A Matlab GUI for use with ISOLA Fortran codes

User’s Guide

by
Efthimios Sokos 1) and Jiri Zahradnik 2)

1) University of Patras, Seismological Laboratory

2) Charles University in Prague, Faculty of Mathematics and Physics

esokos@upatras.gr , jz@karel.troja.mff.cuni.cz

Ver. 2.5
June 2006
Contents

GUI PURPOSE................................................................................................3

REQUIREMENTS ...........................................................................................3

INSTALLATION .............................................................................................4

METHOD .........................................................................................................4

RUNNING THE GUI.....................................................................................10

Crustal Model ....................................................................................................................................11

Event Info...........................................................................................................................................13

Station Selection..............................................................................................................................14

Raw Data Preparation....................................................................................................................15

Trial Source Definition ...................................................................................................................18

Green Function Computation ......................................................................................................20

Inversion.............................................................................................................................................22

Plotting ................................................................................................................................................25

Utilities I and II................................................................................................................................31

RUNNING THE FORTRAN CODES WITHOUT MATLAB ..........................33

ACKNOWLEDGMENTS ................................................................................33
GUI Purpose

The ISOLA GUI has been created using the Matlab GUIDE tool. Its purpose is
to help the user of ISOLA-FORTRAN part with data preprocessing, Green
function preparation, waveform inversion and display of the results. In
addition to the main tools a few utilities are also present that help the user to
inspect the quality of the data using filters, convert to displacement, etc. In
utilities there are also two other options that could help: Backup files and
Create folder structure. The Backup files option saves in a user specified
folder a number of files that are critical for running the inversion; in this way
the user can keep track of various attempts during an experiment. The Create
folder structure option creates the folders needed for ISOLA if these aren’t
present. Finally the import SAC and GCF (Guralp Compressed Format) options
help converting SAC and GCF binary files in ISOLA ascii format (4 columns of
ascii data per station, i.e.,Time, NS, EW and Z component).

Requirements

In order to use ISOLA GUI, you need to have the following software installed
• Matlab 6.5 or 7.0 (the software has not been tested using other
versions of Matlab)
• Generic Mapping Tools (GMT, Wessel and Smith, 1991, 1998)
(the software was tested using version 4.0 and 4.1, other
versions should work also )
• M_Map, this is a freeware Matlab based package, it is used in
map creation within Matlab, you can download from
http://www2.ocgy.ubc.ca/~rich/map.html and you have to add
it in the Matlab path.
• Ghostscript and GsView for Postscript files display
http://www.cs.wisc.edu/~ghost/, and you have to add it in the
system path.
 .

Installation

Installation is easy; just unzip the zipped distribution file in a folder. This will
create the necessary folders and extract the files needed for running the
Matlab GUI. After this operation you should get a folder structure like the
following:
\isola
\gmtfiles
\green
\invert
\pzfiles
\polarity

A typical location would be the Matlab folder e.g. c:\matlab\isola.

Method

ISOLA is based on multiple point-source representation and iterative

deconvolution method, similar to Kikuchi and Kanamori (1991) for teleseismic
records, but here the full wavefield is considered, and Green's functions are
calculated by the discrete wavenumber method of Bouchon (1981). Thus the
method is applicable for regional and local events. Instrumentally corrected
band-pass filtered velocity records are used. The code transforms velocity into
displacement, inverts the displacement, and provides synthetic displacement.
Seismogram s(t) is approximated by a combination of elementary
seismograms ei(t), i=1,2,...6, i.e. s(t)= sum( ai . ei(t)). The elementary
seismograms correspond to basic focal mechanisms as follows:
i strike dip rake
----------------------
1 0 90 0
2 270 90 -90
3 0 90 90
4 90 45 90
5 0 45 90
6 isotropic

Seismograms (3-component), from a set of stations, represent the data, d,

while the coefficients ai are the parameters to be found, m. The linear inverse
problem d= G m is solved by the least-square method: m = (GT G)-1 GT d,
where GT is G transposed and ()-1 is the inverse of the so-called system
matrix (GT G). Eigenvalues of the system matrix (GT G) are calculated to
understand how well-posed (or ill-posed) the inverse problem is. Ratio of
min/max eigenvalue is reported. The lower the value, the worse the results
are (although, formally, the quality of the fit may be good, e.g. variance
reduction can be large). Question how the individual stations contribute to the
well-posed inversion has no simple answer. It is to be carefully analyzed case
by case.

