 |
| Titel |
Objective-function hybridization in adjoint seismic tomography |
| VerfasserIn |
Yanhua O. Yuan, Ebru Bozdag, Frederik J. Simons, Fuchun Gao |
| Konferenz |
EGU General Assembly 2017
|
| Medientyp |
Artikel
|
| Sprache |
en
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 19 (2017) |
| Datensatznummer |
250146946
|
| Publikation (Nr.) |
 EGU/EGU2017-11020.pdf |
|
|
|
|
|
| Zusammenfassung |
| Seismic tomography is at the threshold of a new era of massive data sets. Improving the
resolution and accuracy of the estimated Earth structure by assimilating as much information
as possible from every seismogram, remains a challenge. We propose the use of the
“exponentiated phase’’, a type of measurement that robustly captures the information
contained in the variation of phase with time in the seismogram. We explore its
performance in both conventional and double-difference (Yuan, Simons & Tromp,
Geophys. J. Intern, 2016) adjoint seismic tomography. We introduce a hybrid approach
to combine different objective functions, taking advantage of both conventional
and our new measurements. We initially focus on phase measurements in global
tomography. Cross-correlation measurements are generally tailored by window
selection algorithms, such as FLEXWIN, to balance amplitude differences between
seismic phases. However, within selection windows, such measurements still favor the
larger-amplitude phases. It is also difficult to select all usable portions of the seismogram in
an optimal way, such that much information may be lost, particularly the scattered
waves. Time-continuous phase measurements, which associate a time shift with each
point in time, have the potential to extract information from every wiggle in the
seismogram without cutting it into small pieces. One such type of measurement is the
instantaneous phase (Bozdağ, Trampert & Tromp, Geophys. J. Intern, 2011), which
thus far has not been implemented in realistic seismic-tomography experiments,
given how difficult the computation of phase can sometimes be. The exponentiated
phase, on the other hand, is computed on the basis of the normalized analytic signal,
does not need an explicit measure of phase, and is thus much easier to implement,
and more practical for real-world applications. Both types of measurements carry
comparable structural information when direct measurements of the phase are not
wrapped. To deal with cycle skips, we use the exponentiated phase to take into
account relatively small-magnitude scattered waves at long periods, while using
cross-correlation measurements on windows determined by FLEXWIN to select distinct
body-wave arrivals without complicating measurements due to non-linearities at
short periods. We present synthetic experiments to show how exponentiated-phase,
cross-correlation measurements, and their hybridization affect tomographic results. We
demonstrate the use of hybrid measurements on teleseismic seismograms, in which surface
waves are prominent, for continental and global seismic imaging. It is clear that the
exponentiated-phase measurements behave well and provide a better representation of the
smaller phases in the adjoint sources required for the computation of the misfit gradient. The
combination of two different types of phase measurements in a hybrid approach moves us
towards using all of the available information in a data set, addressing data quality and
measurement challenges simultaneously, while negligibly affecting computation time. |
|
|
|
|
|
|