ars.els-cdn.com€¦ · web viewin the horizontal direction, the limit of the 95% confidence...
TRANSCRIPT
![Page 1: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/1.jpg)
Supplementary Materials for
Earth’s deepest earthquake swarms track fluid ascent beneath nascent arc volcanoes
Lloyd T. White, Nicholas Rawlinson, Gordon S. Lister, Felix Waldhauser, Babak Hejrani, David A.
Thompson, Dominique Tanner, Colin G. Macpherson, Hrvoje Tkalčić, Jason P. Morgan,
Correspondence to: [email protected]
This file includes:
Supplementary Text
Supplementary Figures 1 to 11
Supplementary Tables 1 to 3
Captions for Supplementary Video 1 to 3
Captions for Supplementary Datasets 1 to 5
Captions for Interactive Map Data
Captions for Interactive 3D Model
1
1
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
![Page 2: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/2.jpg)
Supplementary Text
Visualisation of hypocenter data using eQuakes
The location (longitude, latitude and depth) of hypocenter data with magnitudes 1.0 < x <10.0,
and depths between 1 to 1000 km were visualised using eQuakes. The software was further used to
produce cross-sections of the different hypocenter catalogues perpendicular to the strike of oceanic
trenches as well as to visualize the hypocenter data at discrete time intervals (e.g. yearly, monthly,
daily). Multiple cross sections were generated along strike of the trench to interpret the 3D
geometry of the Wadati-Benioff Zone. The interpretation of the 3D slab geometry was cross-
checked with anomalous P-wave velocity structures in seismic tomographic data (Li et al., 2008).
Earthquake location using a fully non-linear grid search algorithm
For the Mariana sub-vertical pipe swarm, a total of 47,322 picks from 637 earthquakes are
used (Supp. Fig. 1a), whereas for the Izu-Bonin pipe swarm, 45,452 picks from 631 earthquakes are
used (Supp Fig 1b). The geographic distribution of picks is generally good (Supp. Fig. 2), but there is
a distinct lack of recording stations in the Pacific region to the east, which will have a detrimental
effect on our ability to accurately locate events. The vast majority of picks correspond to direct P-
and S-arrivals, but where available, other phases such as pP, sP, PcP and PcS are retained to
improve depth location accuracy.
The non-linear inverse problem that we solve is to find the location of each earthquake such
that the differences between the observed and predicted arrival times (through a prescribed Earth
model) are minimized. This is formalized by specifying an objective function S(m), where m is a
vector that defines the latitude, longitude and depth of the earthquake, such that:
2
2
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
![Page 3: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/3.jpg)
S (m )=(dobs−dm (m) )TC0−1 (dobs−dm (m ) ) (1)
where dobsis a set of N observed arrival times (from a given hypocenter) with the mean subtracted,
C0−1 is a data covariance or weighting matrix which accounts for data picking uncertainty, and
dm (m ) is a set of predicted traveltimes with the mean subtracted. By removing the mean, we can
eliminate the origin time as an unknown in the inversion, since source location can be constrained
by the relative on-set times of a phase at a set of stations.
We locate the minimum of S(m) using a nested grid-search, which is a fully non-linear
technique capable of locating the global minimum of S(m) and providing valuable information on
location uncertainty. The method works by computing S(m) at a regular grid of points in latitude,
longitude and depth centered on the ISC location. The grid with the minimum value of S(m) is then
chosen as the updated location, and the grid search is repeated using a finer grid spacing. The final
solution provided by this two-step procedure is then used as the new hypocenter. For both the
Mariana and Izu-Bonin sub-vertical pipe swarms, we use an initial grid spacing of 10 km in all three
dimensions, and a grid which extends 150 km in latitude and longitude and 200 km in the depth
direction from the reference ISC (2013) location. For the second stage, the refined grid has a
spacing of 2 km in all three dimensions and a grid which extends 30 km in latitude and longitude
and 40 km in the depth direction from the initial source location. We found that the final locations
were quite robust with respect to variations in the grid spacing and grid extent. To account for data
outliers, we remove arrival time picks which differ from the corresponding prediction by more than
two standard deviations.
Although a nested grid search is likely to be less computationally efficient than other direct
search methods such as simulated annealing, genetic algorithms and the neighborhood algorithm
(e.g., Billings et al., 1994a; 1994b; Sambridge and Kennett 2001), it does heavily sample model
3
3
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
![Page 4: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/4.jpg)
space, so information on hypocenter uncertainty can be achieved with little additional
computational effort. Here, we adopt the approach described by Sambridge and Kennett (1986) for
defining the 95% confidence region around the solution m¿
obtained from the nested grid search.
This is achieved by identifying the region of parameter space that satisfies:
T (m ,m¿ )=S (m )−S (m
¿ )S (m¿ )
≤χ32 (0.95 )
(N−3 )(2)
where N is the number of data picks and χ32 (0.95 )=7.815 is the Chi-square statistic for a p-value of
0.05 and three degrees of freedom. This approach is valid since the objective function of Equation 1,
with [C0−1 ]ij=δ ijσ ij where δ ij=1 when i=j and δ ij=0 otherwise, can be written as:
S (m )=∑i=1
N [dobsi −dmi (m ) ]2
σ i2
(3)
which has a chi-square distribution with N-M degrees of freedom, where M=3 is the number of
model parameters. In equation 3, σ i represents the picking uncertainty associated with diobs.