Optionally, the weighted inversion is performed, with weights prescribed by

user, different weights for different stations and components. As a rule, up-
weighting records with small amplitudes, and down-weighting large
amplitudes, is recommended. Another strategy is to up-weight stations whose
seismograms appear to be the most sensitive with respect to small
modifications of the problem.
The coefficients ai are related to moment tensor Mpq (where p,q refer to
geographic coordinates x (>0, N), y (>0, E), z (>0, up)):

Mxx=-a4+a6
Myy=-a5+a6
Mzz=a4+a5+a6
Mxy=Myx=a1
Mxz=Mzx=a2
Myz=Mzy=-a3

Moment tensor components in spherical coordinates r, t(=theta), p(=phi),

used in Harvard catalogue, or in GMT, are given by:

Mtt=Mxx
Mpp=Myy
Mrr=Mzz
Mtp=-Mxy
Mrt=Mxz
Mrp=-Myz

Eigenvectors of the moment tensor provide strike, dip and rake. Eigenvalues
provide scalar moment Mo and decomposition of the moment tensor into
three parts: double-couple (DC), compensated linear vector dipole (CLVD),
and voluminial (VOL).

Possible moment tensor (MT) inversion modes are as follows:

-full MT inversion: all six basic focal mechanisms, 1,2,...6; DC+CLVD+VOL
-deviatoric MT inversion: basic focal mechanisms 1,2,...5; DC+CLVD;
VOL%=0;
-DC-constrained MT inversion: DC only; VOL%=CLVD%=0
-known and fixed DC moment tensor (only position and time is searched)
Opposed to strike, dip, rake and Mo, the DC% is strongly unstable. It closely
relates with the fact that a large deviation from pure DC has, as a rule, a very
small effect upon the seismograms. Indeed, it can be demonstrated by many
examples in which synthetics for, say DC%=50 and DC%=100, cannot be
distinguished by eye from each other, and quantitative measure of their fit
with real data is nearly identical (e.g. variance reduction, introduced below,
differs by 1%). Therefore, we can say that the DC% value provided by the
routine moment tensor inversion is rarely of any physical meaning. The same
applies for the volume percentage VOL%.

Optionally, the inversion can be constrained through the Lagrange multipliers

to give a nearly 100% DC solution. The (non-linear) condition that the
moment tensor must have zero-valued determinant is solved iteratively. It
slows down the calculation and may create problems. There is no special
advantage in this option, because the strike, dip and rake of the DC-
constrained solution is usually very near to that of the (unconstrained)
deviatoric MT inversion.

Thus the most recommended inversion mode is deviatoric inversion (VOL%=0)

in which, however, the resulting DC% is considered with great caution.

It is assumed in the above explanation that the source position and time are
known. This code allows their optimization through a grid search over set of
trial source positions and time shifts, pre-defined by the user. The search
seeks the trial source position and time for which the residual error of the
least square method is minimized. It is equivalent to maximize correlation
between the observed and synthetic seismograms. The concept of optimizing
position and time is applicable to each point element (the so-called subevent)
of the generally more complex source model. If the whole earthquake is
represented (in the low-frequency range) by a single source, the optimum
solution provided by the grid search corresponds to the centroid (the centroid
position, centroid time and the centroid MT).
User can check the correlation values as a 2D function of the trial source
position and time. The code has an interactive part in which user can use
automatic search of the optimum correlation, or can make his preferred
choice. The preferred choice may be the source position or time with a
somewhat lower correlation, close to formal maximum, in which however
some other conditions (constraints) are better satisfied. In this way user can
apply, for example, his knowledge about the first-motion polarities,
geologically deduced fault dip, various assumptions about rupture
propagation, etc. The 2D correlation may also help in identifying poor
resolution of some parameters, for example the down-dip source position.

