SOCRATES seemed to see a much smaller probability for this event when it recalculated around the same time:
https://celestrak.com/SOCRATES/search-results.php?IDENT=NAME...
I wonder why?
https://celestrak.com/SOCRATES/search-results.php?IDENT=NAME...
I wonder why?
LeoLabs has a recent paper [1] that evaluates the state prediction accuracy. They're finding the right thing (Mahalanobis distance at various temporal separations between their state estimate and an object position). See secs. 4 and 5.
But: they are not getting the chi-squared distribution that you'd want. They are getting log-normals, with significant probability mass out to a Mahalanobis distance of 10 and higher (!) -- see the figure in sec. 5.
That's a very high deviation. It's not unexpected, because this is a hard problem, but the deviations do not appear to follow the expected law.
If their collision-probability takes this into account, that's good. But if not -- if they're assuming a Gaussian prediction cloud -- then their collision-probability will not be well-calibrated. If my cursory examination of their (well-done!) paper is correct, a poorly-calibrated probability would be very wrong indeed.
[1] https://amostech.com/TechnicalPapers/2019/Astrodynamics/Park...