Equation 3 accounts for the effect of errors in the velocity model (which are usually unclear) by
rescaling S so that S (m¿ )=N−3 which is simply the expectation value of a chi-square distribution
with N-3 degrees of freedom. This is obtained by multiplication with the factor (N−3 ) /S (m¿ ). Both
the objective function (Equation 1) and method for computing confidence intervals assume that the
data errors are Gaussian, which will influence the final hypocenter locations and uncertainty
estimates. The validity of this assumption is tested below, when a comparison is made between
locations obtained using an L2 and L1 measure of misfit.
4
4
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
![Page 5: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/5.jpg)
The method described above is applied to the ISC arrival time dataset in the presence of the
ak135 global reference model (Kennett et al., 1995), which is the default standard model for global
earthquake location (Engdahl et al., 1998). The standard deviation of the picking error for P-phases
and S-phases are set to a uniform value of σ P=0.5 s and σ S=1.0 s which is towards the higher end
of what is considered the normal range for manual picks of teleseismic phases for a trained analyst
(e.g., Douglas et al., 1997; Leonard 2000).
Supp. Fig. 1 compares the relocated events for both the Mariana and the Izu-Bonin sub-vertical
pipe swarms with their respective ISC locations; in this case, the relocated events are colored
according to the limits of the 95% confidence region in depth, which is here defined as the average
distance between source ms=(rs , θs , ϕs ) (radius, latitude and longitude) and the two intersection
points of (r , θs, ϕs ) with the boundary of the 95% confidence region defined by T (m,ms ). In the
horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the
average distance between ms=(rs , θs , ϕs ) and the two intersection points of (r s ,θ , ϕs ) with the
boundary of the 95% confidence region and (2) the average distance between ms=(rs , θs , ϕs ) and
the two intersection points of (r s ,θs , ϕ) with the boundary of the 95% confidence region. In the
following tests, we use these measures as proxies for the vertical and horizontal uncertainty in
earthquake location.
5
5
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
![Page 6: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/6.jpg)
Depth distribution of well constrained events
In general, the more picks available for an earthquake, the more accurate the final location,
although data noise, phase type and geographic distribution of recording stations also plays a role.
Supp. Fig. 3 compares two subsets of located events for the Mariana and Izu-Bonin earthquake
pipes; the first (Supp. Fig. 3, top) only displays earthquake locations which were constrained by a
minimum of 25 picks, while the second (Supp. Fig. 3, bottom) only displays earthquake locations
with a depth uncertainty no larger than 10 km (Mariana) or 15 km (Izu-Bonin). The advantage of
using uncertainty estimates is that they account for the sensitivity of the location parameters to the
arrival time picks, and where redundancy occurs, will take into account inconsistencies in the data.
However, when both the statistical distribution of the noise and its standard deviation are largely
assumed, uncertainty estimates become less reliable. As such, it is useful to consider both sets of
results. For the Mariana and Izu-Bonin sub-vertical pipes, both subsets of events quite clearly
delineate the pipe, although it could be argued, in the case of the Mariana, that gaps are present
between ~120-140 km and ~160-220 km depth. If we gradually relax the depth uncertainly
criterion, these gaps become filled with events that are less accurately located. The common feature
of these events is that they are relatively low magnitude, which makes it unsurprising that they
have fewer high-quality picks and hence are more poorly constrained in depth. Nevertheless, with a
maximum depth uncertainty set to ~20 km, these “gaps” are largely filled, so whether they actually
exist or are zones in which only smaller earthquakes occur (which in itself would be worthy of
further investigation in the future) is unclear. It is worth noting, however, that the double difference
relocation method does not identify any gaps in the Mariana earthquake pipe (see Figure 2 of main
manuscript). These tests support the existence of the two pipe-like swarms of seismicity, since
removing the more poorly constrained events does not result in a loss of the sub-vertical pipe-like
geometry.
6
6
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
![Page 7: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/7.jpg)
Influence of pP and sP phases on focal depth
The arrival times of direct P and S phases can be used to obtain accurate epicenters, but
control on depth is generally poor (largely due to a trade-off between event depth and origin time).
Depth-sensitive phases such as pP and sP can therefore play an important role in estimating
earthquake depth. As was demonstrated earlier, the use of even a single differential arrival time
such as pP-P can provide a well-constrained estimate of depth, and the consistency between such
estimates at different stations can help improve confidence in the result. However, there are issues
surrounding the correct identification of pP, including the potential to misidentify sP and pwP (P-
wave reflection from the ocean surface instead of the ocean bottom) as pP. Waveform modelling
performed using the CMT solutions derived in this paper indicate that pwP occurs as a low
amplitude reverberation following the main pP phase, and are unlikely to be mis-identified. These
same tests also show that the amplitude of pP varies with azimuth, with maximum amplitudes
generally occurring to the north and south, and minimum amplitudes to the east and west, where
sP becomes dominant. Therefore, it is possible that pP phases identified in these regions are
actually sP, which would have the effect of increasing the estimate of focal depth. Our pP-P results
don't indicate that this is a major problem, due to the consistency in depth estimates from stations
at different azimuths. In addition, a comparison between hypocenter depths determined using our
single-station pP-P scheme and the non-linear location scheme with all available phases indicates
that there is little evidence of inconsistency between the two sets of results (Supp. Fig. 4). Another
independent test is to perform the non-linear location without pP and sP phases, and compare the
results with those illustrated in Supp. Fig. 1. As shown in Supp. Fig. 5A-B, there is little difference
between the two sets of results, although it is clear that a majority of locations do not have sP or pP
phase picks. However, if we compare the depths of only those events that have at least one sP or pP
pick, the mean difference in depth is only 20 km for the Mariana pipe and 18 km for the Izu-Bonin
pipe, which given the depth extent of the pipes is not significant.