Narrow band-pass inversions around period T can often provide two

correlation maxima, in time t and t+T/2, characterized by rake and
rake+180° respectively. This is just the case where even a single guaranteed
first motion polarity may help to decide between proper time, t or t+T/2.

The code works iteratively. A point source contribution is called subevent.

Using data code searches for subevent 1. Once subevent 1 is found, its
contribution is subtracted from the data to get the so-called residual data. In
the next step the residual data are processed in the same way as the original
data during the first step. Subevent 2 is found, etc. Original data minus the
last residual seismogram provide the resulting synthetic seismogram.

Optionally, the search of spatial and temporal distribution of subevents can be

performed with fixed (100% DC) moment tensor. The user prescribes the
known (or assumed) strike, dip and rake, and these values are kept fixed for
all subevents. It can be recommended when the unconstrained moment
tensors vary chaotically from one subevent to the other.

When the solution is represented by several subevents of different focal

mechanism, the resulting scalar moment cannot be obtained as the sum of
the subevent scalar moments. That is why we also provide an output file
where, subsequently, the subevent contributions are summed up as tensors
and the scalar moment of the resulting (cumulative) tensor is shown.

Denoting data as d, and synthetics as s, the fit between them is measured by

the so-called variance reduction varred= 1 - |d-s|2/|d|2, where |.| denotes
the L2 norm, e.g., |d|2=suma(di2). The norm is calculated as a single number
summing up over all time series, all components, and all stations. The
optimum value of varred is 1. Wrong synthetics may give even varred < 0 (no
physical meaning).

The variance reduction is determined repeatedly, “when building up” the

synthetics, i.e. after adding every subevent. In other words, when s1, s2,…
are synthetics corresponding to individual subevents, then first we calculate
varred from d and s=s1, then from d and s=s1+s2, etc. As such, varred is a
measure of the convergence of the iterative deconvolution process. Slow
increase of varred with adding subevents indicates that additional subevents
do not further improve the fit.

At this point it is useful to clarify how the variance reduction relates with the
correlation. The relation is simple when all stations and all components are
used, and all weights equal 1, and when we calculate only the first subevent.
Take the observed seismogram as data d, and the first subevent as synthetics,
s=s1. It is an internal property of the least-square minimization of |d-s|,
during the corresponding matrix operations, G, GT G, etc., that we
automatically get the value of c2, where c is the correlation between d and s,
without numerically calculating the correlation integral. (Speaking more
precisely, we get sqrt(c2)=abs(c), i.e. not the correlation c, but only its
absolute value. Anyway, hereafter, we simply say only “correlation”). At the
same time, it can be shown that in c2= 1 - |d-s|2/|d|2 = varred.
When we calculate already the second subevent, s2, the least-square method
minimizes |r-s2|, where r is the residual seismogram defined as difference
between the original data and the first subevent, d-s1. Therefore, from this
least square run, we get automatically, and we report on the code output,
(the absolute value of) the correlation between r and s2. At the same time,
after adding the second subevent, the code calculates varred using d and
s=s1+s2, i.e. varred= 1- |d-(s1+s2)|2/|d|2. Obviously, the varred value
reported after the second subevent is not the same as the (squared)
correlation between r and s2. Moreover, this varred is also not the same as
the (squared) correlation between d and s=s1+s2, because we did not find
the synthetic s=s1+s2 by a single least-square minimization of |d-(s1+s2)|.

When non-unit weights are used, and/or some stations are deselected from a
particular inversion run, the relation between correlation and variance
reduction gets even more complicated.

Running the GUI

In order to run the code, within the Matlab command window, change
directory (using the cd command) to the \isola folder and type isola. This
will run the main GUI file and you should get a form like the following (Fig. 1).
Figure 1. Main form of the ISOLA Matlab GUI.

