1
2
3
4
5
Non-Volcanic Tremor Locations and Mechanisms
6
in Guerrero, Mexico, from Energy-based and
7
Particle-Motion Polarization Analysis
8
9
Víctor M. Cruz-Atienza, Allen Husker,
10
Denis Legrand and Emmanuel Caballero
11
12
Universidad Nacional Autónoma de México, Instituto de Geofísica
13
(Corresponding author email: cruz@geofisica.unam.mx)
14
15
Submitted to the Journal of Geophysical Research – Solid Earth
16
17
June 2014
18
19
20
1
21
Abstract (250 words)
22
23
We introduce the Tremor Energy and Polarization (TREP) method, which is a novel
24
technique that jointly determines the source location and focal mechanism of sustained non-
25
volcanic tremor (NVT) signals. The method minimizes a compound cost function by means
26
of a grid search over a three-dimensional hypocentral lattice. Inverted metrics are derived
27
from three NVT observables: (1) the energy spatial distribution, (2) the energy spatial
28
derivatives, and (3) the azimuthal direction of the particle motion polarization ellipsoid.
29
The cost function combines these metrics obtained from observed and synthetic
30
seismograms for double-couple point dislocations considering frequency-dependent quality
31
factors (Q) in a layered medium. Performance of the TREP method is thoroughly assessed
32
through synthetic inversion tests where the ‘observed’ data were generated using an NVT-
33
like finite difference (FD) model we introduced. The FD tremor source is composed by
34
hundreds of quasi-dynamic penny-shape cracks governed by a time-weakening friction law.
35
In agreement with previous works, epicentral locations of 26 NVTs in Guerrero are
36
separated in two main groups, one between 200 and 230 km from the trench, and another at
37
about 170 km. However, unlike earlier investigations, most NVT hypocenters concentrate
38
at 43 km depth near the plate interface and have subparallel rake angles to the Cocos plate
39
convergence direction. These results are consistent with independent locations of low-
40
frequency earthquakes in the region and further support their common origin related to slip
41
transients in the plate interface. Our results also suggest the occurrence of NVT sources
42
within the slab, ~5 km below the interface.
43
1. Introduction
44
45
The discovery of non-volcanic tremors (NVT) (Obara, 2002) and silent slip earthquakes
46
(SSE) (Dragert et al., 2001) has provided key elements to understand fault mechanics in
47
deep subduction interfaces. In some regions, such as the Cascadia and Nankai subductions
48
zones, these phenomena are usually correlated in space and time, with recurrence intervals
49
from several months to years (e.g., Rogers and Dragert, 2003; Obara et al., 2004). Low
50
(i.e., 1-5 Hz) and very-low (i.e., 0.02-0.05 Hz) frequency earthquakes (LFE and VLF,
51
respectively) with SSE-consistent focal mechanisms (i.e., shallow-thrust faulting) have also
2
52
been observed along the subduction interface of southwest Japan (Shelly et al., 2007; Ito et
53
al., 2007), with the LFEs proposed as the elementary unit of tremor signals (i.e., NVTs are
54
the summation of LFEs) (Shelly et al., 2007; Ide et al., 2007). Since LFE locations in
55
different subduction regions are constrained near the interface of the plates (e.g., Ide et al.,
56
2007; Brown et al., 2009; Bostock et al., 2012; Frank et al., 2013), this hypothesis assumes
57
that the sources of NVT are also located close to the interface. However, tremor locations
58
with an independent method in Cascadia (Kao, 2005; Kao et al., 2009) suggest that tremors
59
are not originated at the plate interface where SSEs are supposed to occur. Recent
60
observations of non-linear crustal deformations during SSEs in Guerrero, Mexico, revealed
61
transient reductions of the rocks resistance (i.e., the shear modulus) in the middle and lower
62
crust (Rivet al., 2011; 2013). Therefore, this phenomenon represents a possible mechanism
63
promoting shear failures in wide spread regions around the silent slipping surface. LFEs are
64
usually detected after stacking tiny signals to obtain higher correlation coefficients between
65
waveforms (e.g., Shelly et al., 2006). These are low-frequency signals embedded in a
66
broader-band long-duration records where a significant amount of energy corresponds to
67
“incoherent” arrivals, namely tremor bursts. In other terms, the band-limited seismic
68
radiation whose source has been identified as shear failures near to the plate interface is
69
only a fraction of the information recorded during tremor episodes. A comprehensive
70
review of these phenomena may be found in Beroza and Ide (2011).
71
SSEs in Mexico were first reported (Lowry et al., 2001) the same year the phenomenon was
72
discovered worldwide (Dragert et al., 2001). Further investigations of silent earthquakes in
73
Guerrero allowed the understanding of the spatial extent and propagation patterns of the
74
transient slips at the plate interface (Iglesias et al., 2004; Franco et al., 2005; Vergnolle et
75
al., 2010; Radiguet et al., 2012; Cavalié et al., 2013). Unlike other subduction zones, NVT
76
episodes in Guerrero are not always concomitant with large, long-term SSEs (Payero et al.,
77
2008; Husker et al., 2012), suggesting that tremors and silent earthquakes may have
78
different origins (Kostoglodov et al., 2010). However, NVT bursts observed during inter-
79
SSE periods may be possibly related to small, short-term slip transients with subcentimeter
80
surface displacements (Vergnolle et al., 2010). Detailed spatio-temporal tracking of the
81
NVT and LFE activity with accurate estimates of the source depth and mechanism may
82
provide key elements to better elucidate the relationship between both phenomena.
3
83
The largest difficulty to locate tremors is their emergent and sustained character, where
84
almost no coherent phases can be visually identified. Most of the current methods to locate
85
NVTs are based on cross-correlation measurements of waveform templates or envelopes
86
(e.g., Obara, 2002; Shelly et al., 2006; Ide et al., 2007; Payero et al., 2008; Wech and
87
Creager, 2008; Brown et al., 2009; Obara, 2010; Frank et al., 2013). Time lags between the
88
maximum cross-correlation coefficients are then used as travel times to locate the source.
89
When dense seismic arrays are available, beamforming analysis (Ghosh et al., 2009) and
90
energy attenuation patterns (Kostoglodov et al., 2010, Husker et al., 2012) may also be
91
used, although these approaches lack of depth resolution so the source depth is usually
92
fixed to analyze epicenter locations.
93
Evidence of NVT in wide spread three-dimensional regions is based on source locations
94
using two different methods: the source-scanning algorithm (SSA) (Kao and Shan, 2004;
95
Kao, 2005; Kao et al., 2009) and the envelope cross-correlation method (Obara, 2002).
96
Envelope cross-correlation locations suffer from a lack of precise arrival-time peaking due
97
to long-duration envelopes used for this purpose (i.e., minutes long). Although
98
improvements have been made to the method (e.g., Obara et al., 2010), this issue may
99
translate as inaccurate depth estimates (e.g., Payero et al., 2008). The other method, the
100
SSA, is based on move-out summation of seismogram windows without considering the
101
source radiation pattern, and so tremor depths may suffer from significant errors, as we
102
shall discuss later. Thus, there are no convincing results about the depths where the
103
sustained tremor bursts are generated. Reliable locations of these seismic events based on a
104
robust, and distinct method may help to further elucidate whether the hypothesis stating that
105
LFEs represent the elementary unit of NVTs is plausible in the entire frequency band where
106
the NVTs are observed.