7
7
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
![Page 8: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/8.jpg)
Exclusion of S-wave phases
If P and S-wave arrival times have similar uncertainty, then the S-phase will likely be more
sensitive to location due to S-wavespeeds being substantially lower than P-wavespeeds. In practice,
the higher levels of noise in S-wave arrival time datasets tends to nullify this advantage (see
Gomberg et al., 1990 for more discussion), but the increased data coverage they provide means that
their inclusion in the non-linear location scheme may provide improved results. To test the
influence of S-arrivals, we perform a test in which events are located using only the arrival times of
P-waves. A comparison between these locations and those of Supp. Fig. 1, in which both P and S
arrival times are exploited, is shown in Supp. Fig. 5C-D. For both sub-vertical pipes, the differences
in event distribution are negligible (see Supp. Table 1), with the exception of a small cluster of
events at approximately 400 km depth in the Mariana pipe, which appear in the inversion of P-
arrivals only, but not in the inversion of P and S. These are likely to be artifacts resulting from
insufficient constraints in the P-wave arrival time dataset.
Use of L2 misfit function
The use of an L2 norm to define the objective function (Equation 1 in main text), assumes that
the data errors have a Gaussian distribution, but in general, this may not be the case. In particular,
incorrect phase identification may result in large outliers, which can significantly influence
earthquake location if an L2 measure of misfit is used. The use of a fully non-linear location scheme
makes it relatively simple to test whether the L2 norm is suitable, via comparison with an L1
definition of the objective function which we know is robust with respect to outliers. For both pipes,
a comparison between the use of an L2 and L1 norm shows that the effect on the final earthquake
locations is minimal (see Supp. Fig. 5E-F and Supp. Table 1).
8
8
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
![Page 9: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/9.jpg)
Influence of 3D velocity structure
Apart from data picking uncertainties, another major source of location error comes from our
assumption that the Earth's velocity structure conforms to the ak135 global reference model
(Kennett et al., 1995). The principal variations in wavespeed in the Earth occur with depth, but it is
well known that lateral variations in velocity structure can affect the arrival times of earthquakes
significantly enough that they need to be taken into account during location (e.g. Rawlinson et al.,
2010).
In a study involving a global earthquake dataset of 450,000 events, Amaru (2007) compared
the results of locations made using the ak135 global reference model (Kennett et al., 1995) and a 3D
global velocity model obtained from body wave tomography. The results showed that while some
events were shifted in location in the 3D model by as much as 50 km with respect to location in
ak135 (Kennett et al., 1995), the vast majority experienced a location shift of less than 20 km, with
an average shift of just 12.8 km. The average location shift in depth was only 3.1 km, with the
largest shifts occurring at shallow depths, where crustal anomalies are relatively large compared to
those in the underlying mantle. If these shifts are roughly indicative of the scale of uncertainty
caused by using ak135 (Kennett et al., 1995) in the location of the Mariana and Izu-Bonin events,
then, like the location uncertainty caused by picking error, they are unlikely to change the seismic
delineation of the two pipes. In fact, any such shift in this case is likely to contain a dominant
component that is common to every event, given that they are clustered in a very similar location. A
DC shift of this kind is unlikely to impact the interpretation of the earthquake clusters at all, unless
it is extremely large. However, both earthquake pipes do lie west of a subduction zone, and the
faster velocities associated with a descending slab may have had the effect of applying an easterly
DC location shift to both clusters of events. A well-known example of location bias due to
9
9
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
![Page 10: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/10.jpg)
subduction was the Longshot explosion, which took place in the Aleutian islands in 1965. The
sampling of the Aleutian slab by nearly all recorded phases caused a 26 km location error (Herrin
and Taggart, 1968). However, even if the DC shift was this large (unlikely, since there is a better
azimuthal distribution of stations in our case), it is hard to see the interpretation changing, given
the scale of the features that are considered.
