
NavList:
A Community Devoted to the Preservation and Practice of Celestial Navigation and Other Methods of Traditional Wayfinding
On improving data
From: Peter Fogg
Date: 2003 Jan 18, 05:17 +1100
From: Peter Fogg
Date: 2003 Jan 18, 05:17 +1100
Dan Allen wrote: > ... like my land sights: some are amazingly accurate while > others are not. We know this because we have the luxury of comparing > our sextant results with an averaged GPS location or a typographic map > location. > > But how would we do without these crutches? How would we do on a > rolling ship in a storm? What statistical measures or rules of thumb > can be > used to help throw out the bad sights and keep the good ones? This is > an > area that I am considering more and more. The common answer is to 'average' a series of sights. The problem with this is that the bad sights pollute the others, and there is no way to know just which is what from just looking at the numbers. But there is a technique to separate the good from the bad, so telling which is which becomes clear with a glance. It has been described before but may bear repeating. The rate a celestial body appears to rise or fall over a 5 minute period can be expressed as a sloping line on a graph. If the sights taken (as many as possible of the same body over the 5 minutes*) are plotted above this line then a second line can be drawn parallel which best fits the pattern of sights - they should also rise or fall. Then any point along this 'line of best fit' can be used for the 'time/sextant altitude' data. Any sights that are wildly off are ignored, so unlike 'averaging' they make no contribution to the result. That's it in a nutshell. The bad sights are obvious - they are the ones far from the 'line of best fit'. The reasonably good ones will lie just above and below the 'line of best fit' - meaning that the line may give a more accurate result than any individual sight. And this is my experience. Its a technique ideally suited for the 'rolling ship in a storm'. An additional advantage is that a precise time for the sight can be chosen in advance, by starting the sight taking process a few minutes before and extending it a few minutes after the exact time required. Whether you took a sight at the right moment and whether that one is any good is irrelevant. This enables, for example, a sight when the body is at 90 or 270 degrees of azimuth, giving a line of position that indicates the longitude. The alternative seems to be to take sights of multiple bodies and discard the LOPs that seem wrong, compared to the others. But this entails much more work, lots of sight reduction, improved possibility of error. Then you are left with a plotting sheet full of confusing lines. With multiple LOPs all over the place I find it difficult to choose the point of fix . If some of these bodies have similar azimuths their usefulness is questionable - they are more likely to promote further confusion. Better, I propose, to have just 3 bodies with nicely separated azimuths, and the assurance in advance that they are quite accurate. If anyone is interested in more detail about this I will be happy to expand further. Its a simple routine and I am mildly surprised that it is not more widely known and practiced - it is simpler to do than to describe. (PS*This reminds me that I've been reading about how the United States was comprehensively surveyed, a mammoth task. They found that the best technique was to get their sights done quickly; that extra fiddling with the theodolite was counter-productive - it tired the operator and, in any case, led to less accurate results.)