107
NVT (and LFE) amplitudes and waveforms depend on both, the source location and
108
mechanism. Solving these source parameters by simultaneously explaining independent
109
properties of the radiated wavefield may lead to robust location estimates. Previous works
110
have demonstrated that both, the energy spatial distribution of the NVT records, and the
111
associated particle motion polarization are two source-dependent properties that are clearly
112
observed in Guerrero and Cascadia (Kostoglodov et al., 2010; Husker et al., 2012; Wech
4
113
and Creager, 2007). In this study, we introduce a new technique that combines these
114
properties to assess the source of sustained tremor signals.
115
116
2. Location Technique
117
118
The location technique, called the “Tremor Energy and Polarization” (TREP) method,
119
consists of a grid search over a three-dimensional hypocenter lattice to investigate the
120
source location (i.e., the three hypocentral coordinates) and mechanism (i.e., the rake angle)
121
that minimize a cost function involving different observables. A similar approach to
122
simultaneously determine the location and focal mechanism of volcanic sources was
123
proposed by Legrand et al. (2000) using the wavefield amplitudes. The observables used
124
here, which are described in the next section, are translated into energy and particle-motion
125
polarization metrics that are simultaneously inverted to determine the source parameters
126
mentioned above.
127
128
A three-dimensional regular grid of potential hypocenter locations was generated below the
129
region where the seismic stations are located. In this study, we chose the middle-America
130
trench axis as the reference direction. Thus the grid was rotated 15 degrees clockwise from
131
the North. The optimal 140 x 60 x 60 km3 grid has 5 km increments in the three Cartesian
132
directions and was located about 130 km away from the trench (dashed rectangle, Figure 1).
133
For each grid node, a set of three-component double-couple seismograms for horizontal
134
point dislocations in space and time (i.e., the convolution of the Green-tensor spatial
135
derivatives and the moment density tensor) with different rake angles were computed
136
between the nodes and each site of the 42 available stations. Assuming a fault strike of 285o
137
(i.e., the trench axis azimuth) and given the slip directions already observed in the region
138
for LFE and VLF (Frank et al., 2013; Ide, 2014), the rake angles ranged between 30o to
139
120o with a 10o increment. To compute the synthetic seismograms database (SSD), we used
140
the wavenumber method (Bouchon and Aki, 1977) considering a layered elastic medium
141
(Campillo et al., 1996) and included the crustal intrinsic attenuation by means of frequency-
142
dependent Q factors empirically determined by García et al. (2004), both suitable for the
143
study region. More than 5.7 million point-source synthetic seismograms were generated
5
144
and saved in a database (i.e., SSD) prior to the inversions for the whole grid and set of
145
stations.
146
147
2.1. Energy and Particle Motion Polarization Estimates
148
149
The energy spatial distribution of a wavefield mainly depends on the source location and
150
mechanism; however, site, scattering and anelastic effects may also affect seismograms
151
depending on the frequency band of interest. For these reasons, to compute the NVT energy
152
distribution along the stations array, we followed the procedure introduced by Kostoglodov
153
et al. (2010) and used by Husker et al. (2012) to locate NVT epicenters in the same region.
154
The spectral signature of tremors in Guerrero emerges between 0.5 and 10 Hz, although the
155
signal to noise ratio is highest in the range of 1 to 2 Hz along most of the MASE stations
156
(Husker et al., 2010). After removing both, the trend and mean of the signals, we proceeded
157
as follows: (1) filtered the three components of the observed seismograms in this
158
bandwidth, (2) applied the corresponding site-effect correction factors determined by
159
Husker et al. (2010), (3) computed the square velocities per component, (4) applied a ±10
160
minute-window median filter to smooth the resulting time series, and (5) computed the
161
energy on each station, , as the time-cumulative values. For a given tremor episode, we
162
thus generated three component energy values per station to finally obtain the three-
163
dimensional energy distribution along the stations array. Figure 3 shows some examples of
164
energy distributions for two tremors computed by means of this technique.
165
166
Similar to gravity anomalies (Telford et al., 1990), the shape of the energy functions (i.e.
167
the width and slopes) is primarily controlled by the source depth. The wider and smoother
168
the function, the deeper is the source. Therefore, to improve the hypocentral depth
169
resolution of the location technique, we incorporated the energy spatial derivatives
170
approximated as the ratio of the energy difference between each pair of stations and the
171
distance separating them (Figure 3).
172
173
Depending on the seismic array configuration, the energy distribution in the three
174
components may not uniquely solve for the source location and mechanism. In our case,
6
175
given the stations alignment (Figure 1, array geometry), this problem becomes critical. For
176
this reason, we also analyzed the particle-motion polarization ellipsoids, which has
177
essential and complementary information about these source properties. For this purpose,
178
we used the method proposed by Flinn (1965) and Jurkevics (1988) that has been proved to
179
be a powerful tool for analyzing NVTs in Cascadia (Wech and Creager, 2007). By solving
180
the eigenproblem of the data covariance matrix, the method allows to determine the
181
rectilinearity of the ground motion and the orientation of the corresponding polarization
182
ellipsoid. The direction of the eigenvector associated with the largest eigenvalue (from now
183
on referred as the main eigenvector) corresponds to the direction of the largest ellipsoid
184
semi-axis, while the degree of rectilinearity, which ranges between 0 (spherical motion) and
185
1 (linear motion), is given by the combination 1
186
and represent the three eigenvalues (Jurkevics, 1988). Figure 2a shows the 24-hour energy
187
spectrogram of March 6, 2005 between 0.5 and 10 Hz (upper panel) at station SATA along
188
with the corresponding rectilinearity spectrogram (Figure 2b) and the 1 – 2 Hz north-south
189
filtered seismogram (Figure 2c). One main observation emerges: although the degree of
190
motion rectilinearity is moderate, there is clear correlation in time and frequency between
191
both spectrograms during tremor episodes (e.g., between 4 – 6 h, 7 – 7.6 h, 12.8 – 13.9 h,
192
and 20 – 21 h). This implies that the ground motion polarization is dominated by coherent
193
tremor signals (i.e. above the ambient seismic noise). To compute the rectilinearity, we
194
used a 5 s moving window with time increments of 2 s and then applied a 10 s median filter
195
over the resulting time series for each 0.5 Hz frequency band. Despite the almost
196
continuous spectral signature of tremors, their polarization patterns are segmented in two
197
frequency bands, one from 1 to 3 Hz and another from 5 to 9 Hz. This observation opens
198
interesting questions about the nature of tremor sources that go beyond the scope of the
199
present work, since we focused on polarization estimates in a frequency range of 1 to 2 Hz
200
(i.e. the same as for the energy estimates) to locate the events.
⁄2
, where
,
201
202
We used the azimuthal direction of the largest polarization-ellipsoid axis as the particle-
203
motion polarization metric in the location technique. This direction is given by the
204
horizontal projection of the main eigenvector. Once projected, the polarization horizontal
205
vector (PHV) is normalized to become proportional to the total energy on each site, as
7
206
required by the cost function described in the next section. Figure 3 shows the PHV along
207
the array for two different NVTs and how the polarization azimuth changes from one event
208
to the other.