Depth phase analysis using USArray data
In order to independently identify depth phases for the Mariana pipe events, broadband
seismic data from USArray, which was deployed near the west coast of USA in 2006-2007, were
analyzed for the duration of the Mariana pipe seismic activity. This provided between 185 and 366
geographically similar stations covering a narrow epicentral distance range (75°-93°). Vertical
component data, acquired from the IRIS Data Management Centre, were initially cut 300s before
and after the predicted P-wave arrival time for all of the Mariana pipe events. Traces were then
tapered and filtered using a 2nd order two-way Butterworth bandpass filter with corner
frequencies of 0.5 Hz and 3 Hz. The data underwent a visual inspection to remove noisy traces and
only data with a clear P-wave arrival were retained for further processing. Initial alignment of
traces was achieved by making a moveout correction using the ak135 global model (Kennett et al.,
1995). Utilizing the similarity of the direct P-waveform across the array, trace alignment was then
refined using the adaptive stacking technique (Rawlinson and Kennett, 2004). Differential moveout
between the P and pP phases varies by less than 1s across the epicentral distance range of the
USArray data (Crotwell et al., 1999), meaning alignment along the direct P-wave and stacking
should also allow identification of the associated depth phases at the frequencies considered in this
study. However, we also align traces along time windows associated with pP using the same
adaptive stacking approach to increase coherency of this phase.
10
10
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
![Page 11: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/11.jpg)
Several of the pipe events exhibit pP-P times (7s to >30s) consistent with earthquake focal
depths within the mantle wedge (40 km to greater than 200 km) as evidenced by clear arrivals both
in the linear and quadratic stacks. These results are in agreement with depths reported by ISC as
well as the non-linear and double-difference methods utilized in this study (Supp. Table 2). We note
that the GCMT catalogue states that some of the events listed below occur at depths of 12–18 km.
However, other depth solutions were obtained following the CSCMT-3D solution (discussed above)
and these are more faithful to the various depths obtained for the equivalent hypocenters (Supp.
Data File 1; Supp. Table 2).
Calculation of potential fluid/melt ascent rate within the sub-arc mantle
We inferred that the rate of ascent of fluid/melt within the sub-arc mantle could be calculated
from each “burst” of seismicity that was identified in Supp. Figs. 9 and 10. A velocity value was
calculated for every pair of hypocenters where one was younger and shallower than another
hypocenter for each “burst” that was identified in Figure 6 of the main paper as well as in Supp.
Figs. 9 and 10. The ascent rate calculations generated numerous velocity values within each “burst”
and we report the median value for each “burst”. The velocity values and the corresponding
calculations are provided in Supplementary File 5.
Our intention of using this cumbersome calculation was to ensure we did not subconsciously
infer shallowing of hypocenters within each “burst”. However, we also considered whether the
same ascent rates could be calculated using depth and time values generated from random number
generators. First of all we created a test to determine if “bursts” of activity could be identified
between randomly generated depths of 0 and 250 km (Haarhr, 2019), as well as randomly
generated dates and times over a two year period (http://random-date-generator.com/). This test
was created to simulate the Mariana sub-arc earthquake pipe (i.e., similar depth and period of
activity). We repeated the test five times by regenerating the depth and time values (using the same
11
11
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
![Page 12: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/12.jpg)
limits). In doing so, we found no evidence of individual “bursts” (Supp. Fig. 12). We are therefore
confident that the bursts that we identified do not reflect a random process.
We conducted additional tests to determine if the ascent rate velocities that we obtained from
real-world hypocenter data were valid and how much these were influenced by different depth
ranges, time ranges and the number of nodes (i.e., hypocenters) used in the calculations. We found
that ascent rates could be calculated using number obtained from random number generators if we
restricted the depth range and time range when requesting randomly generated numbers. We also
found that the calculation is sensitive to the depth and time range used in the calculation (Supp. Fig.
13).
Finally, we calculated 20 pseudo-ascent rates using randomly generated depths between 0 km
and 200 km, a period of 10 days and using 20 nodes – these values were selected to simulate the
approximate length, duration and number of events within the Mariana ‘bursts’). We then
calculated the median ascent rate for each iteration and took an average of 19 out of 20 the results
and the standard deviation to obtain a statistically significant estimate of an ascent rate generated
from randomly generated data (see Supplementary File 5 for the data and calculations). We used
the same test, but restricted the depth range to values between 0 and 60 km to simulate the Izu-
Bonin pipe. The median velocity generated for the Izu-Bonin and Mariana pipes is compared with
the values obtained for each ‘burst’ in Supp. Fig. 14. This work demonstrates that the median
velocity obtained from the randomized data is lower than all velocities obtained for the Izu-Bonin
example, but is within error of Burst 2 and 3. For the Mariana example, similar velocities were
obtained from the randomized data and Bursts 1, 2 and 3. This analysis implies that there is
stronger statistical support for the faster ascent rate velocities (e.g., Izu-Bonin Burst 1 and 4, as well
as Mariana Burst 4).
These tests were a useful exercise, particularly for assessing the sensitivity of the ascent rate
velocity calculations with respect to different inputs. However, these results also need to be taken
12
12
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
![Page 13: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/13.jpg)
with some caution as the randomly generated numbers were generated with particular boundary
conditions (depth range, time range, number of nodes) and Supp. Figure 12 (and Supp. Data 5)
shows that the ‘bursts’ are unlikely to be generated using randomized data.
We expect that this investigation will prove to be of some use for future workers. Our overall
aim here was to develop an unbiased method to infer shallowing as well as to calculate a velocity of
transport. Ultimately, the best test we have at present is to compare our calculated ascent rates of
the HypoDD datasets with ascent rates obtained from independent data (as are discussed in the
main paper).
