Title: Erdbeben%20in%20der%20Schweiz%20fr
1Blindtest onKinematic Source InversionInitial
(sobering) results
P. Martin Mai Institute of Geophysics, ETH
Zurich mai_at_sed.ethz.ch and the SPICE Local
Scale Group (Ampuero, Delouis, Festa, Holden,
Madariaga, Moczo, Vilotte, Zarahdnik )
2OUTLINE
? Goal perform accurate kinematic
finite-source inversion to estimate the rupture
process during earthquake faulting ? Approach
test various source-inversion methods, and assess
their resolution, strength and weaknesses to
finally design the optimal inversion
strategy ? Method generate synthetic
near-source motions for some source rupture model
(with increasing complexity as the project
progresses) that is known to only one person (M.
Mai), while interested researchers can then apply
their source-imaging technique blindly
3The Model Setup
- Earthquake source geometry and station
distribution chosen to be reminiscent of the 2000
Tottori earthquake (M 6.9), but the details of
the rupture process are not known.
4The Model Setup
- Synthetic seismograms are then computed,
assuming (unknown to modelers) constant rupture
velocity, constant rise time, simple
slip-velocity function, and heterogeneous slip. - A discrete wave-number integration method (0.01
lt f lt 3 .0 Hz) is used for wavefield
calculations. - In the first step, no noise is added to the
synthetics later, also the above conditions on
vr, tr will be relaxed. - The modelers are given
- seismic moment 1.43?? 1019 Nm
- dip 90o
- rake 150o
- hypocentral depth 12.5 km
- velocity-density structure
5INITIAL RESULTS
- So far, four researchers have sent in their
inversion solution to this initial problem - ? the results are remarkably diverse !!!!!
- In alphabetical order
- Betrand Delouis Non linear inversion by
simulated annealing - Gaetano Festa back-projection of the S
-amplitudes along the rays - Catherine Holden non linear inversion using the
neighbourhood algorithm - Jiri Zahradnik iterative moment-tensor
deconvolution, combined with forward modeling
6INITIAL RESULTS
- Betrand Delouis Non linear inversion by
simulated annealing - vr 3.1 km/s, tr 1.0 s
? NOTE THE GOOD DATA FITTING !!!
7INITIAL RESULTS
- Gaetano Festa back-projection of the S
-amplitudes along the rays - vr 2.8 km/s, tr 1.0 s
? NOTE THE GOOD DATA FITTING !!!
8INITIAL RESULTS
- Catherine Holden non linear inversion using the
neighbourhood algorithm - vr 2.15 km/s, tr ?? s
? NOTE THE GOOD DATA FITTING !!!
9INITIAL RESULTS
- Jiri Zahradnik iterative moment-tensor
deconvolution, combined with forward modeling - vr 2.6 km/s, tr 1.0 s
? NOTE THE GOOD DATA FITTING !!!
10THE SOLUTION
- and here comes the long awaited answer .
- vr 2.7 km/s and tr 0.8, using an isosceles
triangle as SVF - max. displacement 2.5m, about 15 km north-west
of the hypocenter - the rupture expanded primarily in the
NW-direction -
11CONCLUSIONS
? Can we conclude anything from this exercise
yet?? ? First of all work is needed to make
this solutions agree -- consistent definition
which stations to use and how to perform the
data preparation, to eliminate any bias coming
from there -- differences due to Greens
function calculations ?? -- differences due to
inversion approach (linear vs. non-linear,
parameterization, smoothing, misfit function,
etc) ?? ? Next step quantitative comparison
of inverted slip models -- cross-correlation
analysis with respect to input model -- slip
values at selected points -- overall slip
distribution -- .