Crustal Model

The first step in the inversion procedure is the definition of the crustal model.
This is done by pressing the Define Crustal Model button in the main form,
which will launch the form shown in Fig. 2. There is a fixed option for a
maximum of 15 layers in the crustal model. For each layer, the user should
give the depth of the layer top, the Vp, Vs velocities in km/sec, density in
g/cm3, Qp, Qs and, optionally, a title. In order to allow the program to
understand the last layer, the user should type 999 in the last row as shown
in Fig. 2.

There are buttons for saving the data (Save) in a text format file suitable for
ISOLA Fortran, as well as reading it back in the form (Read). The latter is
useful for editing a previously saved model. The plotting option (Plot)
generates graphs of Vp and Vs versus depth. Automatic calculation of Vs
according to a specified Vp/Vs value is also possible, as well as the calculation
of density according to formula density (g/ cm3) =1.7+0.2*Vp (km/s).

Pressing Save the crustal model file that was created is saved under the
GREEN and POLARITY folders.

Figure 2. The Crustal model definition form.

The native format of the crustal model file in ISOLA Fortran is the following:
Crustal model Tomography model
number of layers
8
Parameters of the layers
depth of layer top(km) Vp(km/s) Vs(km/s) Rho(g/cm**3) Qp Qs
0.0 5.47 2.700 2.560 300 300
2.0 5.50 2.860 2.800 300 300
5.0 6.00 3.230 2.940 300 300
10.0 6.20 3.240 2.940 300 300
15.0 6.48 3.400 2.980 300 300
20.0 6.70 3.800 2.980 300 300
30.0 6.75 3.810 2.980 300 300
40.0 8.00 4.660 3.360 1000 1000
*************************************************************************
After finishing with the crustal model it is time for the user to give some
information related to the event he wants to analyze; this is done by pressing
Event Info in the main form (Fig. 3)

Event Info

Figure 3. The Event Info form.

In this form the user provides information that will be used in all the
subsequent steps of the inversion, like the event Lat, Lon, Depth,
Magnitude, Date (the last two are included just for information), Origin Time
of the event, Start Time of the data file (seismograms), Sampling Frequency,
and, finally, the Time Length of the data that will be used for the inversion.
The time length should satisfy some specific conditions. For user’s
convenience, there are 3 predefined time length intervals that should cover
most applications at local and regional distances. After pressing the update
button the GUI writes or updates the appropriate files e.g. event.isl,
rawinfo.isl, duration.isl; those are ISOLA (*.isl) related text files that hold
the above information and are stored in ISOLA folder.

Next step is the selection of the stations that will be used for the inversion
(Fig. 4).

Station Selection

Figure 4. The Station Selection form.

Options in this form are simple; the user should press Make Map - Select
stations in order to select his (previously prepared) text file containing the
geographical coordinates of his network. The name and location of the file is
arbitrary, the user will be guided to browse. The file should be an ascii file
with three columns, Station Name, Latitude, Longitude, separated by
spaces. The Station Name should have 3 characters. If it is longer a warning
is issued and the station name is reduced to the first three characters.
Pressing Make Map – Select stations (and assuming that M_MAP is properly
installed) a map of the station distribution is created, and the user can select
the stations he wants to include in the inversion. Selection of the stations is
done using the left mouse button; the last station should be picked using the
right mouse button. After pressing the right mouse button a warning window
comes up with the total number of the selected stations. Later during the
inversion stage, if necessary, the user can again “deselect” some stations, i.e.
to remove them from the inversion process, but once the station was selected
the corresponding seismograms must be provided.

After finishing the selection of the stations the user should just press Exit on
the Station Selection form. Doing this the GUI will create files like
stations.isl, station.dat, source.dat (in GREEN folder) and allstat.dat in
INVERT folder.

