BCRA Grade 5

[Larry Fish]

Bob and Wookey,

>>  And Larry Fish has run statistical analysis on large surveys 
>>  to see how accurate they really are.
>
> Larry was looking for large errors that indicate blunders.  He 
> did [not] try to evaluate accuracy or compare the loop error distribution
> to any assumed measurement accuracy.  He has "standard deviation"
> as one of the choices of criteria for evaluating loop quality.
> I was not able to find any description of exactly how he defines
> standard deviation.

I'm not sure I understand your discussion, but I will just
describe what I did. In COMPASS, the user enters an estimate of
the typical errors that would be produced by the instruments.
These are assumed to be random, gausian distributions. Ideally,
they would be derrived by taking a large number of measurments
with the specific instruments used for the survey and by
calculating the Standard Deviation of the errors. However, in the
case of the work I did, they were just educated guesses.

These values are then projected through the loop or network in
three dimensions. This gives a projected three-dimensional
standard deviation that reflects what you would expect to find if
the survey errors are random and consistant with estimated
instrument errors. (Since there are separate X, Y and Z values,
they could be used to describe an elipsoid. In other words, the
projected standard deviation varies according to the
direction-vector of the error.)

You can then take the actual errors for a loop and decide whether
they are consistant with random errors or whether they are more
likely caused by a blunder. This is process is very similiar to
studing any population of measurements. For example, the average
height of an adult male is 5 foot and 9 inches and the standard
deviation is 2.5 inches. Based on this, if you found a 4 foot 10
inch adult male, you would have an instance that is 4 standard
deviations away from the mean, and you might want to do some
medical tests.

COMPASS has the ability to report the percentage of loops that
fall beyond twice the projected standard deviation. This can be
used to estimate the overall quality of the surveys in a cave.
Using these tools, I processed a very large number of caves and
compared the quality of different caves. It was possible to see
that there was a vast difference in survey quality. For example,
Roppel appeared to have no blunders whereas Lechuguilla had a
large number.

I was also able to infer some things about the typical errors that
are produced by the standard instruments. However, these values
were derrived from a more complicated process and I am not very
confident of the results.

I think there are still many question remaining about survey
errors. The one I am most interest in is: what is the accuracy of
the compass, inclinometer and tape? That is why Wookey's survey
course was so interesting. If I remember right you had experienced
surveys getting compass error as large as six degrees on a relatively
simple, above-ground course.

Larry