etas fore 3 3 · 2011. 2. 24. · ! 6! 102!...
TRANSCRIPT
1
The Spatial Density of Foreshocks 1
Emily E. Brodsky 2
Dept. of Earth & Planetary Sciences 3
UC Santa Cruz; [email protected] 4
Abstract 5
Aftershocks follow a well-‐defined spatial decay pattern in the intermediate field. Here I 6
investigate the same pattern for foreshocks. Foreshock linear density decays as r-‐1.5+/-‐0.1 over 7
distances r of 0.1-‐30 km for 15 minutes after magnitude 3-‐4 mainshocks. This trend is the same 8
as that of the aftershocks within the error of the measurement. This consistency of spatial 9
decays can be explained by the clustering inherent in earthquake interactions. No additional 10
preparatory process beyond earthquake triggering is necessary to explain the spatial decay. 11
Introduction 12
Earthquake triggering is a possible route into studying earthquake initiation and, by 13
extension, predictability. Earthquake interactions are generally dissected by looking at 14
relationships between earthquakes that are close in either time or space. For instance, 15
Felzer and Brodsky [2006] examined the spatial pattern of smaller earthquakes following 16
small magnitude mainshocks within 5 minutes. The study found that the linear density ρ of 17
aftershocks followed a well-‐defined pattern of ρ ∝r-‐γ where r is the distance from the 18
mainshock. The persistence of this spatial trend to large distances was interpreted as 19
indicative of a consistent triggering process across the entire range of distances. Since 20
2
seismic waves are thought to be significant triggerers at great distances, the inference was 21
those seismic waves were an important part of triggering at all distances [Hill et al., 1993; 22
Brodsky et al., 2000; West et al., 2005; Hill and Prejean, 2007; Van der Elst and Brodsky, 23
2010]. 24
Foreshocks are the best-‐known predictor of earthquakes. Around ~50% of earthquakes are 25
preceded by observable foreshocks [Abercrombie and Mori, 1996]. Early thinking 26
attributed foreshocks to a stress accumulation process that ultimately culminated in a large 27
earthquake [e.g., Jones and Molnar, 1979]. In this scenario, the spatial distribution of 28
foreshocks is controlled by the stress distribution and fatigue on the fault plane. More 29
recent work has entertained other possibilities such as the foreshocks being symptomatic 30
of an earthquake cascade. Earthquakes are constantly triggering each other. When 31
seismicity is tightly clustered, the likelihood of a large earthquake increases and, after the 32
fact, the seismicity cluster is interpreted as foreshocks [Helmstetter, et al., 2003; Felzer et 33
al., 2004]. This type of clustering results in distinct patterns, like the Inverse Omori’s Law 34
[Helmstetter et al., 2003]. Still other work has suggested that both kinds of foreshocks 35
exist in earthquake catalogs [McGuire et al, 2005; Zhuang et al., 2008]. 36
This paper starts by establishing that the foreshock spatial trend follows a similar 37
relationship to the aftershock one. At first, this observation is disconcerting. If the spatial 38
decay of r-‐γ is generated by aftershocks, then why is it present before the mainshock? I will 39
follow up the observation by showing that a statistical seismicity model in which 40
aftershocks follow the spatial decay law of r-‐γ automatically generates foreshocks with the 41
3
same spatial decay. The trend can be a natural consequence of the clustering provided by 42
the mutual triggering of the earthquake cascade. 43
Observation 44
I begin by using the well-‐located LSH catalog of Southern California earthquakes 1981-‐45
2005 to compare the spatial distribution of aftershocks and foreshocks [Lin et al., 2007]. 46
Only magnitudes >2 are included to ensure completeness. For the purpose of comparison, I 47
use a similar windowing criteria as Felzer and Brodsky [2006] to isolate sequences. 48
Mainshocks are defined as earthquakes that are at least 4 days after and 0.5 day before any 49