13
13
286
287
288
289
290
291
292
293
294
![Page 14: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/14.jpg)
Supp. Fig. 1.
Non-linear location results for (a) Mariana pipe; (b) Izu-Bonin pipe. Colored dots represent the non-
linear location results, which in this case are colored according to uncertainty in depth (a measure
14
14
295
296
297
298
![Page 15: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/15.jpg)
based on the boundary of the 95% confidence region). The size of the coloured dots is proportional
to the magnitude of the event (see scale range in top two panels). Small black dots represent the
earthquake locations determined by the ISC (2013). The Mariana example is defined by a total of
637 events, and the Izu-Bonin example is defined by a total of 631 events.
Supp. Fig. 2.
Locations of recording stations used in this study for (a) Mariana sub-vertical earthquake pipe; (b)
Izu-Bonin sub-vertical earthquake pipe. Stations are colored according to the number of arrival
time picks available.
15
15
299
300
301
302
303
304
305
306
307
308
309
310
311
312
![Page 16: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/16.jpg)
Supp. Fig. 3.
Results of two tests which only plots events when there are more than 25 picks available (top two
panels), or when the uncertainty in depth is less than 10 km (bottom left panel) or 15 km (bottom
right panel). The size of the coloured dots is proportional to the magnitude of the event (see scale
range in top two panels). Small black dots represent the earthquake locations determined by the
ISC (2013). The number of events in each plot is as follows (clockwise from top left): 201, 258, 197,
249.
16
16
313
314
315
316
317
318
319
320
321
![Page 17: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/17.jpg)
Supp. Fig. 4.
Comparison of depths between the non-linear location results (solid black line) and pP-P inversions
(red dots). The error bars are equal to one standard deviation of the spread of depth values when
more than one station has P and pP picks. When an event has P and pP picked at only one station,
no error bar is shown. In the case of the Mariana sub-vertical pipe (A) 51 events are compared,
whereas for the Izu-Bonin sub-vertical pipe (B), 88 events are compared. The grey area denotes the
95% confidence region associated with the non-linear location result (solid black line).
17
17
322
323
324
325
326
327
328
329
![Page 18: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/18.jpg)
Supp. Fig. 5.
Comparison of non-linear event locations that make use of all available data (red dots), versus (as
shown by small black dots) (a), (b) those that exclude pP and sP arrivals; (c), (d) those that exclude
S arrivals; and (e), (f) those that use an L1 measure of misfit. The size of the red dots is proportional
to the magnitude of the event (see scale range in top two panels). The Mariana sub-vertical pipe is
defined by a total of 637 events, and the Izu-Bonin sub-vertical pipe is defined by a total of 631
events in each of the three tests.
18
18
330
331
332
333
334
335
336
337
![Page 19: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/19.jpg)
Supp. Fig. 6.
Histogram of initial (ISC, grey) delay time residuals for Mariana (A) and Izu-Bonin (B) pipe events,
and residuals for the final DD locations (blue). (C-D) Histograms of lateral and vertical relative
location errors, computed from the major axes of the horizontal and vertical projection of the 90%
confidence ellipsoids obtained from a bootstrap analysis of the final double-difference vector based
on 200 samples with replacement. Phase travel time curves for the Mariana (E) and the Izu-Bonin
19
19
338
339
340
341
342
343
344
![Page 20: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/20.jpg)
(F) pipe events showing all available phase arrival time picks in the ISC bulletin (2013) (grey dots),
and phase arrival time picks observed pair wise at common stations and used to form delay time
data (blue dots) that went into the double-difference analysis.
Supp. Fig. 7.
Bathymetric maps (Ryan et al., 2009) that show published geochemical data (71) that have been
collected from the seafloor above and within the vicinity of the Izu-Bonin (top panels) and Mariana
(lower panels). The samples that directly overlie the two sub-vertical earthquake pipes are
primarily basalts and basaltic andesites according to their bulk rock compositions (middle panels).
The Izu-Bonin example also has several andesites and dacites above the pipe (middle panels). The
trace element data for these samples demonstrates typical volcanic arc signatures with respect to
N-MORB compositions (Sun and McDonough, 1989).
20
20
345
346
347
348
349
350
351
352
353
354
355
356
![Page 21: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/21.jpg)
Supp. Fig. 8.
The location of the Izu-Bonin and Mariana sub-arc earthquake pipe swarms compared with the
location of arc volcanoes and their recorded eruptive histories (http://volcano.si.edu). The
bathymetric maps are taken from Global Multi-resolution Topographic database (Ryan et al., 2009).
21
21
357
358
359
360
361
![Page 22: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/22.jpg)
Supp. Fig. 9.
A spatio-temporal analysis of seismic activity within the Izu-Bonin earthquake pipe demonstrates
that there are periods of increased seismic activity over one to several day periods that also occur
over a range of depths. Top panel: Time of occurrence plotted against depth of events. The
subfigures represent zoomed in views of the individual burst events as row 1: Depth vs. Time, row
2: Map View, row 3: Depth vs. Longitude (i.e. a west to east section) and row 4: Depth vs. Latitude
(i.e. a north-south section). The hypocenters are sized according to their magnitude and each burst
22
22
362
363
364
365
366
367
368
369
![Page 23: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/23.jpg)
is colored with respect to time. The grey hypocenters represent the total double-difference re-
located hypocenters.
