of 6
Evidence for universal earthquake
rupture initiation behavior
Men-Andrin Meier
1
, Thomas Heaton
1
, and John Clinton
2
1
Seismological Laboratory, California Institute of Technology, Pasadena, California, USA,
2
Swiss Seismological Service, Swiss
Federal Institute of Technology (ETH), Zurich, Switzerland
Abstract
Earthquake onsets provide a unique opportunity to study physical rupture processes because
they are more easily observable than later rupture stages. Despite this relative simplicity, the observational
basis for rupture onsets is unclear. Numerous reports of evidence for magnitude-dependent rupture onsets
(which imply deterministic rupture behavior) stand in contradiction to a large body of physics-based rupture
modeling efforts, which are mostly based on inherently nondeterministic principles. Here we make use of the
abundance of short-distance recordings available today; a magnitude dependency of onsets should appear
most prominently in such recordings. We use a simple method to demonstrate that all ruptures in the studied
magnitude range (4
<
M
<
8) share a universal initial rupture behavior and discuss ensuing implications for
physical rupture processes and earthquake early warning.
1. Introduction
What is it that decides whether an ongoing earthquake grows into a large rupture, tens or hundreds of
kilometers long, involving several meters of peak slip and causing widespread destruction, or whether it
stopstobecomeanotherinsigni
fi
cantsmalleventonlyperceivedbyhighlysensitiveinstruments?Thisquestion
lies at the heart of a debate on whether or not earthquakes develop in a deterministic and hence potentially
predictable manner.
Numerous authors have suggested that the eventual size of an earthquake may be determined by the char-
acteristics of rupture initiation [e.g.,
Iio
, 1995;
Ellsworth and Beroza
, 1995;
Olson and Allen
, 2005;
Colombelli et al
.,
2014], e.g., because ruptures that start with a stronger initial push may be more likely to overcome mechanical
barriers andthus may tend to propagate further [e.g.,
Heaton
, 1990]. Most ofthese studies, however, have been
contested on an observational basis [e.g.,
Mori and Kanamori
, 1996;
Rydelek and Horiuchi
, 2006]. Furthermore,
the deterministic rupture development that rupture predictability implies would be surprising from the point
of view of rupture mechanics and dynamics: Rupture processes are widely understood to be in
fl
uenced by
feedback mechanisms such as slip rate-dependent friction [
Burridge and Knopoff
, 1967;
Heaton
, 1990;
Rice
,
1993], the
presumably heterogeneous
stress distribution along the rupturing fault [e.g.,
Smith and
Heaton
, 2011], and the dynamics in the ongoing rupture energy budget [e.g.,
Aagaard and Heaton
, 2008;
Elbanna and Heaton
, 2012], all of which render a deterministic rupture development improbable.
At least in part, this debate on rupture predictability is unresolved because the observational basis for the
discussion is itself unclear, with different studies reporting different results. Reasons for the observational
discrepancies may include that observational studies on rupture onsets have analyzed narrow magnitude
ranges [
Iio
, 1992;
Mori and Kanamori
, 1996;
Nakatani et al
., 2000], have been based on small and often
hand-selected data sets [e.g.,
Ellsworth and Beroza
, 1995;
Colombelli et al
., 2014], have used indirect parame-
terizations of ground motion that are dif
fi
cult to interpret [e.g.,
Olson and Allen
, 2005;
Colombelli et al
., 2014],
and have focused on records with large hypocentral distances [
Colombelli et al
., 2014] which are strongly
affected by attenuation that has to be corrected for, typically in simplistic ways.
If indeed rupture onsets are diagnostic of the future rupture development, initial differences between small
and large earthquakes should appear most prominently in short-distance recordings for which complicating
effects from attenuation as well as from nondirect and shear wave phases are minimal [
Kanamori and Mori
,
2000]. Today there are considerable amounts of such near-source recordings available for a wide magnitude
range. From an observational point of view the debate should therefore be resolvable. To this end we have
compiled a waveform data set that contains a majority of publicly available near-source records for large