larger earthquake at any distance in the catalog from any larger earthquake. Aftershocks 50
are defined to be earthquakes that follow a larger identified mainshock within the specified 51
time period Δt. Foreshocks are earthquakes that are followed by a larger identified 52
mainshock within a specified time period Δt. When comparing foreshocks and aftershocks 53
for a particular sequence, the same value of Δt is used to define both types of seismicity. 54
These windowing criteria are not meant to imply a physical limit to interaction. The 55
insensitivity of the aftershock spatial decay to the windowing details was established in 56
previous work [Felzer and Brodsky, 2006]. The point of the present study is simply to 57
compare foreshock decays to aftershock decays given the same conditioning on the data. 58
To measure density, the mainshocks (and their accompanying sequences) for the entire 59
catalog are combined to improve the statistical sampling. A single, ordered vector of 60
aftershock-‐mainshock hypocentral distances is created, r . The linear density of 61
4
aftershocks (and foreshocks) is estimated at the midpoints between each element of the 62
combined aftershock distance vector r by 63
ρ ri + ri+12
⎛⎝⎜
⎞⎠⎟=
1ri+1 − ri
(1) 64
For mainshocks with magnitude <4, the fault rupture length is expected to be less than the 65
spatial precision of the data (~0.1 km) and therefore the point approximation embedded in 66
eq. 1 is appropriate. Larger magnitude mainshocks would require a better geometrical 67
model for earthquake density estimation. 68
The aftershocks following the magnitude 3-‐4 mainshocks yields a fit of γ=1.5+/-‐ 0.08 for 69
data from distances of 0.1-‐30 km for aftershocks for 15 minute (Figure 1). (Error ranges 70
here and throughout the paper are standard deviations of 1000 bootstrap trials on the fit). 71
This observation is similar to Richard-‐Dinger and Stein [2010]. This value of γ is defined 72
slightly differently than Felzer and Brodsky [2006] because it is based on epicenters, not 73
hypocenters. This study uses epicenters to facilitate comparison with the 2-‐D ETAS model 74
below. For Δt= 5 minutes, a similar trend appears, but far fewer foreshocks are recorded 75
(Figure 1b). 76
For close distances, the largest mainshocks in this range can no longer be treated as point 77
sources and so the distance r is likely inaccurate. For large distances, the number of 78
triggered events is sufficiently small that unrelated seismicity overwhelms the signal. The 79
background level can be seen clearly by shuffling the event times randomly for the same 80
earthquake locations and re-‐measuring the density. The thin light lines in Figure 1 show 81
5
100 such reshufflings and the thick light lines show the mode values of the time-‐82
randomized set. As an unusually closely spaced pair of events can easily dominate the 83
median or mean of the density, the mode is the best representative of the ordinary 84
background level of seismicity. This background level demonstrated that indeed the 85
background interferes with recording the aftershock signal at large distances, but an 86
aftershock signal is visible up to at least 30 km distance from the mainshock. 87
Interestingly, the foreshocks follow almost the same trend (Figure 1). For 15 minutes 88
before, the decay of the composite foreshock sequence from the future mainshock site 89
follows r-‐1.5+/-‐0.1 and for the 5 minutes foreshocks the density is r1.6+-‐/-‐0.2. The foreshock 90
decay is also truncated by the background seismicity at large distances. 91
92
Clustering Model 93
In order to recreate the observed foreshock trend, I will employ an Epidemic Triggering 94
Aftershock Sequence (ETAS) model that combines empirical statistical laws of seismicity to 95
generate self-‐consistent synthetic earthquake catalogs [Ogata, 1999]. An earthquake 96
triggers successive earthquakes with a probability determined by the magnitude of the 97
