Re: [find_orb] baffled by longer arcs

Bill Gray Nov 26, 2014

Hi Andy,

> I was reluctant to "Filter Obs" because I didn't want to toss out a
> bunch of observations. Although I probably should have been more careful
> to let (or force) Find_Orb to exclude the same observations that the MPC is.

I'd suggest hitting 'Worst Obs'. If you have enough observations, that
"worst one" will almost certainly be a blunder that you'll look at and say:
No _way_ do I want that thing in my data.

You're then left with the puzzle of deciding just how bad is too bad.
I mostly err on the side of tossing out really obviously bad data. By the
time you're tossing out "probably bad, but not sure" observations, the
gain doesn't appear to be all that great.

A warning: there's no way for Find_Orb to know which observations
MPC has excluded. (With .rwo data from AstDyS/NEODys, there's a flag
for this purpose. They even flag excluded magnitudes separately from
excluded positions... something Find_Orb really ought to do.) In any
case, I'd really recommend using Find_Orb and its 'filter obs' to decide
what to exclude. Find_Orb has a few advantages in this department :

-- The 'over-observing' parameters do a decent job of combatting the
problem of one observatory getting a lot of astrometry, and completely
dominating the orbit solution. Essentially, the idea is that the tenth
observation from observatory A shouldn't get the same weight as the first
observation from observatory B. (AstDyS/NEODyS has recently added a version
of this. I don't think MPC does it.)

-- The 'blunder management' solution enables Find_Orb to say (for example):
This observation is so far off that it could only occur 1% of the time by
chance, but we think about 2% of observations are simply blunders. (Which
is probably a low estimate. It's rare to look at fifty observations and not
see one or two that make you wonder what happened.) In such a case, we can
say that there's about one chance in three (1% compared to 2%) that the
observation is "real". (As far as I know, Find_Orb is the only program
doing this.)

Details on both of the above are at :

http://www.projectpluto.com/blunder.htm

-- Find_Orb includes the effects of timing errors. Admittedly, while
the method is mathematically correct, we still have the problem that you
don't necessarily know what the uncertainty in timing is for a given
measurement. Find_Orb assumes a five-second uncertainty for all observations.
Not great, but I'd argue that this is still better than completely ignoring
the timing problem. (Keeping in mind that it's only a problem if, during
those five seconds, the object moves a noticeable amount. If your astrometry
is only good to 0.5 arcseconds, and the object only moved 0.1" during the
five seconds, then the effects of timing errors are essentially unnoticeable.)

Again, as far as I know, Find_Orb is the only program taking timing
errors into account.

-- Bill