Dear TEM Segmin'ers
For some time I have been wondering how it is that inversion of TEM data
works at all.
My concern is the statistical weighting that must be given to successive
channels of a TEM system in order to make whatever minimization procedure
you are using take at least some notice of the late time data.
How do you decide on a proper set of weights?
I have asked a number of long-time TEM experts this question and each has
come up with different "rules of thumb". I think everyone agrees that the
weights should be proportional to the inverse of the errors (or noise) in
the data, but at that point opinions diverge.
When I talk to an someone with an engineering bent I often get something
like. "Well, the electronic noise in my system is constant, so as the
windows get wider with time, the noise in successive channels should
decrease as the square root of the window width"
Others with lots of field experience say: "Oh that's nonsense", (most
actually use rather more colourful adjectives), "I've done many of repeated
soundings and have observed that the early time data has much more
variability. You need to use a percentage error at early time and an
additive error at late time to make inversions work" ....... Notice here a
tendency to become more empirical.
Still others might have said. "Well you put the loop down in a slightly
different position and you get huge changes at early time. On top of that
the I'm sure that the orebody looks nothing like the simplistic model I'm
fitting. And anyway I don't trust the guy who collected the data..... I
just adjust the weights until I get something that looks right!"
Now we have moved all the way along the spectrum until we are at the point
of that famous joke about the Mining Engineer, the Geologist and the
Geophysicist who were asked "What is 2+2?".
Well, "What answer do you want?" hardly seems appropriate these days and it
would be nice to have some sound reasons for choosing a weighting scheme.
Where does one turn to find them? Are there any good publications on the
issue?
Your collective wisdom on this matter would be appreciated.
Andy Green
...................................................................
Andy Green
CRCAMET
8 Lawley Crescent
Pymble NSW
AUSTRALIA 2073
Telephone (in Australia) (02) 91 44 13 51
(elswhere) (61 2) 91 44 13 51
Fax As for the telephone
e-mail [EMAIL PROTECTED]
...................................................................
_______________________________________________________
List archive: http://www.mail-archive.com/[email protected]