mainshock. Omori’s Law and a spatial decay relationship determine the time and distance 98
separating the aftershock from the mainshock. Successive earthquake magnitudes are then 99
determined by Gutenberg-‐Richter statistics and then each earthquake generates its own 100
aftershocks. 101
6
The ETAS formulation has been explored analytically and numerically in the literature 102
[Helmstetter and Sornette, 2002; Felzer et al., 2004; McGuire et al., 2005; Zhuang et al., 103
2007]. Windowing and selection criteria often result in behaviors that are challenging to 104
evaluate analytically. I therefore use a numerical model directly for this work. The 105
parameters used here are meant to illustrate the relationship between the foreshock and 106
aftershock spatial decay. They are certainly not exhaustive, but are realistic and suffice to 107
elucidate the role of clustering in generating a foreshock spatial pattern. Variations on the 108
basic ETAS model are implemented by utilizing different space or time modifications of the 109
observational laws. As the permutations of the model are important and the practicalities 110
of implementation are sometimes difficult, I will go into some detail on the exact forms 111
used in the numerical calculations done here. 112
ETAS Implementation 113
For this particular implementation, background seismicity is imposed as a Poissonian 114
process with rate λ is over the duration of the observed catalog (24 years) with the 115
magnitude of each earthquake determined by Gutenberg-‐Richter. The background rate is 116
chosen so that the total number of synthesized events (including aftershocks) is 117
approximately equal to the observed number. Each of these background earthquakes 118
generates aftershocks, which in turn spawn their own sequences. The cascade is produced 119
by the combination of four empirical equations cast in terms of probability distributions: 120
I) Gutenberg-‐Richter 121
The number of aftershocks with magnitudes greater than or equal to M is governed by 122
Log N(M)= -‐bM+a (2) 123
7
where a and b are constants. 124
125
To obtain a cumulative density function (CDF), N(M) is normalized by the total number of 126
earthquakes in the catalog, i.e., N(Mmin) where Mmin the minimum magnitude of the 127
simulated catalog. A random number p1 is then selected between 0 and 1 for each 128
aftershock and inverted using the CDF into a magnitude, i.e., 129
M=-‐log (p1(10-‐b Mmin -‐10-‐b Mmax))/b (3) 130
where p1 is a random variable between 0 and 1 and Mmax is the maximum magnitude size to 131
be generated. The extra term in brackets is a normalization that accounts for discarding 132
values of p1 that result in M>Mmax [Sornette and Werner, 2005]. 133
134
II) Omori’s Law 135
The rate of seismicity as a function of time t from a mainshock is 136
dNAS/dt =K/(t+Comori)β (4) 137
where β and Comori are constants and K is a function of mainshock magnitude to be 138
discussed below. The constant Comori has proven particularly problematic to measure as it is 139
often biased by completeness problems at short times after an earthquake. Recent work 140
has suggested that Comori is significantly shorter than previously imagined [Peng et al., 141
2007]. 142
In probabilistic terms, 143
t=((tmax + COmori)1-‐β p2 + (1-‐p2) COmori(1-‐β))1/(1-‐β) -‐ COmori (5) 144
where p2 is a random variable between 0 and 1. 145
146
8
III) Aftershock Productivity 147
The numerator of Omori’s Law determines the number of aftershocks from a mainshock of 148
magnitude M 149
K = C10α(M-‐Mmin) (6) 150
where C and α are constants. 151
Studies agree that α=1 for Southern California seismicity [Felzer et al., 2004; Helmstetter et al., 152
2005]. The productivity constant is much more poorly determined, in part because it depends on 153
the minimum active magnitude [Hardebeck et al., 2008]. Furthermore, the parameter C above is 154
defined for the instantaneous rate in Omori’s Law. In order to use eq. 6, the total number of 155