Raw Data Preparation

Next step is the Raw Data Preparation (Fig. 5), maybe the most important
option of the GUI, since it processes the seismograms (doing the instrument
correction, origin time alignment, resampling) and produces data files ready
for the inversion. The data, i.e. the uncorrected seismograms, must be
available in 4-column text files (with the strictly required column order, i.e.,
Time in seconds , NS, EW, Z in counts, separated with spaces), one file per
station. The filenames are not completely arbitrary; for example, for the LTK
station the filename must start with the letters LTK, where LTK is the station
name used in the Station Selection. The remaining part of the filename
(including the extension) is already arbitrary. The location of the file is also
arbitrary as well as the extension and the part of the filename after the first
three characters, the user will be guided to browse.

The user should make available also the files containing the parameters that
will be used for the instrument correction. An example of such a “pole and
zero file” follows. A few pole and zero files for several instruments are
included with the software distribution also.

A0 Comment
133310 A0 value
count-->m/sec Comment
3.33e-10 conversion counts to m/sec
zeroes (Re and Im in rad/sec) Comment
3 Number of zeroes
0.0 0.0 Zero
0.0 0.0 Zero
51.5 0.0 Zero
poles (Re and Im in rad/sec) Comment
5 Number of Poles
-272 218 Pole
-272 -218 Pole
56.5 0 Pole
-0.1111 0.1111 Pole
-0.1111 -0.1111 Pole

It is strictly required that the file describes the instrument transfer function
for input velocity. The filename must be a combination of the Station Name
and the “BH*.pz” string; for example, for the station named LTK and the EW
component, it must be LTKBHE.pz, thus three pole and zero files per station,
are needed. The location of the “pz-files” is not arbitrary, the user must place
them in the PZFILES folder.

Caution must be devoted to the appropriate units. ISOLA needs the poles and
zeros in rad/sec, and the constant of the transfer function A0 (guaranteeing
that the “plateau”of the amplitude response equals to 1) should be consistent
with that. In case that the user has the zeros and poles in Hz, they all must
be multiplied by 2π, and, correspondingly, A0 must be multiplied by (2π)NP-NZ,
where NP and NZ is the number of the poles and zeros, respectively.
Figure 5. The Data Preparation form.

Pressing Load Ascii file, a standard open file dialog appears and the user
selects a data file. When the file is read, the sampling frequency is displayed
(in red) at the top of the form (Fig. 5), side by side with the resampling
frequency of the data in blue. The resampling is made automatically, based
on specific methodical requirements and the data input from the previous
steps.

After data loading the Cut and Instrument correction buttons are enabled. If
not all of the data is needed using the Cut button the last part of it can be
deleted. By pressing the Inst. Correction button the program searches in
PZFILES folder for the appropriate pole and zero file and then performs the
instrument correction. During this operation a filter can be applied
(Butterworth, 2nd order, zero-phase) along with the DC and trend removal, all
using the built-in Matlab functions. The baseline option is an alternative of the
DC removal, in which the baseline is determined from the beginning of the
seismogram only.
Next step is the time alignment of the seismogram, i.e., irrespectively of its
Start Time, since now it will begin at the earthquake Origin Time; this is done
by pressing the Origin Align button which becomes active after the
instrument correction.

Finally the user should press the Save Data button in order to save the files
in INVERT folder. The files are saved as text files (4 columns, Time in sec,
NS, EW, Z in m/sec), one per a station, always containing just 8192 points of
the appropriately resampled data. The GUI automatically creates the proper
filename for the corrected resampled data, which for station LTK would be
LTKraw.dat. This filename must not be changed by the user.

Trial Source Definition

The final data that the user should prepare is the trial source location and this
is done by pressing the Seismic Source Definition button in the ISOLA
main form. The user has to select Single Source or Multiple Source
definition.

In the case of single source, we can vary the depth only, thus the options in
the form are Starting depth, Depth step and No of Sources, the explanation of
which is given in Fig. 6. In this case, all the trial positions are below the
epicenter. By pressing Exit the program writes the appropriate files in
GREEN and GMTFILES folders. These files are src*.dat files in GREEN and
sources.gmt in GMTFILES. The sources.gmt file is formatted for GMT
(psxy) plotting.