shallow crustal earthquakes, along with a large number of records from smaller events. Based on this data
MEIER ET AL.
UNIVERSAL EARTHQUAKE RUPTURE INITIATION
7991
PUBLICATION
S
Geophysical Research Letters
RESEARCH LETTER
10.1002/2016GL070081
Key Points:
We compare seismic near-source
recordings in magnitude range
4
<
M
<
8
Onsets of small and large earthquakes
are statistically indistinguishable
Rupture onsets are not diagnostic of
fi
nal rupture sizes
Supporting Information:
Supporting Information S1
Correspondence to:
M.-A. Meier,
mmeier@caltech.edu
Citation:
Meier, M.-A., T. Heaton, and J. Clinton
(2016), Evidence for universal earth-
quake rupture initiation behavior,
Geophys. Res. Lett.
,
43
, 7991
7996,
doi:10.1002/2016GL070081.
Received 18 JUN 2016
Accepted 26 JUL 2016
Accepted article online 29 JUL 2016
Published online 14 AUG 2016
©2016. American Geophysical Union.
All Rights Reserved.
set we attempt to compile a model-free and objective description of ground motion onsets and search for
signi
fi
cant differences between the onsets of small and large earthquakes.
2. Materials and Methods
2.1. Composite Near-Source Waveform Data Set
The data set consists of vertical ground motion records with hypocentral distances of
25 km (
near-source
,
Figure 1). It includes all Japanese National Research Institute for Earth Science and Disaster strong motion
records with
M
JMA
4(
K-NET & KiK-net,
2471 records), the records from the Next Generation Attenuation
West 1 data set that include
P
wave onsets [
Chiou and Youngs
, 2008] (
NGA West 1,
36 records), all strong
motion and broadband records from the Southern California Seismic Network with
M
L
4(
SCSN
, 568
records), and the near-source record from the 2008
M
w
7.9 Wenchuan, China, earthquake. A more detailed
description of the data set is given as supporting information.
2.2. Peak Absolute Ground Motion Displacement
pgd
(
t
)
For each record we identify the onset of ground motion using the SBPx
P
phase picking algorithm [
Meier et al
.,
2015] on the band-pass-
fi
ltered acceleration traces. We numerically integrate all records to displacement
time series,
u
(
t
), starting the integration 100 samples before the onset of the
P
wave. We then process the
displacement time series with a causal second-order Butterworth band-pass
fi
lter with corner frequencies
of 0.075 and 30 Hz. We choose an upper frequency limit well below the Nyquist frequency of the used records
(
f
Nyq
40) in order to rule out any in
fl
uence of acausal antialias
fi
ltering that is applied by modern data log-
gers [
Scherbaum and Bouin
, 1997]. We examine how peak absolute vertical displacements evolve over time
for different magnitudes. We de
fi
ne
pgd
(
t
) = max{|
u
(
t
)|} the peak absolute amplitude of the ground motion
displacement time series during the time interval
t
[
t
pick
t
], where
t
pick
is the pick time and
t
is increased in
0.01 s intervals. For the sake of simplicity we do not account for the effects of radiation patterns, and we do
not normalize the
pgd
(
t
) amplitudes to a common distance.
Owing to the heterogeneous nature of our data set, presignal ambient noise levels before the earthquake
signals vary strongly, from ~1e

6 m/s
2
to ~1e

2 m/s
2
. The noise level can affect the
pgd
(
t
) statistic, since
high noise levels cause a detection delay of the signal onset (the time it takes for the signal to reach ampli-
tudes above the noise level). In order to avoid artifacts from such noise-dependent detection delays, we
ensure that all records have comparable noise amplitudes by scaling up the noise levels of the low-noise sig-
nals to a common presignal amplitude level of 1e

3 m/s
2
(95th percentile of the amplitude distribution,
detailed description in the supporting information). Supporting information Figure S1 shows the same
fi
gure
as Figure 2 without the noise scaling; it shows that the results of this study do not depend on the noise
scaling procedure.
Figure 1.
Data set overview. (a) Magnitudes and hypocentral distances of the records used in this study from the four data
sources. (b) Empirical cumulative hypocentral distance distributions for the records in six magnitude bins and number of
records,
n
, in each bin.
Geophysical Research Letters
10.1002/2016GL070081
MEIER ET AL.
UNIVERSAL EARTHQUAKE RUPTURE INITIATION
7992