probability of a LEOsat trail interfering with this process is high. LEOsats already cause loss of data to Pan-STARRS, the Catalina Sky Survey, and other NEO surveys, effectively wiping out a long trail in the focal plane. Trails also generate spurious artifacts that can confuse automated pipelines. Just after evening twilight and just before morning twilight are the only usable parts of the night for detecting NEOs at low solar elongation. As currently understood, either the Starlink Generation 2 or the OneWeb scenario (of order 40,000 satellites) will significantly degrade twilight near-Sun observations [...] The satellite trail masking developed for the LSST pipeline processing is very promising, but it may also unintentionally remove trails originating from NEOs. [...] NEOSM will be sensitive to larger, more distant NEOs, not the myriad close-approachers that must be observed from the ground.
- Spectroscopic observations generally cover smaller fields of view than imaging programs. However, exposure times can be much longer for spectroscopy. A bright satellite crossing could ruin the entire exposure, as it is not known a priori which observations are contaminated, forcing a repeat exposure or possible loss of science opportunity. With an 8 square degree field of view and long exposures (10–20 minutes), [DESI which completed its construction phase and will begin operations in 2021] it is not possible to “point between” satellite trails. [...] MSE will generally always target very faint sources, and so the signal from LEOsats will dominate over the science target. Practically speaking, this means that the observation of those targets will be precluded. LEOsats also leave a much wider trail than the effective size of low surface brightness objects, which will impact the necessarily long integration times for these faint objects. The largest contributor to this effective faint object size is the wide wings of the point-spread function (PSF) in typical turbulent air corresponding to good sky conditions (0.7 arcseconds FWHM seeing). Since the mean separation between fibers or slits in a 0.2–1.0-degree focal plane is comparable to LEOsat trail width, the probability of a LEOsat trail polluting one or more spectra is high with tens of thousands of LEOsats. Owing to the long exposure times, there is no mitigation for the next generation of large spectroscopic facilities where control of mid-exposure shuttering is not possible [...] . To mitigate the impacts, it is important to be able to promptly flag (within ~24 hours after the observations) which fibers were affected by a satellite. This implies developing the ability to access the positions of the satellites with a precision comparable to a fiber diameter, and with a timing accuracy of ~1 second.
- Satellites prevent the use for faint galaxy science of the region of the frames extending up to 60 or more pixels away from their trails (i.e., 120-pixel diameter, equivalent to ~30 arcseconds swath). Bright (V mag < 12) image artifacts can make it difficult or impossible to detect faint galaxies in large regions of the field. This argument is generally true for extended regions of low surface brightness, or for any imaging program with the expectation of uniform signal-to-noise ratio (S/N) over 100% of the areal coverage.
- [Exoplanet detection around] stars that fall near satellite trails will suffer from skewed and less precise photometry, as well as added noise. Exoplanet detection will be impossible for stars that fall directly under a trail. Some of the most severely affected targets will be the M dwarfs, since cooler stars (at a fixed distance) will suffer larger relative effects. With the full constellations deployed, it will be impossible to detect super-Earth planets around M dwarf stars crossed by satellite.
- Arguably the most exciting and important science to come out of current and planned astronomical facilities will
be the discoveries of types of objects and phenomena not yet observed nor predicted by theorists. It is impossible to calculate the risk or the impact of losing such opportunities to discover the unexpected without knowing what we’re missing. But some phenomena will surely go undiscovered as a result of significant interference from LEOsats.
- In addition to the scientific value of the night sky, there is cultural and social value that is difficult, if not impossible, to quantify
 

Summary on mitigations from both parts:

SpaceX has shown that operators can reduce reflected sunlight through satellite body orientation, Sun shielding, and surface darkening. A joint effort to obtain higher accuracy public data on predicted locations of individual satellites (or ephemerides) could enable some pointing avoidance and mid-exposure shuttering during satellite passage. Observatories will need to adopt more dynamic scheduling and observation management as the number of constellation satellites increases, though even these measures will be ineffective for many science program.
With state-of-the-art masking techniques for satellite trails and current understanding of systematics and losses induced by the requirement for such masking, the impact of higher altitude LEOsat constellations ranges from costly additional exposure time per area (at the 10–20% level or higher) to the complete loss of ability to study certain astrophysical problems.