209
210
2.2. Observable Metrics and Cost Function
211
212
As mentioned earlier, the location technique looks for the three hypocentral coordinates and
213
the rake angle that minimize , a compound cost function. This function depends on three
214
different metrics, all of them computed in the same manner for the observed data and the
215
SSD (from now on referred as synthetic data). The metrics are (1) the spatial energy
216
distribution in the three components; (2) their corresponding spatial derivatives; and (3) the
217
azimuth of the particle-motion polarization direction. To compute the misfit between
218
observed and synthetic data for metrics 1 and 2, we first normalize the energy and
219
derivative distributions along the stations array to their corresponding maximum absolute
220
values in the three components. The misfit of the energy metric is then given by
221
∑
, i.e. the 2
, where
and
are the observed and synthetic
222
energy series in the three components, respectively, and
223
multiplied
224
∑
by
three. Similarly, the
, where
and
misfit
of
the
is the number of stations
derivative
metric
is
are the observed and synthetic energy derivative series,
∑
225
respectively. The misfit of the polarization azimuth is defined as
226
i.e. the 1
227
unit vectors) of the observed and synthetic main eigenvectors, respectively, and
228
normalized weighting factors proportional to the total energy along the array (i.e. 0
229
1 such that
230
weight the polarization misfit depending on the amount of energy at every station. Finally,
231
to establish a well-balanced cost function,
232
, ,
̂
̂
,
, where ̂ and ̂ are the horizontal projections (i.e. two-dimensional
are
1 at the station where the energy is maximum). These factors properly
, combining all metric misfits, we define
as the average of the three misfit functions defined above after carrying out
233
the following normalization. Let function
234
given by
235
optimize the problem resolution by maximizing the gradient of
236
absolute minimum.
min
⁄ max
be ,
min
or
. Then, normalized function
is
. This definition has been found to
in the surroundings of its
8
237
238
2.3. Location Technique Verification
239
240
In this section we introduce a novel tremor-like source model that is approximated by
241
means of a finite difference approach. This model is then used to generate theoretical
242
data for several synthetic inversion tests in order to assess and verify the TREP
243
location method introduced in previous sections.
244
245
2.3.1. Non-Volcanic-Tremor Source Model
246
247
We developed a 3D quasi-dynamic source model to simulate tremor-like seismograms by
248
means of a fine difference approach (Cruz-Atienza, 2006; Cruz-Atienza et al., 2007).
249
Although the model does not aim to describe the physics of NVT sources, it does provide a
250
means to calculate sustained tremor-like signals (Figure 5a) along the MASE array within a
251
heterogeneous earth model (i.e. the structure by Campillo et al., 1996) to be used as the
252
‘observed data’ in our location technique.
253
254
Since the earth model we have considered is a simple 1D layered medium, in order to
255
introduce random variability in the simulated wavefield as observed in the real earth due to
256
scattering effects, the synthetic source is composed by 250 quasi-dynamic penny-shape
257
horizontal cracks following a spherical Gaussian distribution (with a radius of 5 km and a
258
of 1 km) (Figure 4b) with a constant (radial) rupture speed (2.9 km/s, i.e., 0.8 times the S-
259
wave speed at 40 km depth) and a constant stress drop (0.5 bar). The stress-breakdown
260
process during the cracks rupture propagation is governed by a time-weakening friction law
261
(Andrews, 2004) with a characteristic time of 0.6 s. The shear pre-stress (i.e., initial fault-
262
tangential traction) orientation varies randomly between cracks (i.e., 15°) (white arrows
263
in Figure 4b) around a main prescribed direction (black arrow in Figure 4b) so that the slip
264
direction (i.e., the rake angle) also varies in space. The radiuses of the cracks are also
265
randomly generated with a Gaussian distribution around a value of 1 km and a
266
(black circles in Figure 4). The rupture initiation time per crack is randomly set between the
267
starting and ending simulation times (i.e., between 0 and 80 s). The source model is
of 0.5 km
9
268
numerically approximated by means of a 3D partly-staggered finite difference scheme
269
designed to simulate the dynamic rupture of faults (Cruz-Atienza, 2006; Cruz-Atienza et
270
al., 2007). The numerical model introduced by these authors has been adapted to the source
271
features mentioned above and used with a grid size of 150 m and a time step of 0.015 s so
272
that, considering the lowest S-wave velocity in our crustal model (i.e., 3.1 km/s, Campillo
273
et al., 1996), the method guarantees a good numerical accuracy up to 2 Hz (i.e., 10 grid
274
points per minimum wavelength) (Bohlen and Saenger, 2006).
275
276
2.3.2. Synthetic Inversion Results
277
278
Using this source model, we have generated tremor-like synthetic seismograms for
279
frequencies below 2 Hz (e.g., see Figure 5a for the station XALI) along the MASE array
280
(Figures 1 and 4a), which are the frequency bandwidth and the stations array used to
281
analyze real data in this present work. Results for two synthetic inversion tests are
282
illustrated below, where the target-sources have the same epicenters (green stars in Figure
283
6a and 6d) but different depths (i.e., 30 and 40 km) and pre-stress directions (i.e., main
284
azimuths of 50 o and 130o, green arrows in Figure 6a and 6d, respectively). The associated
285
seismograms were inverted using a low-resolution hypocentral grid with 20 km-horizontal
286
and 5 km-vertical increments (crosses in Figure 4a).
287
288
Figure 6 shows the results for the best-fit solution models yield by the two synthetic
289
inversions. In the second and third row panels (i.e., panels b, e, c and f) we find fits
290
between the “observed tremors” (red) and the synthetic data (blue) for both, the energy
291
distributions and the associated spatial derivatives. Although significant misfits are found
292
in the vertical components (less energetic), the overall similarities in the horizontal
293
components are remarkably good. Relative amplitudes between NS and EW signals are
294
self-explained as well as the derivative profiles, which control the shape of the energy
295
functions and thus the source depth. It is interesting to emphasize that, although both
296
epicenters were collocated, the energy maxima is spatially shifted from one event to
297
another. While the maximum of the NS component in the first event (i.e., with rake of 50 o,
298
Figure 6b) lies around 215 km from the trench, the maximum in the second event (i.e., with
10
299
rake of 130o, Figure 6e) was about 20 km further inland. This is even clearer in the EW
300
components, where the maxima of both NVTs are shifted at about 50 km. This means that
301
the energy spatial distribution is controlled by both, the hypocenter location and the source
302
mechanism. On other hand, notice how the particle-motion polarization direction is very
303
sensitive to the hypocentral depth and slip direction (compare horizontal vectors for both
304
events, Figure 6a and 6d). Actually, many tests (not shown) have demonstrated that
305
including the horizontal polarization direction as observable in the inversion technique is
306
critical for uniquely solve both, the source mechanism and location. Arrows in the vertical
307
sections are only indicative of the particle-motion ellipsoid inclination and were not used in
308
the inversion. Cross-sections of function
309
locations (color maps) reveal that the minima of the functions coincide with the target
310
source locations (i.e., those used to generate the observed data; compare yellow and green
311
stars) and that the NVT mechanisms have been retrieved (compare yellow and green
312
arrows). Other tests for different mechanisms and source locations along the array yielded
313
similar results proving that, although resolution is significantly better in the horizontal
314
directions (compare horizontal and vertical gradients of function ), the hypocentral depth
315
and source mechanism are retrieved in all synthetic cases.
passing through the best-fit hypocenter
316
317
It is important to point out some implications of our synthetic inversion tests. All results
318
were obtained assuming the same 1D layered medium to compute the SSD and the
319
‘observed data’. However, when comparing the misfit metrics for the best-fit point-source
320
synthetic seismograms (Figures 5b, 5c and 5d) with those for the synthetic tremor (Figure
321
5a) it is clear that our tremor-like source model has intrinsic properties that make the
322
radiated wavefields from both kind of sources remarkably different. This is due to several
323
factors such as: (1) the volumetric support of the tremor source, (2) the variable pre-stress
324
conditions in the penny-shape cracks, (3) the diffracted waves at the crack edges (i.e.,
325
stopping phases) and; (4) the randomly generated rupture times of the cracks. In spite of
326
these fundamental differences in the physics of both kinds of sources, our synthetic
327
inversion tests have shown that the wavefield from simple point-dislocation sources is able
328
to explain satisfactorily the tremor-like signal properties we have selected by correctly
329
retrieving both, the target-source locations and mechanisms.
