Thread by @ProfHughLewis

In advance of my monthly analysis of #Starlink conjunction data I wanted to share some additional analysis undertaken over the last few days. It's a work in progress but here's a thread looking a little deeper at the #SpaceX approach to #Starlink orbital space safety [1/n]

#SpaceX provided some relatively detailed information about its approach in a briefing to the #FCC (here: ecfsapi.fcc.gov/file/108107102…) and in an update on the website (here: spacex.com/updates/index.…) [2/n]

My focus has mostly been on understanding the implications relating to the choice of the probability threshold for collision avoidance manoeuvres. With the #SOCRATES #Starlink data now running across nearly 3 years we can gain some insights that may be useful [3/n]

As per #SpaceX filings and publications, #Starlink satellites manoeuvre if the collision probability with a non-Starlink (or inactive Starlink) object is greater than 1-in-100,000. This is 10 times lower than the threshold used for #NASA EO satellites [4/n]

It's a point that is used by #SpaceX as an example of its commitment to orbital safety. I wanted to look a little more carefully at the implications of this choice, because the #Starlink constellation has an unprecedented number of satellites and I had concerns [5/n]

I revisited the #SOCRATES data to (as best as I could) isolate the unique conjunctions predicted for Starlink since May 2019. The data reveal nearly 1 million close approaches at 5 km or less that involve Starlink and a non-Starlink object [6/n]

With these data, I can count the number of close approaches at, or exceeding, a particular collision probability level. As one might expect, there are relatively few close approaches with high collision probability but many with low collision probability [7/n]

Using a slightly different way of presenting the information, I can see the proportion of those 1 million close approaches at, or exceeding, a particular collision probability level [8/n]

Nearly all (i.e. 100%) of the close approaches are associated with collision probabilities larger than 1-in-10,000,000 (i.e. 1E-7). 10% had collision probabilities larger than 1-in-1,000,000 (i.e. 1E-6) & 1% had probabilities larger than 1-in-100,000 (i.e. 1E-5) [9/n]

It's the 1% at or above a probability of 1E-5 that are of interest because those close approaches are the ones where #Starlink will manoeuvre. I.e. the #SOCRATES data predicts about 10,000 collision avoidance manoeuvres since May 2019 [10/n]

But what about the 99% of close approaches that do not reach the 1E-5 collision probability threshold? That's more than 900,000 close approaches since May 2019. Nearly all of these would have had collision probabilities greater than 1E-7 according to #SOCRATES [11/n]

With some approximations I estimated the chance of at least 1 collision occurring as a result of conjunctions with a probability above 1E-5 (i.e. the avoided collision probability) & again for those with a probability below 1E-5 (i.e. the residual collision probability) [12/n]

For those ~1 million close approaches since May 2019 the chance of 1 or more collisions from events with Pc > 1E-5 was ~0.9. With manoeuvres, these would be avoided. The chance of 1 or more collisions from events with Pc < 1E-5 was ~0.4. This represents the residual risk [13/n]

With the #SOCRATES data, I can perform these calculations for any manoeuvre threshold. Here's a plot showing the results. It's a little complicated so I will explain [14/n]

Let's say #SpaceX adopted a collision probability of 1E-4 for the #Starlink manoeuvre threshold (x-axis value). #Starlink satellites would possibly have manoeuvred 1000 times since May 2019 (blue dashed line, y-axis on the right-hand side)... [15/n]

...The chance of at least 1 collision arising from the conjunctions where a manoeuvre would have taken place would possibly have been ~0.87 (orange line, y-axis on the left-hand side)... [16/n]

...The chance of at least 1 collision arising from the conjunctions that were ignored would possibly have been ~0.55 (grey line, y-axis on the left-hand side). [17/n]

By changing the accepted collision probability level (i.e. the manoeuvre threshold, x-axis) I can gain an understanding of the trade-offs between the possible manoeuvre burden, the risk that's been avoided (removed by manoeuvring) and the risk that's been ignored [18/n]

The only way to have removed all of the risk would have been to perform about 1 million collision avoidance manoeuvres. I think that would have been impossible even for #SpaceX [19/n]

On the face of it the 1E-5 manoeuvre threshold seems like a great step to have taken - it is 10 times better than NASA after all. However, the huge number of close approaches involving #Starlink satellites means that the residual collision probability is likely to be high [20/n]

What does this mean? Well, it means that #SpaceX are doing well at avoiding some collisions but it also means that #SpaceX are perhaps not doing enough to prevent some collisions from occurring [21/n]

The problem is that there isn't much #SpaceX can do to solve this. Removing or even reducing the residual collision probability involves a huge increase in the number of manoeuvres. To halve the current residual risk would need 10 times the number of manoeuvres [22/n]

Instead of 10,000 collision avoidance manoeuvres since May 2019, it would have been 100,000 manoeuvres. Even with that, the residual probability would still have been about 1-in-5 [23/n]

Now, consider the future. In my most recent thread on this I posted the results of a simple model for predicting the number of close approaches & collision avoidance manoeuvres for an increasing number of #Starlink satellites [24/n]

https://twitter.com/ProfHughLewis/status/1509885716603588614

There will be a huge increase in the number of close approaches that will be ignored. The residual risk will increase even if #SpaceX were to lower the manoeuvre threshold further. In the end, the law of very large numbers will probably deliver a collision [25/n]

I can almost hear cries of "so what?" #Starlink satellites operate at low LEO altitudes so any collision debris would re-enter quickly, right? [26/n]

Although larger (compared with #Starlink 1st generations satellites at least) the effects might not be unfamiliar: spacenews.com/russian-asat-d… [27/n]

This issue is also not just limited to #SpaceX & #Starlink. All large constellations and large fleets of satellites will encounter the same problem: it may not be possible to prevent collisions involving these space systems using collision avoidance manoeuvres alone [28/n]

Should governments & companies continue to launch knowing this? Are we saying that we are willing to accept this collateral damage in exchange for progress? Remember, these would be collisions possibly involving other operators (figure from: spacex.com/updates/index.…) [29/n]

Improving space surveillance & tracking will help as it can reduce the number of important close approaches, but it is not a panacea. Satellites require time to perform manoeuvres and this means making predictions of future orbital states, which continues to be difficult [30/n]

You might be wondering how I can remain optimistic, so I thought I'd end (except for acknowledgements) with this wonderful quote from @mayagabeira that really sums up how I feel about the #SpaceDebris problem & the challenges posed by new space systems:

https://twitter.com/oceana/status/1512839258989641728?t=pa_Js8TszDMkK6aMSxpizA&s=09

Thanks, as always, to the amazing @TSKelso who runs & maintains #SOCRATES @CelesTrak. All data used here are in the public domain as a result of the work by TS.

#SOCRATES predictions are based on TLE data published by @SpaceTrackOrg & should be used carefully. I try to avoid focusing on individual conjunction events (which I acknowledge is sometimes difficult to do).

I use my own code to analyse the #SOCRATES data and freely acknowledge that I am not a particularly good coder. I have also made assumptions & used approximations to deliver the results shown in this thread. You use the results at your own risk!

Finally, I wanted to offer my appreciation to the students enrolled in my 'Space Environment' module who saw these results for the first time today, asked some excellent questions & had some great insights that have helped me to better my own understanding [fin]