The computation of the elementary synthetic seismograms
for Isola MT inversions in Dobrá Voda region
using the 3D finite-difference method
AIM second annual meeting
September 29-30, 2011 Prague
in cooperation with industrial partner:
Juraj SEKEREŠ Dagmar SEKEREŠOVÁ
Progseis, Ltd., Trnava, Slovakia
Martin GÁLIS1,2 Lucia FOJTÍKOVÁ1,3 Jiří ZAHRADNÍK4
1 Geophysical Institute, Slovak Academy of Sciences, Bratislava2 Comenius University Bratislava
3 Institute of Rock Structure and Mechanics,Academy of Sciences of the Czech Republic, Prague
4 Charles University Prague
introduction
the area of our interest
used names
Dobra Vodaor
Little Carpathiansor
Male Karpaty
the region withsignificant seismic activity
with respect to Slovak territory (1906 Ms 5.7)
3 methods for moment tensor inversionwere tested and compared
with respect to their accuracy and stability :
FOCMEC Pg and Pn wave polarities
AMT P-wave amplitudes
ISOLA full waveform
introduction
motivation
introduction
motivation
recent earthquakes
national network
MKNET
introduction
ISOLA
first step
synthetic elementary seismogramsare computed for a set of stations
elementary seismogram=
ground motion at the stationdue to one elementary focal mechanism
first step
synthetic elementary seismogramsare computed for a set of stations
elementary seismogram=
ground motion at the stationdue to one elementary focal mechanism
second step
the coefficients of the linear combinationof synthetic elementary seismogramsare find by minimization of the misfit between the real data and synthetics
introduction
ISOLA
introduction
motivation
ISOLA is distributed as a packagewith a programs to computethe elementary seismograms
only for 1D medium
the limitations of 1D mediumwith respect to realistic 3D structures
are obvious
therefore we decided to try to use ISOLAwith synthetics for 3D medium
from model definition to computational model
model by Geofyzika, 1985
extended model
from model definition to computational model
spline interpolation- presence of false details, high-frequency content
2D interpolation
from model definition to computational model
spline interpolation- presence of false details, high-frequency content
bi-linear interpolation on quadrilaterals- presence of stair-like structures
2D interpolation
from model definition to computational model
spline interpolation- presence of false details, high-frequency content
bi-linear interpolation on quadrilaterals- presence of stair-like structures
linear interpolation on triangles with the same orientation- generally good results, but still some local sharp corners
2D interpolation
from model definition to computational model
linear interpolation on triangles with adapted orientation
for each quadrilateralthe orientation
with smaller gradient at the triangle’s contactis used for the interpolation
2D interpolation
from model definition to computational model
2D interpolation
from model definition to computational model
linear interpolation on triangles with adapted orientation
we applied 2D interpolation in successive stepsto xy, xz and yz planes
to obtain a 3D grid with spacing 100m
we applied volumetric smoothingto remove the artifacts of linear interpolation
(discontinuous derivatives at the triangle edges)
3D interpolation and smoothing
from model definition to computational model
model visualization
4
2
6
3
5
Vp[km/s]
51 km
31 km
depth = 0 km
from model definition to computational model
4
2
6
3
5
Vp[km/s]
51 km
31 km
depth = 1 km
model visualization
from model definition to computational model
4
2
6
3
5
Vp[km/s]
51 km
31 km
depth = 2 km
model visualization
from model definition to computational model
the model by Geofyzika is definedonly in terms of Vp (P-wave) velocity
for 3D model we used the same Vp / Vs ratio as for 1D model:
Vp / Vs = 1.75
Vp to Vs
from model definition to computational model
-0.50
0.00
0.50
1.00
1.50
2.00
2.50
3.00
3.50
4.00
0 1000 2000 3000 4000 5000 6000 7000 8000
ND Cornwell
ND Brocher
Gardner (standard)
1D model
den
sity
[g
.cm
-3]
Vp [m.s-1]
Vp to density
from model definition to computational model
we used FD scheme and program developed in Bratislavamainly by Peter Moczo and Jozef Kristek
finite-difference method
brief info
we used FD scheme and program developed in Bratislavamainly by Peter Moczo and Jozef Kristek
finite-difference method
brief info
in two recent exercises,ESG2006 and E2VP-Cashima project,
participated teams using different numerical methods(finite-difference, finite-element,
spectral element, discontinuous galerkin, pseudo-spectral)
we used FD scheme and program developed in Bratislavamainly by Peter Moczo and Jozef Kristek
finite-difference method
brief info
in two recent exercises,ESG2006 and E2VP-Cashima project,
participated teams using different numerical methods(finite-difference, finite-element,
spectral element, discontinuous galerkin, pseudo-spectral)
these exercises showed thatif our FD scheme was applicable to the problem configuration
(for example if planar free surface was an acceptable approximation)it was the most accurate and the most efficient method
V14
results
0 3 6 9 12 15 18-0.00004
-0.00002
0.00000
0.00002
0.00004-0.00004
-0.00002
0.00000
0.00002
0.00004-0.00004
-0.00002
0.00000
0.00002
0.00004
0 3 6 9 12 15 18
A
E
I
comparison of 1D and 3D synthetic seismogramsSMOL – very close bedrock station
results
Z
NS
EW
0 3 6 9 12 15 18-0.000004
-0.000002
0.000000
0.000002
0.000004
0.000006-0.000004
-0.000002
0.000000
0.000002
0.000004
0.000006-0.000004
-0.000002
0.000000
0.000002
0.000004
0.000006
0 3 6 9 12 15 18
comparison of 1D and 3D synthetic seismogramsHRAD – distant bedrock station
results
Z
NS
EW
0 3 6 9 12 15 18-0.0002
-0.0001
0.0000
0.0001
-0.0002
-0.0001
0.0000
0.0001
-0.0002
-0.0001
0.0000
0.0001
0 3 6 9 12 15 18
results
SPAC – distant station on sediments
comparison of 1D and 3D synthetic seismograms
Z
NS
EW
conclusions
we prepared the 3D computational modelof the Dobra Voda regionfor 3D FD computations
(however the model should be considered as very preliminary)
we applied the 3D FD method to compute3D synthetic elementary seismograms
for ISOLA moment tensor inversion
we tested this procedureon two events, V03 and V14
(the results of the analysis will be presented in next talk)
we are ready to apply this procedureto more events
with the same modelor with improved model once it will be available
thank you for the attention