Dear UBC-GIF

I am interested in using MAG3D but I know little about what this program can
do, or about the data necessary for inversion. Could you kindly outline
what the data should be?
----------------------------------------------------------------------------
Hello,

To answer your question concerning data files ...

DATA INPUT FILE
The data input file is essentially in XYZ format except for the top three lines.
Most people generate an XYZ file with northing, easting, elevation, and value,
then use a spreadsheet or hand-written software to add the error column. The
inversion problem becomes inefficient if there are more than 1000 or 2000 data
points, so handling data with spreadsheets or custom routines is usually not a
problem. At UBC, the focus is on research and development of new algorithms and
inversion methodologies, so we do not carry out routine processing of large
numbers of data sets.

NUMBER OF DATA POINTS
The solution to the 3D inversion problem involves working with matricies that
have a size of (# data) times (# of discretization cells). For 1000 data points
and an earth model consisting of 10,000 cells this is a fairly large problem -
a matrix with 10,000,000 elements. A great deal of research has gone into
reducing the mathematical size of this type of problem.

STANDARD PRACTICE FOR DATA PREPARATION
(a) discretizing the earth into cells with a size appropriate for the problem
you are trying to solve. The total number of cells is commonly in the range of
10,000 - 100,000. For example a grid that is 50 x 50 x 25 cells would be
considered large but reasonable.

(b) decimating the data so that there are only one or two data points per
surface cell. A reasonable number of data values for inversion is around 1000-5000,
although more can be used at the expense of inversion speed. If the computer
has to continuously hit the disk during the inversion you need more memory or
fewer data points (or a smaller number of cells).

(c) upward continuing the data so that they represent data gathered at an
elevation of roughly half a cell-width,

(d) removing any regional trend that would require the model to include
susceptible material outside the region specified by the mesh.

Step (c) may not be necessary for airborne data if the elevations average
roughly half (or more) of a cell width.

Step (d) may not be necessary if the anomalies are is well contained within the
survey area. If there the anomalies extend off the edges of the data, this step
is very important - see the 14 slides starting at
http://www.eos.ubc.ca/research/ubcgif/tutorials/mag3d/practical/sld001.htm.
Adding padding cells is less important
than for solving the DC resistivity or IP inversion problem (DCIP2D), but it
will help if you have anomalies that extend beyond the data set.

Errors arise from many sources including position errors, elevation errors,
instrument noise, etc. Sometimes it is reasonable to assign a fixed value of nT
for all data points - say around 5% of the average value. Some people apply
errors as a percent of datum plus a fixed value; for example 5%datum + 5nT.

If you have more questions let me know. Most of the information is on our
website under tutorials (http://www.eos.ubc.ca/research/ubcgif/).

Regards,
UBC-GIF