Supp. Fig. 10.
A spatio-temporal analysis of seismic activity within the Mariana earthquake pipe demonstrates
that there are periods of increased seismic activity over one to several day periods that also occur
over a range of depths. Top panel: Time of occurrence plotted against depth of events. The
23
23
370
371
372
373
374
375
376
![Page 24: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/24.jpg)
subfigures represent zoomed in views of the individual burst events as row 1: Depth vs. Time, row
2: Map View, row 3: Depth vs. Longitude (i.e. a west to east section) and row 4: Depth vs. Latitude
(i.e. a north-south section). The hypocenters are sized according to their magnitude and each burst
is colored with respect to time. The grey hypocenters represent the total double-difference re-
located hypocenters.
24
24
377
378
379
380
381
![Page 25: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/25.jpg)
25
25
382
![Page 26: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/26.jpg)
Supp. Fig. 11. A summary of the depth (and 90% confidence intervals) compared with timing of the
hypocenters (relocated using HypoDD) that define ‘bursts’ within the broader (a-d) Mariana and (e-
h) Izu-Bonin sub-vertical earthquake pipe swarms. Note that the apparent difference in error bar
lengths between the Mariana and Izu-Bonin examples is simply due to different scale of the y-axis.
Supp. Fig. 12. An example of randomly generated time and depth values covering a two-year period
(the approximate duration of the Izu Bonin and Mariana earthquake pipes) and over a depth range
of 250 km (the approximate length of the Mariana pipe).
26
26
383
384
385
386
387
388
389
390
391
392
![Page 27: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/27.jpg)
Supp. Fig. 13. A series of box-and-whisker plots showing the variation in ascent rate velocities
obtained using different depth ranges and time ranges, as well as different number of ‘nodes’ (i.e.
hypocenters) used within the calculations. These calculations were performed in an attempt to
understand whether the ascent rates calculated using real-world hypocenter data (i.e., Mariana and
Izu Bonin ‘Bursts’ 1 to 4) actually reflected random values. The variation in median velocity values
demonstrates that the outcomes of the calculations are influenced by the depth range of the values
within an individual burst and the duration of activity within an individual burst.
27
27
393
394
395
396
397
398
399
400
401
![Page 28: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/28.jpg)
Supp. Fig. 14. A series of box-and-whisker plots showing the range, median and standard deviation
of velocities calculated for each ‘burst’ (see Supp. Data 5 for data and calculations). These are
compared with the average of 19 median velocities (black line) and their standard deviation (grey
box) obtained for randomly generated values used to simulate activity within (a) the Izu-Bonin pipe
(i.e., depth range of 60 km, duration of 10 days and with 20 nodes/hypocenters), and (b) the
Mariana pipe (i.e., depth range of 200 km, duration of 10 days and with 20 nodes/hypocenters).
Note that the grey box would shift if different boundary conditions were used to obtain the
randomly generated numbers (as per Supp. Fig. 12 and 13).
28
28
402
403
404
405
406
407
408
409
410
411
412
![Page 29: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/29.jpg)
Supp. Table 1.
Differences in hypocenter determination obtained by comparing the results of the non-linear
location scheme using all available phases with several other scenarios.
Mariana sub-vertical pipe Izu-Bonin sub-vertical pipe
Depth
(km)
Latitude
(km)
Longitude
(km)
Depth
(km)
Latitude
(km)
Longitude
(km)
Mean difference between ISC
and non-linear locations
20 6 12 42 14 17
Mean difference between non-
linear locations with and
without sP and pP phases
20 5 5 18 5 5
Mean difference between non-
linear locations with and
without S phases
10 3 6 8 2 3
Mean difference between non-
linear locations using L2 and
L1 measures of misfit
10 3 6 15 7 9
Supp. Table 2.
Comparison of the depth of the 20 earthquakes temporally and spatially associated with the
Mariana mantle-pipe earthquake swarm – this includes hypocenter depths as reported in the ISC
29
29
413
414
415
416
417
418
419
420
![Page 30: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/30.jpg)
bulletin (2013) and centroid moment tensors as reported in the GCMT (Ekström et al., 2012)
database, as well as the depths determined from various relocation techniques employed in this
study*. These are listed in order from deeper to shallower events (according to the HypoDD
solution) and color coded according to the level of confidence in the assigned depth (high: ≥4
solutions within 15 km depth; moderate: 3 solutions within 15 km depth; low: ≤2 solutions within
15 km depth).