11
330
331
3. Location of Non-volcanic Tremors in Guerrero
332
333
We have applied the TREP technique to 26 tremors episodes in the state of Guerrero
334
recorded between March 2005 and March 2007 along the Meso-America Seismic
335
Experiment array (MASE, 2007) (triangles, Figure 1). These tremors were taken from the
336
catalog developed by Husker et al. (2010). Figure 3 shows individual locations for two
337
events and the associated misfits for the three inverted observables. The spectrograms and
338
the signal from one of the NVT episodes (occurred on March 6, 2005, 12:40 – 14:00) are
339
depicted with red dashed lines in Figure 2. Observed particle-motion horizontal-
340
polarization directions (red arrows, upper panels in Figures 3a and 3d) significantly change
341
from one event to the other (e.g., compare station XALI, where the polarization azimuths
342
differ about 35o i.e., from ~85o to ~110 o), indicating possible changes in the source location
343
and mechanism. Vertical polarizations were not inverted due to low signal-to-noise ratios
344
and are only indicative (Figures 3a and 3d, lower panels). The best solution models for both
345
events, satisfactorily explain the polarization azimuths (compare blue and red arrows),
346
which have different locations but similar (although not identical) source mechanisms (i.e.,
347
rake angles). Thus, to explain such significant variations of the motion polarization, the
348
epicenters must lie in different areas (i.e., different sides) of the array.
349
350
We recall that magnitudes of polarization vectors are proportional to the total energy, which
351
is decomposed along the array in its three cardinal directions for inversion purposes
352
(Figures 3b and 3e). Vertical energy distributions are not shown because their amplitudes
353
are about two times smaller than those of the horizontal components. Although fits between
354
observed and synthetic metrics are not perfect, the energy spatial distributions are well
355
explained by the best-fit solution sources (compare red and blue curves and symbols).
356
Differences are primarily due to the ambient noise, which is only present in real
357
seismograms. This can be clearly seen at stations located far from the epicenters, where the
358
energy of the NVTs are negligible, as predicted by the theoretical blue curves (e.g. see
359
stations QUEM and MAZA for the March 6, 2005 event; and stations QUEM to ZURI for
360
the November 2 event). Despite the absence of noise in our model predictions, the overall
12
361
shapes of the energy profiles are well explained, as confirmed by the quality of the energy-
362
derivative fits (Figures 3c and 3f).
363
364
Unlike the work by Husker et al. (2012), lower panels of Figures 3a and 3d show that the
365
tremor epicenters are not collocated with the maxima of the energy functions. For these two
366
events, there is a spatial shift of about 20 km between them (either to the north or south
367
direction). This is in accordance with our synthetic inversion tests (Appendix A) and was
368
expected because the approach used by Husker et al. (2012) did not consider the source
369
radiation patterns. As a consequence of the one-dimensional geometry of the station array,
370
horizontal cross-sections of the cost function
371
panels) show that the resolution of the epicentral locations is better along the axis array
372
(i.e., along the coast-perpendicular direction; compare gradients of
373
directions). This was also expected from previous studies that used the same set of data to
374
locate similar events (Payero et al., 2008; Husker et al., 2012; and Frank et al., 2013). Some
375
tests we performed (not shown) revealed that uniqueness of the inverse problem is strongly
376
enhanced by the combination of the three observables. For instance, if the polarization
377
vector is not inverted, the energy-based observables may not solve the epicenter location
378
along the array-perpendicular direction (i.e., the method yields multiple anti-symmetric
379
solutions with respect to the array). Although in general our compound cost function
380
exhibits prominent and unique minima in the horizontal plane, this is not always the case in
381
depth. The lower panel of Figure 3d shows an example where
382
minima, the absolute one at 50 km depth and a second one 10 km shallower (i.e., at the
383
plate interface). Since we do not have arguments based on our location technique to
384
determine which of the minima is more likely to be correct, interpretations must be done
385
carefully.
(background color maps in the upper
in the two Cartesian
has two comparable
386
387
Source model solutions for the whole set of NVT events are shown in Figures 1 and 4.
388
Circles with 5 km radius (i.e., the space increment of the hypocentral lattice) represent
389
tremor locations and their colors correspond to the logarithm of the NVTs per cubic
390
kilometer. The most recent locations of NVTs in the region by Husker et al. (2012) and
391
Low Frequency Earthquakes (LFEs) by Frank et al. (2013) are also plotted in Figure 1 with
13
392
green and red dots, respectively. Three main observations come out: (1) in accordance with
393
the work by Payero et al. (2008) and Husker et al. (2012), our tremor locations are
394
separated in two main groups, one at the “sweet spot” (Husker et al., 2012), between 200
395
and 230 km from the trench, and another one closer to the trench at about 170 km; (2) fault
396
mechanisms of both, NVTs and LFEs are consistent between each other and subparallel to
397
the Cocos plate convergence direction (compare red arrows with both gray and blue
398
arrows); and (3) most of the LFEs reported by Frank et al. (2013) seem to be located farther
399
from the trench than the NVTs.
400
401
Figure 7 shows our NVTs locations (colored circles) compared with those for LFEs (blue
402
circles) reported by Frank et al. (2013) projected into a vertical trench-perpendicular cross-
403
section. The interface between the Cocos and the North American plates is sketched with a
404
black line following the geometry proposed by Pérez-Campos et al. (2008). To make the
405
comparison of foci locations clearer, in the left we show normalized histograms for both
406
kinds of events as a function of depth (red and blue curves). Although the NVTs histogram
407
is bimodal with a secondary peak at 48 km depth (i.e., inside the slab) both, NVTs and
408
LFEs have their maxima of occurrence around 43 km, indicating that most of the sources of
409
these two kinds of signals originate at the same depths and probably on the plate interface.
410
Because of the orientation of the cross-section and since the alignment of NVTs at the
411
sweet spot is not parallel to the trench (see Figure 1), the horizontal position of NVTs
412
relative to LFEs may not be well appreciated in the figure. However, at least for the events
413
reported here, it seems that most LFEs occur in the northern edge of the region where the
414
NVT activity happens.
415
416
4. Discussion and Conclusions
417
418
In this study we have introduced the TREP method, which is a novel technique to locate
419
NVT signals. Using energy-based and particle motion polarization metrics, the method
420
determines the location and mechanism of the tremor source that minimize a compound
421
cost function considering frequency-dependent quality factors (Q) in a layered medium.