aftershocks must be computed for a sequence and so in practice what is required is the integral of 156
eq. 4 over the finite time tmax, that the aftershock sequence will be simulated. The practical 157
equation is therefore, 158
NAS = C’ 10α(M-‐Mmin) 159
where C’=C((tmax + COmori)1-‐β -‐ COmori (1-‐β))/(1-‐β) (7) 160
Eq. 7 is implemented deterministically based on aftershock magnitudes. In reality, aftershock 161
productivity probably varies in time and space, but those deviations are not sufficiently well 162
mapped out to justify a more sophisticated implementation at this time. In the simulations, I 163
apply a value of C’ that fits the observed productivity in the LSH data set with the aftershock 164
identification criteria used here (Figure 2). 165
For non-integer value of NAS, the fractional part is interpreted as a probability. For each 166
sequence, a uniform random variable p is generated between 0 and 1 and if the value is less than 167
the non-integer fraction, an extra aftershock is added. 168
9
169
IV) Distance Decay 170
The distance fall-‐off is implemented by normalizing the distribution to form a CDF 171
assuming that the aftershocks are distributed a distances greater than dmin from the 172
mainshock. Since γ>1, no upper bound to the distance fall-‐off is necessary. The 173
probabilistic form for determining r for each aftershock is 174
r = dmin (1-‐p3) 1/(1-‐γ) (8) 175
where p3 is a random variable between 0 and 1. As there are no constraints on the angular 176
distribution an angle θ is selected with uniform probability from 0 to 2π. 177
Combining eqs. 3, 5, 7 and 8 yields an ETAS catalog. In the implementation of ETAS, it is 178
helpful to note the order of calculation. First the total number of aftershocks must be 179
calculated for each extent earthquake using eq. 7, then each aftershock is assigned a time, 180
magnitude and location using eqs. 3, 5 and 8. Then the aftershocks are allowed to generate 181
their own aftershocks by the same procedure until the cascade comes to an end. 182
Simulation Results 183
The ETAS catalog was made to mimic the catalog of Figure 1 as closely as possible. The 184
parameters of Table 1 were used and the same windowing criteria were used to isolate 185
mainshocks as in Figure 1. The same stacking procedure and the same windows for 186
identifying 15 and 5-‐minute aftershock and foreshocks were used. 187
10
The overall trends of the Figure 1 are reproduced in the simulated catalog (Figure 3). Both 188
the aftershock and foreshock densities have a best-‐fit spatial decay consistent with the 189
imposed exponent. This consistency in foreshock and aftershock behavior exists even 190
though there was no specific foreshock preparatory process in the ETAS model. The utility 191
of the statistical simulation is that it elucidates that non-‐intuitive behavior, like Figure 1, 192
emerges naturally from the earthquake sequences. Like the Inverse Omori’s Law, the 193
spatial decay of the foreshocks is symptomatic of the type of increased seismicity that is 194
likely to trigger a large earthquake. Figure 3 demonstrates that the observed parallel 195
between the foreshock and aftershock decay can arise naturally from a clustering model. 196
At very large distances, the simulated catalog has larger clusters due to unisolated ongoing 197
sequences. The isolation distance of 100 km based on the SCSN catalog is insufficient for 198
the simulation because of the more homogeneous distribution of epicenters, i.e., there are 199
no simulated faults. 200
There is one conspicuous difference between the simulations and the real catalog. The ratio 201
of identified aftershocks to foreshocks in the real catalog is much lower than in the 202
simulated catalog. In the actual catalog, the ratio is 2.2 for Δt=5 min and 2.8 for Δt=15 min., 203
while the simulations generally result in ratios between 7 and 8 for the parameters used 204
here. Aftershock-‐foreshock ratios have been analyzed extensively elsewhere [e.g., McGuire 205