ISC EVENT
Number
Depth (km) of earthquake hypocentre or moment tensor:
Confidence of
HypoDD*
depth
ISC
database
pP-P
arrival
time*
HypoDD* GCMT
catalogue
CSCMT-3D*
solution 1
(“shallow”)
CSCMT-
3D*
solution 2
(“deep”)
Indicative
depth from
adaptive
stacking of
depth
sensitive
phases
(USArray)
8956511 109.5 107.5 105.6 12 - 58–127 ~110 High
8956586 90.3 102.3 91.4 12 15–19 41–123 ~90 High
12662964 83.8 69.8 77.9 12 15–22 36–109 - High
30
30
421
422
423
424
425
426
427
![Page 31: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/31.jpg)
11705384 35.0 165 71.6 17.9 15–17 45–120 ~79 Moderate
11705445 65.2 49.2 70.4 12.6 - 44–121 - Moderate
12664872 65.2 67.2 61.6 12 15–20 33–114 ~65 High
12652669 51.8 51.8 49.6 12.2 - 48–130 - High
8956585 50.4 68.4 47.5 12 15–19 55–120 - High
12660440 42.0 48 44 12 15–19 57–106 - High
8823380 41.4 47.4 43.4 12 15–19 48–136 ~41 High
8767893 43.8 59.8 37.8 12 - 30–108 - Moderate
11705431 45.9 29.9 36 12 - 46–125 - High
12670232 28.9 - 32.1 12 15–19 42–115 - High –
although
uncertainty
with which
GSCMT-3D
solution fits
best
8956569 34.2 34.2 31.1 12 15–19 41–123 - High –
although
uncertainty
with which
GSCMT-3D
31
31
![Page 32: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/32.jpg)
solution fits
best
12661269 42.3 32.3 29.8 12 15–18 37–106 - High –
although
uncertainty
with which
GSCMT-3D
solution fits
best
12664883 37.3 - 29.8 12 - 31–111 - Moderate
9137769 26.9 - 24.3 12 - 43–127 - Low
10396345 19.7 21.7 20.9 12 15–17 42–125 25–30 High
11505273 20 18 16.1 12 - 36–106 - High
12652398 4.7 18 15.1 12 15–19 46–115 - High
32
32
428
429
430
![Page 33: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/33.jpg)
Supp. Table 3.
Summary of the results of the calculated ascent velocities for each “burst” that was identified in the
Izu-Bonin and Mariana sub-vertical earthquake pipe swarms.
Izu-Bonin sub-vertical
pipe
Mariana sub-vertical pipe
Median Velocity (km/hr) Median Velocity (km/hr)
Burst 1 33.98 1.36
Burst 2 0.47 0.69
Burst 3 0.78 1.17
Burst 4 3.11 7.27
Combined 1.95 1.27
33
33
431
432
433
434
435
436
437
![Page 34: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/34.jpg)
Supp. Video 1. (separate file)
Animation showing the temporal nature of the seismic events within the most clearly defined
earthquake pipe located beneath the Mariana arc/back-arc. This animation also shows that the
majority of hypocenters that define the sub-vertical, pipe-like structure occur within 2006 and
2007, primarily within several distinct month-long pulses of activity. The hypocenter data were
sourced from ISC (2013).
Supp. Video 2. (separate file)
Animation showing a series of cross-sections generated perpendicular to the plate boundary to
show that that the earthquake pipes are not part of the same structure as the down-going Pacific
Plate. This animation also shows regions of aseismicity along the length of the slab that may
indicate the location of slab tears. The hypocenter data were sourced from ISC (2013).
Supp. Video 3. (separate file)
Animation showing both the spatio-temporal nature of the Izu-Bonin and Mariana earthquake
pipes. These animations show the cumulative location of hypocenters across the region between
1970 and 2009. As time passes, the subducted slab geometry becomes more apparent in both cross-
sections, but in 1985-1986 and in 2006-2007 there are pulses of earthquake ruptures that are
clearly distinguished from the subducted Pacific slab. The hypocenter data were sourced from ISC
(2013).
34
34
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
![Page 35: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/35.jpg)
Supp. Data File 1. (separate file)
Earthquake location, arrival time and relocated data from the Mariana pipe that were used and
were generated as part of this project.
Supp. Data File 2. (separate file)
Earthquake location, arrival time and relocated data from the Izu-Bonin pipe that were used and
were generated as part of this project.
Supp. Data File 3. (separate file)
Selection of stacked events from the Mariana sub-vertical earthquake pipe as recorded by USArray.
In each plot, the top trace shows the quadratic stack, while the trace immediately below shows the
linear stack. All remaining traces are the traces from each station which went into the stack. Global
phase arrival time predictions based on the ISC bulletin (2013) are shown in red. Possible pP and
sP arrivals in the stack are highlighted in yellow and green respectively, although when there is no
second arrival marked, the highlighted phase could be pP or sP. In most cases, the observations are
broadly consistent with the predictions, which provides independent evidence that the earthquakes
are distributed between the subducted slab and the surface, and cannot all be clustered near the
surface. The title of each plot contains the following information: stack filename, latitude, longitude
and depth of event, origin date and origin time (based on ISC Bulletin (2013)).
35
35
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
![Page 36: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/36.jpg)
Supp. Data File 4. (separate file)
Summary figures showing the CSCMT-3D solutions for 20 earthquakes within the mantle wedge
pipe. These show contour plots of correlation between the observed and synthetic waveforms in
the depth versus time, the double-couple percentage for each moment tensor, the broadband
stations that were used for the calculation and the fit between the observed waveforms (thick gray
line), CSCMT-3D solution synthetics (black line) and the GCMT solution synthetics (red line). The
numbers above and below the waveforms show the correlation between the observed and CSCMT-
3D synthetics (black text) and the correlation between the observed and GCMT synthetic
waveforms (red text). A collection of depth vs. correlation plots are also included to show where
one or two depth solutions were possible (green lines indicate values within 5% of the maximum
correlation value).
