Mars Asteroid Impact More Likely Than Before

sheldie writes "The probability of asteroid 2007 WD5 impacting Mars has been revised following further observations. The chance of impact has increased from 1.3% to 3.9%" This is a follow-up to earlier coverage of this asteroid from last week.

Comments

Re:Versus Jupiter

[jmichaelg]

Your question is currently the question on the Minor Planet Mailing List. This morning's post by Alan Harris summarizes the situation quite well:

the problem we see is the error
model, or put slightly different, data weighting. The specific problem
with the 2007 WD5 orbit is that the block of six positions reported by
Spacewatch II on December 4-5 are isolated by a week or two before and
after from any other observations. Of the six, three have residuals
exceeding 1 arcsecond (two of the three by 2 arcsec), leaving some
uncertainty as to which three are the bad observations. Spacewatch II has
RMS fit residuals on all reported positions of about 0.5 arcsec, according
to Andrea's NEODyS site, so two of the observations, the ones discarded in
Aldo's solution, deviate by 4 times the average dispersion of Spacewatch II
observations. Chesley did something similar when he "up-weighted some of
the low noise astrometry", but how does one know? And does this imply that
Spacewatch II astrometry, with RMS errors of 0.5 arcsec, is not "low noise
astrometry"? In this instance, decrying the "lack of a rigorous
statistical error model" seems like trying to sew a silk purse from a sow's
ear -- what we really have here is simply some bad observations. Even more
useful that trying to beg 8-m time to take a few more images of 2007 WD5
would be for the Spacewatch folks to go back and look at their December 4-5
observations and try to find the problems with them, and if possible
correct the errors or at least identify which ones are bad (e.g., due to
image confusion). That's not going to tell us definitively whether it will
hit Mars (although it could tell us definitively that it won't), but it
would go a long way toward showing that impact probability estimates can be
done better than an order of magnitude uncertainty (0.3 vs. 3% impact
probability) if one only takes the time to make sensible (even if not
optimum) uncertainty evaluations.

I think it's worth noting that if it's this hard to predict whether a rock is going to hit a planet when there's confidence in the computational model but uncertainty in the observations, then it's absurd to talk about the climate in 2030-2100 when we have neither certainty that our models are accurate nor do we have very good data.