It is assumed that the Single Source option will be always used to retrieve (at
each trial depth) just a single subevent, that one which predominates in the
low-frequency inversion. The grid search over the depth then serves to
provide the optimum source position characterized by the best fit between the
waveform data and synthetics. Thus we get a first approximation of the
centroid. By “the first approximation” we mean that (opposed to assumption
of this option) the centroid is not necessarily below the epicenter.

Starting Depth

Depth step

Figure 6. The Single Source definition form.

If the user wishes to use the Multiple Source inversion a new form appears
(Fig. 7). This inversion has three options: sources along the strike of the fault,
sources along the dip, or sources on a plane. The user has to select the fault
strike and dip, the number of the trial sources he wants to try, their spacing,
and the first source position. After pressing the Calculate button the
program writes the source files in GREEN folder and produces a map with
the trial source distribution.

The Multiple Source option has two different applications. The first one is to
prepare trial positions for the grid search of the centroid, inspecting usually
positions in the vicinity of the optimum depth found with the Single Source
inversion. In this application we do design a Multiple Source stencil, but make
inversion in such low frequencies that just a single subevent represents the
whole source. The second possible application of the Multiple Source option is
to design the trial source positions for inversion of the earthquake into a
complex source model, consisting of more than a single subevent. Obviously,
as a rule, the latter is achieved by the inversion including higher frequencies
and using a denser grid of the trial positions than of the first application.

This manual describes the code options. It cannot describe all possible
strategies how to use the code. For example, the user may use the Multiple
Source option in the beginning of his study to find the centroid (the first
application), which might need trial positions relatively far from the
hypocenter, and with a relatively large grid spacing. Then he may wish to
repeat the same procedure with finer grid spacing to find the exact position of
the centroid. Finally, he can try the application of the second type, i.e. to
resolve several subevents. The latter may often fail, due to lacking data, or
simply because the earthquake has no clear complexity, then his result
remains to be just the precise centroid solution (with no additional details).
This discussion explains why it is nearly impossible to give an “always working
hint” of where and how densely the trial source positions should be designed.

6
S trike
4
} 5
S pa c in g
3
2
Sh ift to N orth

1 Ne w Hy p oc ente r a t so u rc e 1

Sh ift to E a st
H yp o ce nte r

Figure 7. The Multiple Source definition form.

Green Function Computation

After finishing with the trial source definition the inversion can start and the
first step is to prepare the Green functions. This is done by pressing Green
Function Computation in the ISOLA main form. The greenpre.m is
running (Fig. 8) and, for simplicity, the user has only one option to fill in, this
is the maximum frequency of the Green functions to be computed (fmax).
(Obviously, no higher frequency than this one can be requested later in the
inversion.) Pressing Run the necessary files are created in GREEN folder and
the executable Fortran codes are called from Matlab: gr_xyz.exe and
elemse.exe, for the Green function and elementary seismograms,
respectively. The execution is monitored through the Command Window. The
user should watch the execution of the Fortran programs and press OK in the
“Copy files in invert folder” dialog, when the procedure ends. After that,
copying starts and a message is issued when the files are copied in INVERT
folder and the user can proceed to the inversion.

For simplicity, this version of the code does not specify any particular slip-
rate time function. It means that the assumed slip rate is that given by the
Green function calculation. By definition, the Green function is response of the
medium to the impulsive (Dirac delta function) source, but note that we work
in a limited frequency band. Therefore, working with the band-limited delta
function, the actual slip-rate function applied in the code has duration
approximately equal to dur=1/fmax, where fmax is the maximum frequency
considered in the inversion. The only correct application of this code is for
subevents whose true durations are shorter than dur.

Figure 8. The Green function computation form.

Inversion

The last step starts by pressing Inversion in the ISOLA main form; it runs
invert.m that produces the following form (Fig. 9).

Figure 9. The Inversion form.

Before running the inversion, the use has to select his options regarding the
Filter, Type of Inversion, Number of Subevents to be retrieved, and
parameters of the Time Search (temporal grid search of the subevents). The
same filter is automatically used for the data and synthetics.

