From: John Halleck (John.Halleck_at_utah.edu)
Date: Thu Feb 08 2001 - 19:08:55 CET
Received: (from mdom_at_localhost) by karto.ethz.ch (8.9.3/8.9.3/SuSE Linux 8.9.3-0.1) id TAA31991 for cavexml-outgoing; Thu, 8 Feb 2001 19:08:34 +0100 Received: from cor.oz.cc.utah.edu (nahaj_at_cor.oz.cc.utah.edu [155.99.2.2]) by karto.ethz.ch (8.9.3/8.9.3/SuSE Linux 8.9.3-0.1) with ESMTP id TAA31987 for <cavexml_at_cartography.ch>; Thu, 8 Feb 2001 19:08:33 +0100 Received: from localhost (nahaj_at_localhost) by cor.oz.cc.utah.edu (8.9.2/8.9.2) with ESMTP id LAA22455; Thu, 8 Feb 2001 11:08:56 -0700 (MST) Date: Thu, 8 Feb 2001 11:08:55 -0700 (MST) From: John Halleck <John.Halleck_at_utah.edu> To: martinl_at_talk21.com cc: cavexml_at_cartography.ch, lfish_at_nyx.net Subject: Re:Groupings In-Reply-To: <20010208171314.YGEN20154.t21mta01-app.talk21.com_at_t21mtaV-lrs> Message-ID: <Pine.GSO.4.05.10102081019560.15462-100000@cor.oz.cc.utah.edu> Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-cavexml_at_karto.baug.ethz.ch Precedence: bulk Reply-To: cavexml_at_cartography.ch
On Thu, 8 Feb 2001 martinl_at_talk21.com wrote:
> [...]
> This sounds like a good reason for passing processed data so that results
> can be compared between different processors. If the loop closure techniques
> result in grossly different answers it would be good to know why! I think
> John Halleck's website has some thoughts about this...
start religious issue
I think that for most purposes any of the tecniques are "good enough", in
absence of blunders, for anything we might be interested in doing. And
if there are blunders the standard surveying techniques involve loop
heristics to fix them, not totally the "adjustment".
Surveying books on Survey Adjustment Computations cover this in much
greater detail than most here are likely to care about.
What one does get out of a correct Least Squares Adjustment are the
statistics that allow you to know how bad the data is. What is done
after that is partly a matter of judgement and experience. I'd trust
some people's years of experience and intuition and about a blunder
much better than I'd trust the statistics out of almost any program.
To a large degree, I don't really care how people process data, if it
is documented and does what it says. I will, however, scream if
it is claimed to be what it is not. "Least Squares" adjustments
that don't produce proper statistics, or that do something formally
invalid [Such as doint X, Y, Z in separate adjustments) will get me
going.
My other religious issue is that cavers sometimes reinvent the wheel,
because they have read only caveing publications rather than going
out and reading real survey books. Several books in my library even
came with already written code to do adjustments (and they agree!).
end religious issue.
In the current context I have an interest in seeing that the information
needed to do a full analysis is preserved. (And you'll note that those
are the issues I mainly reply to.)
I don't think it is at all reasonable for the standard to state HOW
adjusted information is adjusted, as long as it is marked with the method.
It would be nice in a data exchange environment if the data from
several types of adjustments could coexist in the same file.
Start discussion of reality "... it would be good to know why!":
The cases where differing methods produce notibly different results
are generally those cases where there are blunders in the survey.
Treatment of Blunders is a major topic, for which there are many
many different articles discussing details, and most advanced
survey books include some details.
Least Squares techniques (my prefered method) have built in statistical
assumptions. Blunders violate those assumptions. A blind adjustment
of the network produces SOME result, and statistics that show that
result is wrong. Survey books in this case cover reweighting and
relinearization to deal with that problem as well as LS can.
HOWEVER, those same books also cover (loop oriented) methods of
getting the blunders identified so that you don't have bad data
in the adjustment.
Other methods of adjustment (such as "best loops first") come from
different traditions, and have different mathematical properties.
They were in use before LS methods were practical, and have years
of experience (>200 years for some methods) in dealing with practical
issues. They were used because they were "good enough" for even
national surveys before computers.
In the absence of blunders, a proper LS adjustment will produce
the mathematically "most probable" result, which will be close
(I.E. good enough for anything cavers will be doing).
Now consider the case of blunders.
A "best first" adjustment will isolate the blunder to a very
small section of the cave. This will aid identifying the
blunder. The loops in the adjustment have information
valuable to identifying the error.
(Allthough few programs actually do that sort of blunder
analysis).
A simple LS adjustment will smear that blunder all over
everything in the area. (And will put out statistics
so say it did so.) If that is all that is done, the
result may be much worse than the previous method, because
the statistical assumptions being made by the adjustment
are wrong.
(Most cave survey programs doing LS do JUST this...
but are defeneded giving the mathematical justifications
that only apply if the data matches the statistical model)
A proper LS adjustment will reweight (and possibly relinearlize)
the adjustment and adjust again. If there are blunders this
will give an adjustment that is a lot more reasonable, and
more appropriate statistics. Adjustment computation books then
advise that you identify and remove the blunders, using
traditional methods (which generally involve working from loops.
(This is not usually done for cave programs [who cares?]
and commercial surveying programs don't do it because they
expect you to remove the blunders if a simple adjustement
has bad statistics.)
Of course, this doesn't even address the issues of missimplemtations,
or "simplifications" that have undocumented ramifications..
Bottom line: Different programs produce different "adustments".
Some because they are wrong, some just because different methods
produce differing answers in the face of blunders, because they
have differing assumptions. That doesn't mean that they are
"WRONG", only that the data doesn't have the accuracy claimed for
this data.
> [...]
This archive was generated by hypermail 2b30 : Thu Mar 01 2001 - 18:00:00 CET