422
Although the horizontal resolution of the locations depends on the geometry of the stations
14
423
array, combining different physical properties of the seismic wavefield allows TREP
424
algorithm to satisfactorily solve hypocentral depths. Based on previous research, we have
425
assumed that tremors are produced by horizontal shear failures, which is a hypothesis that
426
may introduce location errors. However, this constrain may be easily relaxed in regions
427
where no information of the source mechanism is available. Since the TREP technique
428
doesn’t require any temporal information of the observed wavefield (e.g., waveform
429
templates), it can be used to locate previously detected LFEs and VLFs provided that
430
relative amplitudes between stations are preserved. This would provide the possibility of
431
comparing focal locations and mechanisms of different signals (i.e., NTV, LFE and VLF)
432
by means of the same technique and thus, have more confidence in their relative positions
433
and origins.
434
435
Tremor amplitudes recorded in a stations array depend upon several factors, such as the
436
hypocentral distance and the attenuation, site and scattering effects. However, if some
437
stations lie in a nodal source direction, amplitudes may be negligible despite their proximity
438
to the epicenter. As shown in this work, this is why considering the source mechanism is
439
critical to properly locate tremors. One possible explanation of the NVT widespread
440
locations reported by Kao et al. (2005, 2009) is that the SSA detects and locates events
441
from stacked waveforms considering only theoretical travel times (i.e., relative move outs),
442
which implies neglecting possible radiation patterns from double-couples. Another example
443
where such kind of mechanisms was not considered (i.e., an isotropic source was assumed)
444
to locate volcanic events using seismic amplitudes is the method introduced by Battaglia et
445
al. (2003).
446
447
Using the TREP method, we determined the epicenters of 26 NVTs in Guerrero, which are
448
consistent with previous studies in the region (Payero et al., 2008; Husker et al., 2012). The
449
epicenters are separated in two main groups, one at the “sweet spot” (Husker et al., 2012),
450
between 200 and 230 km from the trench, and another one closer to the trench at about 170
451
km. However, hypocentral depths are clearly different as compared to the only available
452
estimates (Payero et al., 2008). Our estimates show that sustained tremors are produced by
453
shear failures near the plate interface (i.e., about 60% of the whole NVT sample) with rake
15
454
angles that are subparallel to the Cocos - North America plates convergence direction.
455
Thus, NVT locations in Guerrero are in agreement with independent depth and source
456
mechanism estimates for LFEs by Frank et al. (2013), with a similar hypocentral depth
457
distributions for both types of events with maxima at 43 km depth. This result further
458
supports the idea proposed by Shelly et al. (2007) and Ide et al. (2007) stating that the
459
NVTs have the same origin as the LFEs and that the NVTs are consistent with the SSEs
460
source mechanism.
461
462
Our results also show a deeper region within the subducting slab at ~48 km depth (i.e., in
463
the oceanic crust) with ~40% of the NVT activity. This suggests that intraslab NVT sources
464
may exist in Guerrero and be related to the concentration of free fluids in the Sweet Spot as
465
a consequence of the SSEs-induced strain fields inside the Cocos plate (Cruz-Atienza et al.,
466
2011). Although the NVT events considered here do not correspond to a complete catalog,
467
our results suggest that there is almost no tremor activity in the continental crust. To assess
468
whether transient reductions (i.e., non-linear response) of the middle and deep continental
469
rocks resistance induced by the SSEs in Guerrero (Rivet et al., 2011, 2013) may trigger
470
NVTs above the plate interface, a temporal and detailed analysis of the complete tremor
471
catalogs of Guerrero should be undertaken.
472
473
5. Acknowledgments
474
475
We thank the Meso-America Subduction Experiment (MASE, 2007) for the data used in
476
this study (http://www.gps.caltech.edu/~clay/MASEdir/data_avail.html). We also thank
477
Vladimir Kostoglodov for fruitful discussions and suggestions, Eiichi Fukuyama for fruitful
478
discussions and Elena García Seco for scientific writing corrections. This work was
479
possible thanks to the UNAM-PAPIIT grant number IN113814 and the Mexican “Consejo
480
Nacional de Ciencia y Tecnología” (CONACyT) under grants number 130201 and 178058.
481
482
483
484
485
6. References
•
Andrews, D. J. (2004), Rupture Models with dynamically determined breakdown
displacement, Bull. Seismol. Soc. Am., 94(3), 769–775.
16
486
•
Battaglia, J., and K. Aki (2003), Location of seismic events and eruptive fissures on
487
the Piton de la Fournaise volcano using seismic amplitudes, J. Geophys. Res.,
488
108(B8), 2364, doi:10.1029/2002JB002193.
489
•
490
491
Rev. Earth Planet. Sci. 39:271–96.
•
492
493
Beroza, G. and Ide, S. (2011), Slow Earthquakes and Nonvolcanic Tremor, Annu.
Bohlen, T., and Saenger, E. H. (2006), Accuracy of heterogeneous staggered-grid
finite-difference modeling of Rayleigh waves, Geophysics, 71, T109–T115.
•
Bostock, M. G., A. A. Royer, E. H. Hearn, and S. M. Peacock (2012), Low
494
frequency earthquakes below southern Vancouver Island, Geochem. Geophys.
495
Geosyst., 13, Q11007, doi:10.1029/2012GC004391.
496
•
497
498
Bouchon, M. and K. Aki (1977), Discrete wavenumber representation of seismic
source wave fields, Bull. Seism. Soc. Am., 67, 259-277.
•
Brown, J. R., G. C. Beroza, S. Ide, K. Ohta, D. R. Shelly, S. Y. Schwartz, W.
499
Rabbel, M. Thorwart, and H. Kao (2009), Deep low-frequency earthquakes in
500
tremor localize to the plate interface in multiple subduction zones, Geophys. Res.
501
Lett., 36, L19306, doi:10.1029/2009GL040027.
502
•
Campillo, M., S. K. Singh, N. Shapiro, J. Pacheco, and R. B. Herrmann (1996),
503
Crustal structure south of the Mexican volcanic belt, based on group velocity
504
dispersion, Geofís. Int. 35, 361–370.
505
•
Cavalié, O., E. Pathier, M. Radiguet, M. Vergnolle, N. Cotte, A. Walpersdorf, V.
506
Kostoglodov, F. Cotton (2013), Slow slip event in the Mexican subduction zone:
507
Evidence of shallower slip in the Guerrero seismic gap for the 2006 event revealed
508
by the joint inversion of InSAR and GPS data. Earth and Planetary Science Letters,
509
367, 52–60.
510
•
511
512
nonplanar faults with a finite difference approach. Geophys. J. Int., 158, 939-954.
•
513
514
Cruz-Atienza, V. M. and J. Virieux (2004), Dynamic rupture simulation of
Cruz-Atienza, V. M. (2006), Rupture dynamique des failles non-planaires en
différences finies, Ph.D. thesis, p. 382, University of Nice Sophia-Antipolis, France.
•
Cruz-Atienza, V. M., J. Virieux and H. Aochi (2007), 3D Finite-Difference
515
Dynamic-Rupture
516
doi:10.1190/1.2766756.
Modelling
Along
Non-Planar
Faults.
Geophysics,
72,
17
517
•
Cruz-Atienza, V. M., D. Rivet, V. Kostoglodov, A. L. Husker; D. Legrand, M.
518
Campillo (2011), Toward a Unified Theory of Silent Seismicity in Central Mexico.
519
American Geophysical Union, Eos Trans. AGU, 92, Fall Meet. Suppl., S23B-2264.
520
•
Dragert, H., K. Wang, and T. S. James (2001), A silent slip event on the deeper
521
Cascadia
522
/science.1060152.