Then the user may press the Compute weights button which automatically
calculates weights to be applied in the inversion. (If not pressing this button,
unit weights will be used.) The automatic weights are inversely proportional
to peak values of the individual components in the used frequency band. In
fact what are computed by the Compute weights function are just the peak
values for the traces and latter isola.exe takes the 1/peak value as the weight.
Finally, pressing Run, the main executable Fortran code isola.exe is running
in the Command window (Fig. 10). When the code makes a pause it prompts
the user to accept the automatic selection of the trial source position (that of
maximum correlation), or to select some other user’s preferred trial source
position. To simplify the decision, the user can plot a correlation diagram
using the Plot Correlation option (Fig. 11); GMT and Gsview is needed. In
this 2D diagram the correlation value and focal mechanism are plotted against
the trial source position and time shift for the subevent under study.

Pressing the Deselect Stations button, the allstat.dat file opens using
notepad, and the user can remove a station from the inversion by editing
the second column of the file (1=use the station, 0=not use it). The columns
3, 4, 5 contain the above mentioned peak displacement values for the NS, EW,
Z component, respectively. The user can manually edit them, thus making his
preferred specification of weights. Note that the larger the number in
allstat.dat, the lower is the weight.

See also the Method chapter to learn more about the weights and correlation.

Example of the allstat.dat file.

GUR 1 6.057019e-005 2.130505e-004 3.082369e-004
LTK 1 6.179501e-005 2.366762e-004 2.285469e-004
DID 1 1.139388e-004 1.451932e-004 1.765126e-004
PYL 1 1.145303e-004 1.020548e-004 1.030644e-004
Figure 10. Example of the inversion run.

Figure 11. Example of the correlation plot.

Plotting

The inversion code isola.exe creates a number of output files that can be
plotted after the inversion run. This is done by pressing Plot Results button
of the main ISOLA form. The plotres.m is called and produces the form of
Fig. 12. There are options for plotting Synthetic versus Real data, the MT
results, as well as for comparing the solution with first motion polarities.

The MT (moment tensor) plotting is done using GMT, and the user has the
option for a single source summary plot (Fig. 13) a plot of Correlation versus
Source number and DC% (Fig. 14) or multiple source results plot in map view,
pressing Plot Sources. In all the plots (except the one in the single source
summary) the fault plane solutions are displayed by beach balls without
showing the non-DC component, because, as explained above, the non-DC
part is usually unstable.

Note that the single-source plotting option applies also for the inversion which
resulted in several subevents. The code prompts the user to specify the
subevent he wishes to plot in Fig. 13 and Fig. 14.

The variance reduction given in the Single Source plot in Fig. 13 then refers to
the situation after adding the chosen subevent (for example, when plotting
subevent 2, the varred value applies for the fit between data and the
synthetics composed of subevent 1 and subevent 2). The fit is calculated only
from stations used in the inversion (the deselected stations are not
considered).
Figure 12. The Plotting form

Pressing the Plot Sources button the plsource.m opens. This creates the
form of Fig. 15. The user has a lot of options in order to create a map with
the results (Fig. 16). Most of the options are self explanatory, but some
knowledge of GMT is needed. The most critical are the Beach ball shift and
Beach ball X shift to prevent overlapping of the beach balls. The plot formally
includes all the calculated subevents. Lot of care is needed to find out which
of them are physically justified. This complicated physical problem is not
solved here in the manual. See, for example, Zahradnik et al. (2005).
Figure 13. The Single source summary plot

Figure 14. The Correlation vs Source number plot. By the source number we
mean the sequential number of the trial source position. The values above the
beach balls give the time shift.
Figure 15. The multiple source plotting form .

Figure 16. Plot of the multiple source inversion results.

Full numerical results of the inversion are available in the files inv1.dat,
inv2.dat, inv2c.dat, inv3.dat, inv4.dat, located in the folder INVERT. For
a detailed explanation, the readme files of the ISOLA Fortran guide should be
consulted (see below). Most often, simpler output (that provided in the
Plotting form) is sufficient.

