NavList:
A Community Devoted to the Preservation and Practice of Celestial Navigation and Other Methods of Traditional Wayfinding
Re: Sun sights at the Gulf of Mexico
From: Stan K
Date: 2016 Feb 25, 10:01 -0500
From: Stan K
Date: 2016 Feb 25, 10:01 -0500
Bill,
David Burch of Starpath wrote an article explaining a better way to eliminate outliers than least-squares linear regression, and why linear regression might have you include a bad sight or eliminate a good sight:
https://www.starpath.com/resources2/sight_average.pdf
As far as how far away from local noon linear regression should/could be used, that a tough call. Linear regression should be used when the data approximates a straight line, so, even away from local noon, it shouldn't be used for an extended period of time. I'm pretty sure that away from local noon 20 minutes would be fine. Around local noon, though, it becomes a function of the time of year, i.e. just how flat the curve is at the time of interest. Maybe someone else can come up with a more quantitative answer, but I would suggest avoiding linear regression within 30 minutes of local noon.
FWIW, if you expect to eliminate sights based on linear regression, the Sight Averaging tool of Celestial Tools (latest beta attached) does a least-squares linear regression of the provided data, then calculates the residuals, the differences between the actual value and the point on the linear regression line. You can use it to simply determine the sight closest to the line, or select or eliminate sights from the list for averaging. Ron Jones' Celestial Navigator spreadsheet (www.calmseas.org/boating_links/education/celestial_navigator.xlsm) also does a plot of the data. There are also several other programs, some by other NavList members, that eliminate the need for manual plotting.
Stan
-----Original Message-----
From: Bill Lionheart <NoReply_Lionheart@fer3.com>
To: slk1000 <slk1000@aol.com>
Sent: Thu, Feb 25, 2016 5:50 am
Subject: [NavList] Re: Sun sights at the Gulf of Mexico
From: Bill Lionheart <NoReply_Lionheart@fer3.com>
To: slk1000 <slk1000@aol.com>
Sent: Thu, Feb 25, 2016 5:50 am
Subject: [NavList] Re: Sun sights at the Gulf of Mexico
I took six sights and what I had done is just base the calculation on the one I thought looked best. When I plotted a regression line, and removed one as an outlier, I get a position line 106m from my position given by Google Maps staelite image (and confirmed by my highly accurate Samsung S4 phone with GPS and GLONASS). So the moral in that is that the errors are due mainly to the variability of my technique (and small ripples on my tray of water), and linear regression is a good way of reducing these errors.
I was wondering though when close to local noon (but either before or after) is linear regression still a good idea? The curve altitude against time is less straight the closer you get to noon. As a rule of thumb how far away from local noon should you use linear regression?