et al., 2005] and are not the focus of this study. Nonetheless, some assessment of the source 206
of the discrepancy is in order. 207
One possibility is that the observed catalog artificially depresses the aftershock/foreshock 208
ratio because of completeness issues. It is generally more difficult to detect aftershocks 209
11
than foreshocks because the mainshock and temporally high seismicity rate can obscure 210
individual events. Therefore, the earthquakes that are above the magnitude of catalog 211
completeness for aftershocks can be significantly higher than normal [e.g., Peng et al., 212
2007]. 213
I assess the effects of catalog incompleteness following mainshocks issue in two different 214
ways. First, I restrict the data to a higher completeness level consistent of M=2.5 (Figure 4). 215
This restriction decreases the simulated aftershock/foreshock ratio as the ratio is sensitive 216
to the range of magnitudes between the mainshock and Mth [McGuire et al., 2005]. For 217
Δt=15 minute, the simulated ratio in the realization of Figure 4c is 3.8. In contrast, the 218
observed aftershock to foreshock ratio for the Δt= 15 minute data increases slightly to 2.9. 219
This improved consistency between simulation and observation is expected based on 220
previous work on foreshock rates in the region [Felzer et al., 2004] and indicates that the 221
lack of aftershock detection is an important issue in the observations. However, the 222
simulated ratios here still do not exactly match the observations. It is possible that the 223
completeness threshold needs to be raised still further to ensure aftershock detection, but 224
any further restrictions limit the dataset to an unreasonable small size. 225
As an alternative method of probing time-‐variable completeness, I exclude a short window 226
immediately before and following the mainshock in measuring the aftershock and 227
foreshock densities (Figure 5). This strategy has the advantage of preserving the number of 228
events at larger times when the completeness threshold drops to the overall catalog value. 229
Excluding a window of 1 minute on either side of the mainshock increases the 230
aftershock/foreshock ratio to 3.3 for Δt= 15 min. This ratio is still smaller than the 231
12
simulated ratio, but it does again indicate that early-‐time aftershock detection is likely a 232
problem in the observations. 233
The aftershock-‐foreshock ratio can be adjusted in the simulations reducing the minimum 234
magnitude of simulation Mmin or increasing the productivity constant C’, both of which have 235
the effect of increasing the average number n of aftershocks per mainshock [Sornette and 236
Werner, 2005]. As n increases, the system approaches a critical state and the foreshock rate 237
should increase relative to the aftershock rate [McGuire et al., 2005]. The calculations also 238
become prohibitively expensive. Thus far, the observed foreshock to aftershock ratio has 239
not been simulated. One possible explanation is that there is a genuine difference in the 240
ETAS process from the observations. If so, the aftershock/foreshock ratio is the only 241
evidence of such a process. Another explanation is that aftershock completeness is still the 242
overwhelming observational problem. 243
More importantly for this study, the spatial decay of the foreshock density in both Figures 4 244
and 5 remains consistent with the aftershock decay. This consistency of Figure 5 is 245
particularly useful in light of the commentary of Richards-‐Dinger et al. [2010] that the 246
observed aftershock density decay may be controlled by events that occurred before the 247
arrival of the seismic waves. No such events are included in the fits of Figure 5 as P waves 248
travelling at 6 km/s will have passed through 360 km in the first minute. 249
Conclusions 250
Earthquakes are preceded by a halo of earthquakes that decays with distance much like 251
