![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
NASTF: A Bayesian catalogue of source parameters from teleseismic body waves |
VerfasserIn |
Simon Stähler, Karin Sigloch, Ran Zhang, Heiner Igel |
Konferenz |
EGU General Assembly 2013
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 15 (2013) |
Datensatznummer |
250078327
|
|
|
|
Zusammenfassung |
Large earthquakes and those in densely instrumented areas are now being studied in great
detail and in extended-source frameworks like finite-fault or back-projection. However,
smaller earthquakes (below MW - 7.5) and especially remote ones with sparse data
coverage are still approximated best by a point source. For earthquakes larger MW - 5.5 it is
generally possible to invert for the temporal evolution and describe it in the form of a moment
rate or Source Time Function (STF).
A reliable STF is crucial for broadband waveform tomography. Its uncertainty is hard to
quantify, especially since it is correlated with the estimated source depth, e.g. when
surface-reflected phases are mapped into the STF. While the inversion for the STF and the
moment tensor is linear, the depth inversion is inherently non-linear. Experience shows that
data from shallow earthquakes can often be fitted well by several distinct depths.
Therefore it is hard to linearize the inversion for the depth. We therefore propose a fully
Bayesian inversion scheme for the Source Time Function, depth and moment tensor as
follows:
The STF is parametrized in empirical source wavelets. Those wavelets are
derived from a catalogue of STFs for intermediate size earthquakes, which were
determined beforehand. The STF can be described by 8-10 weighting factors of
these wavelets.
As a misfit criterion we use the waveform coherence in the P- and SH-window.
While it is not possible to derive an Likelihood distribution for this measure
analytically, we use a large number of synthetic waveforms calculated with
catalogue source solutions to find such a Likelihood function empirically. This
is then used to determine the Likelihood of any given solution in the ensemble.
The parameter space then has 14-16 dimensions, and is sampled with the
Neighbourhood Algorithm. Therefore the inversion is derivative-free and not
disturbed by the nonlinearity of the problem.
Synthetic waveforms are calculated using a reflectivity code on the IASP91
mantle with local crust models from CRUST2.0. One forward solutions takes
around 1 CPU-second.
Exhaustive sampling of the model space requires around 105 forward solutions to be
calculated, which means less than one hour on a modern multicore computer. The
resulting ensemble offers more than just a best solution, but rather probability density
functions for all parameters, from which the extend of likely solution regions can be
estimated.
While this method may be too intensive for rapid-source-determination, all kinds of
broadband body wave studies could benefit from a reliable STF catalogue with actual
uncertainties. |
|
|
|
|
|