
By: Joel E. Rodríguez-Ramírez.
Versamedium Web Services.
info@versamedium.com
Rochester, N.Y. USA.
January 2011.
This tutorial will help you get started with the 2-D Niblett-Bostick
magnetotelluric inversion scheme. The program was thought to be used
by any Internet web browser, ranging from the included in old PC's to
the new smartphone browsers.
The whole process in this tutorial will take in the order of 15 minutes.
You will need the J-format file original, uncorrected, COPROD2 data:
http://mtnet.dias.ie/data/coprod2/cop2_orig_dat.zip.(240 Kb zip compressed).
Download and store it on your computer.
Go to the following web page and register as user:
http://www.versamedium.com/geoinverse/register.html.Registration is free. And only user name, e-mail and password are required.
The user name will keep all your data files and results within the same context.

If you are already registered go to the page below and log into your account:
http://www.versamedium.com/geoinverse/login.html.Once you successfully log into your account, you should see the following image:

Upload the cop2_orig_dat.zip (J-file):
In your web browser click in the Browse box, and select the cop2_orig_dat.zip file
as shown in the following image, press Open.

After the path to the file cop2_orig_dat.zip has been specified in your computer,
click in the Upload box. In your web browser.
Once the file has been Uploaded, the name of the file will appear in a hyper-link:
click the name of the file in the hyper-link in order to start processing. See figure
below.

Select the Magnetotelluric 2-D data:

After selecting the hyper-link, a progress bar will appear, the data parsing process
will take about 1 minute for the current data file.
At this stage in the process you should see a topographic and a satellite
maps, the stations are depicted, and parsing of information over the
stations data files have been performed.


After reviewing the geographic information, proceed to select the horizontal
discretization between stations. Select the 2-Cells discretization option,
and then click the Generate initial model discretization (See figure below).

After selecting the Generate initial model discretization button, a progress bar will appear,
the process will take some 10 seconds for an initial model to be generated.
As shown in the figure below, an initial model with 100 km. depth, will be generated.

Now you can select your desired maximum depth, the discretization scheme
and maximum/minimum periods to be included during inversion. For this tutorial
purpose, choose as shown below.

Once you have selected your desired discretization parameters
Click on the Regenerate the Model with new parameters button.
After a few seconds, you should obtain a new proposed model, like
shown below.

You can repeat the above process until the desired model is achieved.
If you have choosed the proper discretization parameters, and you have
click the Regenerate the Model with new parameters button for the last time.
Then click on the Proceed with the proposed model setup button (shown below).

At this stage the computation of the Linear System of Equations has started (see figure below),
This is a somehow lengthy process and it would take about 6 minutes to complete, please be patient.

At this stage the Linear System of Equations have been generated and we will use
neural computation to find the solution to the system. There is currently 3 Linear
Systems, for modes: Series , Parallel and combined Series & Parallel.
(combined Series & Parallel takes twice as much to complete iterations).
Select the Parallel mode

Starting from a generic initial condition for model values, we start
solving for the Linear System of Equations by selecting the highest Beta=1.0
and a window_size=4x4 values. Then 5 iterations will be performed as shown in
the figure below.
Select the Parallel mode

After the first 5 iterations, choose to perform another 10 iterations over
the current model, as shown in the figure below, repeat the process until RMS error
decreasing rate ``slows down''.

At some point you should obtain a graph as shown below.

Perform another 10 iterations over the current model with window size of 3x3,
shown below are the new parameters and the solution obtained, repeat the process
until the RMS error decreasing rate ``slows down''.


From now on, the process becomes a bit intuitive. A good strategy to achieve
best RMS (lower) is to start with Beta=1.0 and to diminish the window size
from 4x4 to 1x1. Then with the window size=1x1 start diminishing the Beta value,
do it until Beta=0.5 in 10 iteration steps. Then you can alternate between
window_size values 2x2 and 1x1 (with Beta=0.5), as it might diminish the RMS a bit more.
Our final solution is shown in the figure below.

You can leave the process anytime, and go back and continue the
RMS diminishing process. All your data files and graphics will remain
in our servers for further consultation (unless you delete them), whether while
traveling or prospecting on site.
Rodríguez J., Esparza F.J. and Gómez-Treviño E., (2010),
" 2-D Niblett-Bostick magnetotelluric inversion ", Geologica Acta (Vol. 8, 1).
http://www.geologica-acta.com/pdf/vol0801a03.pdf.
Rodríguez-Ramírez J.E., (2010),
http://www.versamedium.com/Versamedium-white-paper.pdf.