523
•
524
525
subduction
interface,
Science,
292,
1525–1528,
doi:10.1126
Flinn, E., 1965. Signal analysis using rectilinearity and direction of particle motion.
Proc. IEEE, 53, 1874–1876.
•
Franco, S. I., V. Kostoglodov, K. M. Larson, V. C. Manea, M. Manea, and J. A.
526
Santiago (2005), Propagation of the 2001–2002 silent earthquake and interplate
527
coupling in the Oaxaca subduction zone, Mexico. Earth Planets Space, 57, 973–985.
528
•
Frank, W. B., N. M. Shapiro, V. Kostoglodov, A. L. Husker, M. Campillo, J. S.
529
Payero, and G. A. Prieto (2013), Low-frequency earthquakes in the Mexican Sweet
530
Spot, Geophys. Res. Lett., 40, 2661–2666, doi:10.1002/grl.50561.
531
•
García, D., S. K. Singh, M. Herraíz, J. F. Pacheco, and M. Ordaz (2004), Inslab
532
earthquakes of central Mexico: Q, source spectra and stress drop, Bull. Seismol.
533
Soc. Am., 94, 789– 802.
534
•
Ghosh, A., J. E. Vidale, J. R. Sweet, K. C. Creager, and A. G. Wech (2009), Tremor
535
patches in Cascadia revealed by seismic array analysis, Geophys. Res. Lett., 36,
536
L17316, doi:10.1029/2009GL039080.
537
•
Husker, A. L., V. Kostoglodov, V. M. Cruz-Atienza, D. Legrand, N. M. Shapiro, J.
538
S. Payero, M. Campillo, and E. Huesca-Pérez (2012), Temporal variations of non-
539
volcanic tremor (NVT) locations in the Mexican subduction zone: Finding the NVT
540
sweet spot, Geochem. Geophys. Geosyst., 13, Q03011, doi:10.1029/2011G
541
C003916.
542
•
Husker, A., S. Peyrat, N. Shapiro, and V. Kostoglodov (2010), Automatic non-
543
volcanic tremor detection in the Mexican subduction zone, Geofis. Int., 49(1), 17–
544
25.
545
546
•
Ide, S., D. R. Shelly, and G. C. Beroza (2007), Mechanism of deep low frequency
earthquakes: Further evidence that deep non-volcanic tremor is generated by shear
18
547
slip
548
doi:10.1029/2006GL028890.
549
•
on
the
plate
interface,
Geophys.
Res.
Lett.,
34,
L03308,
Iglesias, A., S.K. Singh, A. R. Lowry, M. Santoyo, V. Kostoglodov, K. M. Larson
550
and S. I. Franco-Sánchez (2004), The silent earthquake of 2002 in the Guerrero
551
seismic gap, Mexico (Mw=7.6): Inversion of slip on the plate interface and some
552
implications, Geofísica Internacional, 43(3), 309-317.
553
•
Ito, Y., K. Obara, K. Shiomi, S. Sekine, H. Hirose (2007), Slow Earthquakes
554
Coincident with Episodic Tremors and Slow Slip Events, Science, 315, 503-506,
555
doi: 10.1126/science.1134454.
556
•
557
558
Soc. Am., Vol. 78, No. 5, pp. 1725-1743, 1988.
•
559
560
Jurkevics, A. Polarization Analysis of Three-Component Array Data. Bull. Seismol.
Kao, H. and S.-J., Shan (2004), The source-scanning algorithm: mapping the
distribution of seismic sources in time and space. Geophys. J. Int., 157, 589–-594.
•
Kao, H., S-J., Shan, H. Dragert, G. Rogers, J. F. Cassidy and K. Ramachandran
561
(2005), A wide depth distribution of seismic tremors along the northern Cascadia
562
margin, Nature, 436, 841–844, doi:10.1038/nature03903.
563
•
Kao, H., S.-J. Shan, H. Dragert, and G. Rogers (2009), Northern Cascadia episodic
564
tremor and slip: A decade of tremor observations from 1997 to 2007, J. Geophys.
565
Res., 114, B00A12, doi:10.1029/2008JB006046.
566
•
Kostoglodov, V., A. Husker, N. M. Shapiro, J. S. Payero, M. Campillo, N. Cotte,
567
and R. Clayton (2010), The 2006 slow slip event and nonvolcanic tremor in the
568
Mexican
569
doi:10.1029/2010GL045424.
570
•
subduction
zone,
Geophys.
Res.
Lett.,
37,
L24301,
Legrand, D., S. Kaneshima, and H. Kawakatsu (2000), Moment tensor analysis of
571
near-field broadband waveforms observed at Aso Volcano, Japan, J. Volcanol.
572
Geotherm. Res., 101, 155–169, doi:10.1016/S0377-0273(00)00167-0.
573
•
574
575
576
Lowry, A. R., K. M. Larson, V. Kostoglodov, and R. Bilham (2001), Transient fault
slip in Guerrero, southern Mexico, Geophys. Res. Lett., 28, 3753– 3756.
•
MASE (2007), Meso America Subduction Experiment. Caltech. Dataset.
doi:10.7909/C3RN35SP.
19
577
•
578
579
Obara, K. (2002), Nonvolcanic deep tremor associated with subduction in southwest
Japan, Science, 296, 1679–1681.
•
Obara, K., H. Hirose, F. Yamamizu, and K. Kasahara (2004), Episodic slow slip
580
events accompanied by non-volcanic tremors in southwest Japan subduction zone,
581
Geophys. Res. Lett., 31, L23602, doi:10.1029/ 2004GL020848.
582
•
Obara, K., S. Tanaka, T. Maeda, and T. Matsuzawa (2010), Depth-dependent
583
activity of non-volcanic tremor in southwest Japan, Geophys. Res. Lett., 37,
584
L13306, doi:10.1029/2010GL043679.
585
•
Payero, J. S., V. Kostoglodov, N. Shapiro, T. Mikumo, A. Iglesias, X.
586
Pérez Campos, and R. W. Clayton (2008), Nonvolcanic tremor observed in the
587
Mexican
588
doi:10.1029/2007GL032877.
589
•
subduction
zone,
Geophys.
Res.
Lett.,
35,
L07305,
Pérez-Campos, X., Y. Kim, A. Husker, P. M. Davis, R. W. Clayton, A. Iglesias, J.
590
F. Pacheco, S. K. Singh, V. C. Manea, and M. Gurnis (2008), Horizontal subduction
591
and truncation of the Cocos Plate beneath central Mexico, Geophys. Res. Lett., 35,
592
L18303, doi:10.1029/2008GL035127.
593
•
Radiguet, M., Cotton, F., Vergnolle, M., Campillo, M., Walpersdorf, A., Cotte, N.,
594
Kostoglodov, V., 2012. Slow slip events and strain accumulation in the Guerrero
595
gap, Mexico. J. Geophys. Res. 117 (B04305).
596
•
Rivet, D., Campillo, M., Shapiro, N. M., Cruz-Atienza, V., Radiguet, M., Cotte, N.
597
and Kostoglodov, V. (2011), Seismic evidence of nonlinear cristal deformation
598
during a large slow slip event inMexico, Geophys. Res. Lett., 38, L08308,
599
doi:10.1029/2011GL047151
600
•
Rivet, D., M. Campillo, M. Radiguet, D. Zigone, V. M. Cruz-Atienza, N. M.