aftershocks. These foreshocks are an expected consequence of earthquake interaction and 252
13
clustering. Their existence and location near the eventual rupture does not in itself indicate 253
a preparatory strain accumulation process beyond ordinary earthquake triggering. 254
A foreshock preparatory process could exist in the Earth exists beyond the clustering 255
process, but it is not required by the observations. The only potential evidence seen here 256
for such a preparatory process is in the ratio of total number of foreshocks to aftershocks. 257
However, the data indicate that catalog completeness is a serious problem that could 258
artificially depress the aftershock/foreshock ratio and therefore this ratio is not the most 259
reliable indicator of the underlying physics. The spatial decay of both the aftershocks and 260
foreshocks is well explained by the clustering model. 261
262
Acknowledgements 263
Many thanks to Karen Felzer, Keith Richards-‐Dinger, Peter Shearer, Ross Stein and Agnes 264
Helmstetter for many helpful conversations and ideas. This work was funded in part by the 265
Southern California Earthquake Center. 266
267
14
a267
b 268
Figure 1. Aftershocks and foreshocks separated from the mainshock by time (a) Δt= 15 minutes and (b) Δt =5 269
minutes for mainshock magnitudes 3-4. A least-squares fit of the 0.1-30 km data yields γ=1.5+/- 0.08 for 270
aftershocks and 1.5+/-0.1 for foreshocks. For 5 minute data, γ =1.5+/-0.1 and 1.6 +/- 0.2 for aftershocks and 271
foreshocks, respectively. Time-randomized catalog super-imposed to illustrate background level of density. The 272
15
thin, light lines illustrate each of 100 time-randomized catalogs for the same aftershock (blue) and foreshock 273
(red) selection criteria as used in the original data. The thick lines are the mode values for logarithmic bins of 274
these 100 realizations. 275
276
16
276
277 Figure 2. Simulate and observed identified aftershocks as a function of magnitude using the parameters in Table 278
1. The fit between the simulation and observations for the magnitudes in the study range (M 3-4) is used to 279
calibrate the simulation aftershock productivity C’. Solid lines indicate best fit trends to the simulations and 280
observations for M 3-4 mainshocks. The apparent depletion of aftershocks for mainshocks with magnitude> 4 is 281
likely a completeness issue as documented elsewhere (Felzer et al., 2004; Helmstetter et al., 2005]. 282
283
284
17
a284
b 285
Figure 3 Simulated aftershock and foreshock decay from the ETAS parameters in Table 1 with (a) Δt=15 minutes 286
and (b) Δt=5 minutes. The best-fit decay exponent is γ=1.5/-0.04 for the aftershocks and 1.4+/-0.09 for the 287
foreshocks for Δt=15 minutes. For Δt=5 minutes, γ=1.5+/-0.05 for the aftershocks and 1.5 +/-0.1 for the 288
foreshocks. 289
18
290
a291
b 292
19
c293
d 294
Figure 4. Observed and simulated aftershock and foreshock sequences with a completeness magnitude of 295
Mth=2.5. All other simulation parameters are the same as Table 1. (a) Observed earthquake densities for Δt=15 296
minutes and (b) Δt=5 minutes. (c) Simulated earthquake densities for Δt=15 minutes and (d) Δt=5 minutes. 297
298
20
a298
b299
21
c300
d 301
Figure 5. Observed and simulated aftershock and foreshock sequences excluding a 1 minute interval both before 302
and after the mainshock. All other simulation parameters are the same as Table 1. (a) Observed earthquake 303
densities for Δt=15 minutes and (b) Δt=5 minutes. (c) Simulated earthquake densities for Δt=15 minutes and (d) 304
Δt=5 minutes. 305
306
22
306
307
308
309
310
Parameter Values Rationale
Completeness threshold Mth 2 Catalog completeness
tmax 24 years Catalog duration
Mmin 0 Computationally limited
Spatial Decay exponent γ 1.5 Aftershock density spatial fit (Figure 1)
Productivity Constant C’ 0.03 Earthquakes/day Aftershock productivity fit (Figure 3)
Omori Exponent β 1.34 Hardebeck et al. [2008]
Background earthquake rate λ for
M>Mmin