Finally the user has the option to plot the MT inversion results along with
selected polarities (Fig. 17) in order to check validity of the inversion. A few
files need to be updated for this option. Namely the station.dat file has to
be manually edited to assign the observed polarities per stations. This can be
done within the GUI by pressing Input Polarities after which the station.dat
file opens in an editor. The other two files that need to be created are
onemech.dat and moremech.dat that the user has to create in the
POLARITY folder. The onemech.dat contains data to be shown in red (e.g.
the solution from ISOLA), while moremech.dat can include, for comparison,
several other solutions (e.g. those from several agencies). The polarity
plotting is still under heavy development. Present version needs some
knowledge of the ISOLA Fortran manual. In particular, the mentioned files
should share the same format as that of the output file INV2.DAT.
Figure 17. Plot of polarities with inversion results.

Moremech.dat example (a space-sepated text file; only columns 4,5,6 =

strike,dip,rake are important here)
4 6.66 .185761E+18 151. 76. 77. 13. 19. 131. 251. 61. 44. 32. 88.9 .7718E+00
1 3.96 .560486E+17 201. 50. 144. 316. 63. 46. 76. 83. 175. 39. 68.4 .8426E+00
Utilities I and II

Besides the inversion related tools, a number of options is provided for

assisting the user in converting his binary data to ISOLA ascii format and
inspecting the data quality. Tools are available for importing data in two
formats: SAC and GCF (Guralp Compressed Format). The tools are quite easy
to handle; just read the files, component by component, and then save the
file (Fig. 18). If the three components at a station don’t share the same
starting time, or don’t have the same length, options for their trimming are
provided.

Figure 18. The GCF file import form.

The tools for importing data are based on free Matlab routines from the Web
(see acknowledgment part of the manual). The user can use them as an
example to easily add his preferred format conversion to the present GUI.

After converting the data to ISOLA format the user has the option to inspect
the data quality either using the Inspect Data tool or the Try Filters tool. Data
can be filtered in various frequency bands in order to define the proper band
for inversion (i.e. the band with a good signal-to-noise ratio). Data can be
also inspected as the displacement. It is extremely important to inspect the
displacement without filtration and instrument correction since it allows
detection of various disturbances that are often hidden in the velocity records
and/or in the displacement band-passed records (Zahradnik and Plesinger,
2005).

Finally the data can be shifted using the Shift data option. This option may be
used (with great caution) when, for example, we have an indication that the
employed crustal model is not appropriate at some station as regards the
arrival time.
Running the Fortran codes without Matlab

ISOLA can be used also without the Matlab GUI. In case like that the user
should study the readme files that are provided with the Fortran files
distribution at http://???

For compactness of this Guide all the readme files of the ISOLA Fortran
version are copied below. Users of the Matlab GUI can skip this part.

Instruction:
Go trough directories and follow readme files there.
Start in DATA, continue with SOURSTAT, GREEN, INVERT,
POLARITY (and, optionally, also SYNT). Use sample (input
and output) data files to learn the package. The sample
files do not contain *.exe files because of their length
and because users should be able to compile all *.for
codes themselves. In case of problems, ask the author to
send the *.exe files. The example is highly simplified to
allow quick test run. For more details, see the
application referenced below.

Acknowledgments

Special thanks go to Petra Adamova for extensive testing of ISOLA GUI. The
following m-files are part of other programs that we found useful and added
to the GUI.
• rsac.m, lh.m by Michael Thorne,
 • readgcffile.m by M. McGowan, Guralp Systems Ltd.,
• Bandpass.m Coral distribution.

In the Fortran part, several subroutines were used, written by other authors:
M. Bouchon, O. Coutant, J. Sileny, J. Jansky and O. Novotny, J. A. Snoke, P.
Reasenberg and D. Oppenheimer. Codes from Fortran Numerical Recipes
were also used. This work was supported by the EC project 004043-3HAZ-
Corinth and by the Charles University in Prague grant, GAUK 279/2006/B-
GEO/MFF.