601
Shapiro, V. Kostoglodov, N. Cotte, G. Cougoulat, A. Walpersdorf and E. Daub
602
(2013), Seismic velocity changes, strain rate and non-volcanic tremors during the
603
2009–2010 slow slip event in Guerrero, Mexico. Geophys. J. Int., doi:
604
10.1093/gji/ggt374.
605
•
Rogers, G., and H. Dragert (2003), Episodic tremor and slip on the Casca- dia
606
subduction zone: The chatter of silent slip, Science, 300, 1942 – 1943,
607
doi:10.1126/science.1084783.
20
608
•
609
610
Shelly, D. R., G. C. Beroza, and S. Ide (2007), Non-volcanic tremor and lowfrequency earthquake swarms, Nature, 446, 305–307, doi:10.1038/nature05666.
•
Shelly, D. R., G. C. Beroza, S. Ide, and S. Nakamura (2006), Low-frequency
611
earthquakes in Shikoku, Japan, and their relationship to episodic tremor and slip,
612
Nature, 442, 188–191, doi:10.1038/nature04931.
613
•
Taisne, B., F. Brenguier, N. M. Shapiro, and V. Ferrazzin (2011), Imaging the
614
dynamics of magma propagation using radiated seismic intensity, Geophys. Res.
615
Lett., 38, L04304, doi:10.1029/2010GL046068.
616
•
617
618
Telford, W. M., L. P. Geldart and R. E. Sheriff (1990), Applied Geophysics.
Cambridge University Press, U.K. ISBN: 9780521339384.
•
Vergnolle, M., A. Walpersdorf, V. Kostoglodov, P. Tregoning, J. A. Santiago, N.
619
Cotte, and S. I. Franco (2010), Slow slip events in Mexico revised from the
620
processing of 11 year GPS observations, J. Geophys. Res., 115, B08403,
621
doi:10.1029/2009JB006852.
622
•
623
624
625
Wech, A. G., and K. C. Creager (2007), Cascadia tremor polarization evidence for
plate interface slip, Geophys. Res. Lett., 34, L22306, doi:10.1029/2007GL031167.
•
Wech, A., and K. C. Creager (2008), Automated detection and location of Cascadia
tremor, Geophys. Res. Lett., 35, L20302, doi:10.1029/2008GL035458.
626
21
627
Figure Captions
628
629
Figure 1 Study region and tectonic setting. Colored circles with red arrows represent NVT
630
locations and mechanisms (i.e., slip directions) from this study. The figure also shows NVT
631
epicenters (green dots) determined by Husker at al. (2012); LFE epicenters (red dots) and
632
the associated focal mechanism (gray beach ball) determined by Frank et al. (2013); the
633
MASE array stations (blue triangles) used in this study (i.e. those inside the black
634
rectangle); the horizontal projection of the 3D hypocentral lattice used by the TREP method
635
(dashed rectangle); and the rupture area of major earthquakes in Mexico (shaded forms).
636
637
Figure 2 Spectral analysis of a 24-h seismogram recorded on March 6, 2005 in the north-
638
south component of station SATA (Figure 1). (a) Energy spectrogram of the signal; (b)
639
rectilinearity spectrogram of the particle motion; and (c) 1 – 2 Hz band-pass filtered signal.
640
Vertical dashed lines depict the NVT located in Figure 3a.
641
642
Figure 3 Location of two NVTs recorded on March 6, 2005 (left column) and November 2,
643
2005 (right column) using the TREP method. (a and d) Horizontal and vertical cross-
644
sections of cost function
645
(yellow stars), observed and synthetic particle motion polarization directions (red and blue
646
arrows), and best solutions slip directions (yellow arrows); (b and e) best-fits for the energy
647
distribution along the stations array; and (c and f) best-fits for the spatial derivatives of the
648
energy distributions.
(background colors) through the best hypocenter location
649
650
Figure 4 (a) MASE stations array (circles, see Figure 1) and hypocentral lattice
651
(crosses) used in the synthetic inversion tests; and (b) penny-shape cracks with
652
variable shear pre-stress fields, τ (white arrows) around a main direction (black
653
arrow).
654
655
Figure 5 (a) Synthetic tremor-like seismograms computed at station XALI (see Figure
656
6) with the source model described here; and (b, c and d) synthetic seismograms in
657
the three components along the array (Figure 4) corresponding to the best Green
22
658
functions and source mechanism find by the TREP method in the inversion test shown
659
in the left column of Figure 6.
660
661
Figure 6 Location of two NVTs tremor-like synthetic sources (green stars and arrows)
662
using the TREP method. (a and d) Horizontal and vertical cross-sections of cost function
663
(background colors) through the best hypocenter location (yellow stars), observed and
664
synthetic particle motion polarization directions (red and blue arrows), and best solutions
665
slip directions (yellow arrows); (b and e) best-fits for the energy distribution along the
666
stations array; and (c and f) best-fits for the spatial derivatives of the energy distributions.
667
668
Figure 7 Locations of NVTs using the TREP method (colored circles) and LFEs by Frank
669
et al. (2013) (blue circles) projected into a vertical trench-perpendicular cross-section. On
670
the left, comparison of normalized histograms for both kinds of events (red and blue curves
671
for NVTs and LFEs, respectively).
