 |
| Titel |
Kurtosis based automated P-S phase picking procedure for hypocenter determination: Vanuatu region case study |
| VerfasserIn |
C. Baillard, W. Crawford, V. Ballu, C. Hibert |
| Konferenz |
EGU General Assembly 2012
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 14 (2012) |
| Datensatznummer |
250065621
|
|
|
|
|
|
| Zusammenfassung |
| Abstract
Automatic P and S phase picking is indispensable for seismologists dealing
with large amounts of data. Robust algorithms, based on short term and long
term average ratio comparison (Allen, 1982), are now commonly used for
event detection, but further improvements can be made in phase picking and
identification. We present a picking scheme using consecutively Kurtosis-derived
Characteristic Functions (CF) and Eigenvalue decompositions on 3-component
seismic data. When computed over a sliding window of the signal, a sudden
increase in the CF reveals a transition from a gaussian to a non-gaussian
distribution, characterizing the phase onset (Saragiotis, 2002). One strong point is
that it requires much fewer adjustable parameters than competing methods.
We modified the Kurtosis CF to improve pick precision, computing the CF
over several frequency bandwidths, window sizes and smoothing parameters.
Once phases were picked, we determined onset type (P or S) using polarization
parameters such as rectilinearity, azimuth and dip resulting from Eigenvalue
decompositions of the covariance matrix (Cichowicz, 1993). Finally, we removed
bad picks using a clustering procedure and a signal-to-noise ratio (SNR). The pick
quality index was also assigned based on this SNR value.
We applied this procedure to data from a network of 30 wideband seismometers
(including 10 oceanic bottom seismometers) in Vanuatu. Events were extracted
from 10 months of continuous data using STA/LTA algorithm, then picked using
our method. We estimated the reliability of our picking procedure by comparing
4301 manual and automatic picks. We found a mean difference of 0.07 ± 0.24
s overall; for high quality picks (quality index 0) the difference is 0.03 ± 0.13s.
After inversion with HYPOCENTER, more than 30% of earthquakes have a RMS
-¤ 0.7 s and localization uncertainty -¤ 20 km before velocity model improvement.
The method demonstrates good performance to independently pick S and P waves
with only a few parameters to adjust. |
|
|
|
|
|
|