But sure, all impacts are negligible or trivially solvable. Detractors of brute-force megaconstellations are just luddite irrational contrarians refusing to be illuminated by the NewDawn(C) of obscenely powerful billionaires who have the nerve to publicly claim now not even the sky is the limit for their greediness because no amount of riches is enough to ColonizeDeserticHell(TM) and trickle it down to the proles they're going to be secularly screwing.

Lost unique science that has demonstrably and hugely advanced human life quality, knowledge, perception and even philosophy is undeniably fringe compared to the untold benefits of gaming/streaming video in the middle of nowhere from LEO (instead of the boring passively-available proper infrastructure investment), or military global strike comms, or something. As with climate change, gish galloping, censoring, self-censoring sect-style, and sealioning attacks on any attempt at conservation will ruthlessly proceed while the goalposts are moved, until "inevitability" and "it is what it is" can be called upon.

[Frogstar_Robot]

But sure, all impacts are negligible or trivially solvable. Detractors of brute-force megaconstellations are just luddite irrational contrarians refusing to be illuminated by the NewDawn(C) of obscenely powerful billionaires who have the nerve to publicly claim now not even the sky is the limit for their greediness because no amount of riches is enough to ColonizeDeserticHell(TM) and trickle it down to the proles they're going to be secularly screwing.

Lost unique science that has demonstrably and hugely advanced human life quality, knowledge, perception and even philosophy is undeniably fringe compared to the untold benefits of gaming/streaming video in the middle of nowhere from LEO (instead of the boring passively-available proper infrastructure investment), or military global strike comms, or something. As with climate change, gish galloping, censoring, self-censoring sect-style, and sealioning attacks on any attempt at conservation will ruthlessly proceed while the goalposts are moved, until "inevitability" and "it is what it is" can be called upon.

Huh? If you were intending to come across as an irrational contrarian, you succeeded.

The irony of course, is that scientists are getting a taste of their own medicine, and they don't like it.

Do scientists have a responsibility for the consequences from their discoveries?

[eeergo]

But sure, all impacts are negligible or trivially solvable. Detractors of brute-force megaconstellations are just luddite irrational contrarians refusing to be illuminated by the NewDawn(C) of obscenely powerful billionaires who have the nerve to publicly claim now not even the sky is the limit for their greediness because no amount of riches is enough to ColonizeDeserticHell(TM) and trickle it down to the proles they're going to be secularly screwing.

Lost unique science that has demonstrably and hugely advanced human life quality, knowledge, perception and even philosophy is undeniably fringe compared to the untold benefits of gaming/streaming video in the middle of nowhere from LEO (instead of the boring passively-available proper infrastructure investment), or military global strike comms, or something. As with climate change, gish galloping, censoring, self-censoring sect-style, and sealioning attacks on any attempt at conservation will ruthlessly proceed while the goalposts are moved, until "inevitability" and "it is what it is" can be called upon.

Huh? If you were intending to come across as an irrational contrarian, you succeeded.

The irony of course, is that scientists are getting a taste of their own medicine, and they don't like it.

Do scientists have a responsibility for the consequences from their discoveries?

I summarize and quote a scientific, professional consensus document full of technical conclusions (with positive input from the parties involved in causing the problem in the first place, let's not forget that) that directly contradicts irrational half-cooked claims that "everything's fine because I say so with partial skewed data to back it up" posted a bit upthread... and my post is irrational contrarianism?

Your post links a random philosophical generic opinion comment board and provides no counterfacts against the above, blaming scientists for corporate plans, and you're the voice of reason? It'd be laughable if it wasn't so sad these points of view are flooding any debate.

[tyrred]

What a tricky balance. Watch the stars from Earth and reach for the stars from above?