23
20
North America plate
Mexico City
ATLA
Cocos plate
AMAC
MAXE
LFEs
PLLI
ge
er
PLAT
MAZA
ACAH
Sl
ab
co
nv
20
14
100
85
PLAY
14
17
1979
20
19
Acapulco
cm
/yr
196
2
1 95
19
MA
T
-101
−1
−1.5
−2
−2.5
7
89
20
16
-102
15
T r en 0
ch-pe
rp
HUIT
nc
1943
12 1 9
82
Guerrero
1995
1937
85
e
19
Log of NVTs per
cubic kilometer
TOMA
SATA
5.4
Latitude, N°
BUCU
20
endic 0
ular d
2
istan 50
ce (k
m)
SJVH
SAFE
18
30 0
Pacific plate
19
-100
Longitude, E°
-99
1996
1 96
8
-98
Frequency (Hz)
a
Frequency (Hz)
b
10
9
8
7
6
5
4
3
2
1
10
9
8
7
6
5
4
3
2
1
0.8
0.75
0.7
−7
c
Velocity (cm/s)
0.85
x 10
5
0
−5
0
1
2
3
4
5
6
7
8
9
10
11
12
13
Time (hr)
14
15
16
17
18
19
20
21
22
23
24
a
d
Tremor Location and Polarization Directions
Tremor Location and Polarization Directions
Coast−Parallel Distance (km)
80
Event Date: 2005/03/06 (13:00)
Event Date: 2005/11/02 (17:30)
N
N
60
Source Rake
40
5.4 cm/yr
QUEM
XALI
PLLI
PLAT
20
TONA
ZACA
5.4 cm/yr
TOMA CIEN PALM SAFE
SATA
VEVI
VEVI
PLAT
−0.2
ZURI
XOLA
HUIT
MAZA
TONA
QUEM
AMAC
CASA
BUCU
ZACA
XALI
CIEN PALM SAFE
SATA
CASA
AMAC
−0.2
ZURI
ACAH
Source Rake
PLLI
HUIT
−0.4
0
Observed
Synthetic
−0.4
Observed
−0.6
Synthetic
−20
−0.6
−0.8
−0.8
10
−1
Depth (km)
20
−1
−1.2
30
Cocos Plate
Cocos Plate
40
Source Hypocenter
50
Source Hypocenter
60
150
200
250
Trench−Perpendicular Distance (km)
b
Observed
2
QUEM
PLAT
MAZA
ZURI
8
HUIT
BUCU
XALI
VEVI
SAFE
CIEN
PLLI
PALMAMAC
Normalized Energy
0
East-West Component
6
4
2
0
c
17.2
2
17.4
17.6
17.8
Latitude
18
18.2
18.4
0
2
Synthetic
East-West Component
0
−2
17
17.2
17.4
17.6
17.8
Latitude
18
18.2
18.4
18.6
SATA
TONA
PLLI
Synthetic
ZACA
VEVI
QUEM
ACAH
XOLA
ZURI PLAT
0
SAFE
CIEN
HUIT
8
AMAC
PALM
East-West Component
6
4
2
0
17
Energy Derivative
Observed
Observed
4
f
North−South Component
−2
6
2
XALI
North−South Component
8
18.6
Energy-Derivative-Distribution Best-Fit
−17
x 10
Energy Derivative
ZACA
4
17
Energy Derivative
SATA
Synthetic
Energy Derivative
Normalized Energy
Normalized Energy
6
TOMA
300
Energy-Distribution Best-Fit
−12
x 10
TONA
North−South Component
150
200
250
Trench−Perpendicular Distance (km)
e
Energy-Distribution Best-Fit
−13
x 10
8
100
300
Normalized Energy
100
17.2
17.4
17.6
17.8
Latitude
18
18.2
18.4
18.6
18.4
18.6
Energy-Derivative-Distribution Best-Fit
−16
x 10
4
2
North−South Component
0
−2
Observed
−4
Synthetic
4
2
East-West Component
0
−2
−4
17
17.2
17.4
17.6
17.8
Latitude
18
18.2
a
b
250
Horizontal distance (km)
12
North-South (km)
200
150
100
50
0
10
8
6
4
0
0
50
100
150
East-West (km)
τ
2
0
2
4
6
8
10
Horizontal distance (km)
12
a
b
−5
North−South Component
Velocity (m/s)
x 10
SAFE
PALM
BUCU
CIEN
−5
TOMA
East−West Component
TEPO
0.5
ZACA
SATA
5
Velocity (m/s)
Velocity (m/s)
CASA
0.6
−5
0
−5
TONA
MAXE
0.4
XALI
PLLI
VEVI
0.3
HUIT
PLAT
−5
5
ZURI
Vertical Component
x 10
MAZA
0.2
ACAH
XALI
RIVI
TICO
0.1
0
XOLA
PLAY
EL40
QUEM
0
−5
10
20
30
40
50
60
70
0
80
Time (s)
c
APOT
AMAC
AMAC
CASA
CASA
CIEN
TOMA
TOMA
Velocity (m/s)
TONA
MAXE
XALI
PLLI
VEVI
HUIT
MAXE
XALI
PLLI
VEVI
0.1
HUIT
PLAT
PLAT
ZURI
ZURI
MAZA
ACAH
0.05
RIVI
RIVI
TICO
TICO
XOLA
XOLA
PLAY
PLAY
EL40
EL40
QUEM
0
60
TONA
ACAH
0
50
SATA
MAZA
0.05
40
ZACA
0.15
SATA
0.1
60
TEPO
ZACA
0.15
50
BUCU
CIEN
TEPO
0.2
40
PALM
0.2
BUCU
0.25
30
Time (s)
SAFE
PALM
0.3
20
Vertical Ground Velocity
APOT
SAFE
0.35
10
d
East−West Ground Velocity
0.4
Velocity (m/s)
AMAC
0.7
0
x 10
Velocity (m/s)
North−South Ground Velocity
APOT
5
QUEM
0
10
20
30
Time (s)
40
50
60
0
10
20
30
Time (s)
d
Cost Function and Polarization Directions
100
N
80
MAXE
QUEM
60
EL40
TICO
PLAY
40
PLAT
ZURI
ACAH
XOLA
PLLI
VEVI
XALI
TONA SATA TOMA
BUCU CASA
AMAC
APOT
HUIT
−0.5
MAZA
RIVI
Source Rake
20
Observed
0
ZACA TEPO CIEN
PALM SAFE
−1
Synthetic
−1.5
Best-fit
40
50
100
Normalized Deri
North−South Component
5
0
ZACA
MAXE
10
XALI
XOLA
EL40
VEVI
PLAY TICO RIVI ACAH MAZA ZURI PLAT
QUEM
HUIT
TEPO
PLLI
TOMA
BUCU
CIEN
SAFE
CASA
APOT
PALM
AMAC
East-West Component
10
5
0
20
Observed
15
Synthetic
10
5
0
85
105
125
145
165
185
205
225
245
265
285
Trench−Perpendicular Distance (km)
North−South Component
5
0
−5
Observed
Synthetic
0
−5
East-West Component
−10
CASA
APOT
−0.5
MAZA
RIVI
Source Rake
Observed
−1
Synthetic
−1.5
−2
30
Target
Moho
Best-fit
40
100
150
200
250
Energy Distribution Best-Fit
−6
SATA
North−South Component
1.5
0
TEPO
TOMA
ZACA
1
0.5
300
Trench−Perpendicular Distance (km)
x 10
2
XOLA
EL40
XALI
PLAY TICO RIVI ACAH MAZA ZURI PLAT VEVI
QUEM
PLLI
HUIT
2
TONA
CIEN
SAFE
CASA APOT
BUCU
PALM
AMAC
MAXE
East-West Component
1.5
1
0.5
0
2
Observed
1.5
Synthetic
1
0.5
0
85
105
125
145
165
185
205
225
245
265
285
265
285
Trench−Perpendicular Distance (km)
Energy-Derivative Distribution Best-Fit
−11
x 10
5
North−South Component
0
−5
Observed
−10
5
BUCU
20
f
Energy-Derivative Distribution Best-Fit
−11
x 10
HUIT
10
Normalized Energy
20
15
XOLA
0
e
SATA
TONA
ZURI
ZACA TEPO CIEN
PALM SAFE
AMAC
TONA SATA TOMA
ACAH
50
300
Energy Distribution Best-Fit
−7
−10
Normalized Deri
250
Trench−Perpendicular Distance (km)
x 10
15
c
Synthetic
5
0
−5
East-West Component
−10
5
Normalized Deri
Normalized Deri
200
PLAT
TICO
PLAY
20
Normalized Deri
Normalized Energy
Normalized Energy
Normalized Energy
b
150
XALI
VEVI PLLI
EL40
40
Depth (km)
Target
Moho
MAXE
QUEM
60
Normalized Energy
30
N
80
Normalized Deri
Depth (km)
20
20
100
−2
10
Cost Function and Polarization Directions
120
Coast−Parallel Distance (km)
120
Normalized Energy
Coast−Parallel Distance (km)
a
0
−5
−10
85
105
125
145
165
185
205
225
Trench−Perpendicular Distance (km)
245
265
285
5
0
−5
−10
85
105
125
145
165
185
205
225
Trench−Perpendicular Distance (km)
245
Logarithm of NVT per Cubic Kilometer
Depth (km)
0
NVTs this study
LFEs by Frank et al.
Plates Interface
20
North American Plate
−1.5
−2
40
NVTs
60
100
LFEs
Cocos Plate
150
200
250
Trench Perpendicular Distance (km)
−2.5
300