Supp. Data File 5. (separate file)
Ascent rate calculations derived from analyses of the double-difference relocated hypocenters
generated as part of this study.
Caption for Interactive Map Data (separate file)
KMZ file showing the location of earthquake hypocenters (ISC) that define the Mariana and Izu-
Bonin mantle earthquake pipes. Note that these are projected to the surface, but depth information
and other metadata is captured for individual points.
36
36
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
![Page 37: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/37.jpg)
Caption for Interactive 3D Model (separate file)
An interactive 3D PDF was built using the ISC hypocenter locations that define the Mariana sub-arc
mantle earthquake pipe. This file shows the location of the earthquake hypocenters that define the
pipe-like structure with respect to the subducted Pacific Plate and the overriding Philippine Sea
Plate.
37
37
502
503
504
505
506
507
508
![Page 38: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/38.jpg)
References
Amaru, M.L. 2007. Global travel time tomography with 3-D reference models”, PhD thesis, Utrecht
University, Utrecht, Netherlands
Billings, S.D. 1994. Simulated annealing for earthquake location, Geophysical Journal International
118, 680-692
Billings, S.D., Kennett, B.L.N., Sambridge, M.S. 1994. Hypocentre location: genetic algorithms
incorporating problem-specific information, Geophysical Journal International 118, 693-706
Crotwell, H.P., Owens, T.J., Ritsema, J. 1999. The TauP toolkit: flexible seismic travel-time and ray-
path utilities, Seismological Research Letters 70, 154-160
Douglas, A., Bowers, D., Young, J.B. 1997. On the onset of P seismograms, Geophysical Journal
International 129, 681-690
Ekström, G., Nettles, M., Dziewonski, A.M. 2012. The global CMT project 2004-2010: Centroid-
moment tensors for 13,017 earthquakes, Physics of the Earth and Planetary Interiors 200-201,
1-9, doi:10.1016/j.pepi.2012.04.002
Gomberg, J.S., Shedlock, K.M., Roecker, S.W. 1990. The effect of S-wave arrival times on the accuracy
of hypocenter estimation, Bulletin of the seismological Society of America 80, 1605-1628
Haahr, M. 2019. RANDOM.ORG: True Random Number Service. Available at:
https://www.random.org [last accessed 23 May 2019]
Herrin, E., Taggart, J. 1968. Source bias in epicenter determinations. Bulletin of the Seismological
Society of America 88, 722-743
International Seismological Center (ISC). 2013. On-line Bulletin, International Seismological Center,
Thatcham, United Kingdom, http://www.isc.ac.uk
Kennett, B.L.N., Engdahl, E.R., Buland, R. 1995. Constraints on seismic velocities in the Earth from
traveltimes. Geophysical Journal International 122, 108-124
38
38
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
![Page 39: ars.els-cdn.com€¦ · Web viewIn the horizontal direction, the limit of the 95% confidence region is calculated as the mean of (1) the average distance between m s = r s , θ](https://reader033.vdocuments.net/reader033/viewer/2022050604/5fabfb951914575b2c14f4c1/html5/thumbnails/39.jpg)
Leonard, M. 2000. Comparison of manual and automatic onset time picking, Bulletin of the
Seismological Society of America 90, 1384-1390
Li, C., van der Hilst, R. D., Engdahl, E.R., Burdick, S. 2008. A new global model for P wave speed
variations in Earth’s mantle. Geochemistry Geophysics Geosystems 9, Q05018, doi:
10.1029/2007GC001806
Rawlinson, N., Kennett, B.L. 2004. Rapid estimation of relative and absolute delay times across a
network by adaptive stacking. Geophysical Journal International 157, 332-340
Rawlinson, N., Pozgay, S., Fishwick, S. 2010. Seismic tomography: a window into deep Earth, Physics
of the Earth and Planetary Interiors 178, 101-135
Ryan, W.B.F., Carbotte, S.M., Coplan, J.O., O’Hara, S.O., Melkonian, A., Arko, R., Weissel, R.A., Ferrini,
V., Goodwillie, A., Nitsche, F., Bonczkowski, J., Zemsky, R. 2009. Global multi-resolution
topography synthesis. Geochemistry, Geophysics, Geosystems 10 doi:10.1029/2008GC002332
Sambridge, M.S. Kennett, B.L.N. 1986. A novel method of hypocenter location, Geophys. J. R. Astr. Soc.
87, 679-697
Sambridge, M.S., Kennett, B.L.N. 2001. Seismic event location: nonlinear inversion using a
neighbourhood algorithm, Pure and Applied Geophysics 158, 241-257
Sun, S.-s, McDonough, W.F. 1989. Chemical and isotopic systematics of oceanic basalts: implications
for mantle composition and processes in Magmatism in the Ocean Basins, A. D. Saunders, M. J.
Norry, Eds., (Special Publication 42, Geological Society London), 313-345
39
39
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553