300 Earthquakes/day Matches observed catalog rate with aftershocks are included
Minimum aftershock distance dmin 100 m Minimum location accuracy
Omori Delay Comori 130s Peng et al. [2007]
b 1 Standard value
α 1 Felzer et al. [2004]; Helmstetter et al. [2005]
Table 1. ETAS Parameters for simulations in Figures 2-5. 311
312
23
References 312
Abercrombie, R.E. and J. Mori (1996), Occurrence patterns of foreshocks to large 313
earthquakes in the western United States, Nature, 381, 303-‐307. 314
Brodsky, E. E., V. Karakostas, and H. Kanamori (2000), A New Observation of Dynamically 315
Triggered Regional Seismicity: Earthquakes in Greece Following the August, 1999 Izmit, 316
Turkey Earthquake, Geophys. Res. Let., 27, 2741-‐2744. 317
Felzer, K. R., R. E. Abercrombie, and G. Ekstrom (2004), A common origin for aftershocks, 318
foreshocks, and multiplets, Bull. Seism. Soc. Am., 94, 88-‐98. 319
Felzer, K. R. and E. E. Brodsky (2006), Decay of aftershock density with distance indicates 320
triggering by dynamic stress, Nature , 441, 735-‐738. 321
Helmstetter, A. and D. Sornette (2002) Subcritical and supercritical regimes in epidemic 322
models of earthquake aftershocks, J. Geophys. Res., 107, B10, 2237, doi: 323
10.1029/2001JB0001580. 324
Helmstetter, A., D. Sornette, and J.-‐R. Grasso (2003), Mainshocks are aftershocks of 325
conditional foreshocks: How do foreshock statistical properties emerge from aftershock 326
laws, J. Geophys. Res., 108(B1), 2046, doi:10.1029/2002JB001991. 327
Helmstetter, A., Y. Y. Kagan, and D. D. Jackson (2005), Importance of small earthquakes for 328
stress transfers and earthquake triggering, J. Geophys. Res., 110, B05S08. 329
Hill, D. P., et al. (1993), Seismicity remotely triggered by the magnitude 7.3 Landers, 330
California, earthquake, Science, 260(5114), 1617-‐1623. 331
24
Hill, D. P., and S. G. Prejean (2007). Dynamic triggering, in Earthquake Seismology, Treatise 332
on Geophysics, H. Kanamori (Editor), Elsevier, Amsterdam. 333
Jones, L. M., and P. Molnar (1979), Some characteristics of foreshocks and their possible 334
relationship to earthquake prediction and premonitory slip on faults, J. Geophys. Res., 335
84(B7), 3596–3608. 336
Lin, G., P. M. Shearer, and E. Hauksson (2007), Applying a three-‐dimensional velocity 337
model, waveform cross correlation, and cluster analysis to locate southern California 338
seismicity from 1981 to 2005, J. Geophys. Res., 112, B12309, doi:10.1029/2007JB004986. 339
McGuire, J.J., M.S. Boettcher and T.H. Jordan (2005), Foreshock sequences and short-‐term 340
earthquake predictability on East Pacific Rise transform faults, Nature 434, 457-‐461. 341
Ogata, Y. (1999), Seismicity Analysis through Point-‐process Modeling: A Review, Pure Appl. 342
Geophys. 155, 471-‐507. 343
Peng, Z., J. E. Vidale, M. Ishii, and A. Helmstetter (2007), Seismicity rate immediately before 344
and after main shock rupture from high-‐frequency waveforms in Japan, J. Geophys. Res., 345
112, B03306, doi:10.1029/2006JB004386. 346
Richards-‐Dinger, K.B., R. Stein, and S. Toda, Decay of aftershock density with distance does 347
not indicate triggering by dynamic stress. Nature . 467, 583-‐586, 2010. 348
Sornette, D. and M.J. Werner (2005), Apparent clustering and apparent background 349
earthquakes biased by undetected seismicity, J. Geophys. Res. 110, B09303 350
doi:10.1029/2005JB003621. 351
25
Van Der Elst, N.J. and E. E.Brodsky (2010), Connecting near-‐field and far-‐field earthquake 352
triggering to dynamic strain. Journal of Geophysical Research 115, B07311, 2010. 353
West, M., J. J. Sanchez, and S. McNutt (2005), Periodically Triggered Seismicity at Mount 354
Wrangell, Alaska, After the Sumatra Earthquake, Science, 308, 1144-‐1146. 355
Zhuang, J., A. Christophersen, M. K. Savage, D. Vere-‐Jones, Y. Ogata, and D. D. Jackson (2008), 356
Differences between spontaneous and triggered earthquakes: Their influences on 357
foreshock probabilities, J. Geophys. Res., 113, B11302, doi:10.1029/2008JB005579. 358
359