Date post: | 07-Apr-2019 |
Category: |
Documents |
Upload: | phungxuyen |
View: | 214 times |
Download: | 0 times |
Supplementary information
“Seismic Evidence for the Depression of the D” Discontinuity beneath the Caribbean:
Implication for Slab Heating from the Earth’s Core”
Justin Yen-Ting Koa, b, Shu-Huei Hunga,*, Ban-Yuan Kuoc, and Li Zhaoc
This file includes:
1. Data Processing
2. Data Measurement and Correction
3. Synthetic Experiments on Data Sensitivity to Model Variables
4. Synthetic Example to Illustrate the Waveform Modeling Procedure
5. Data Misfits and Lateral Variation of the Resulting 1-D Models
6. Modelled 1-D Structure for Waveform with no SdS Arrival
7. Detectability of the D” Discontinuity
8. Large SdS Amplitudes Observed in the Central North American Continent
Supplementary References
Supplementary Table 1
Supplementary Figures S1-S11
1. Data Processing
For each earthquake we align all the selected waveforms along the SH phase
arrivals by an adaptive stacking method [Rawlinson and Kennett, 2004]. This process
would correct the arrival time moveout with epicentral distances and suppress as
much of the traveltime perturbation as possible resulting from local velocity
1
2
3
4
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
heterogeneity beneath individual stations and difference in station elevations. Finally,
to remove the difference in waveform shapes from the variability of finite-source
processes, we determine a common source wavelet for each event by stacking the
aligned S arrivals at the stations at 70 degrees or less which have little contribution of
seismic energy from later triplication arrivals. An iterative time-domain
deconvolution with a Gaussian width factor of 0.4 (equivalent to a low-pass filter with
the cutoff frequency of 0.2 Hz) [Ligorría and Ammon, 1999] is then applied to
removal of the obtained source wavelet from individual aligned traces. An example of
data processing is summarized in Figure S1.
2. Data Measurement and Correction
As the differential traveltime and relative amplitude measurements are less
influenced by event mislocation and upper mantle heterogeneity and provide strong
constraints on lowermost mantle structure (Fig. 1c), we construct our dataset that
comprises the observed differential ScS-S and SdS-S traveltimes (∆ T ScS−S and ∆ T SdS−S
) and ScS/S and SdS/S amplitude ratios (∆ AScS /S and ∆ ASdS/S). Overall waveform
similarities between observed and synthetic traces in the modeled time window are
also included as part of the measures of optimization criteria in the search of the best-
fit model. Prior to data modeling, the measured differential traveltimes are corrected
for contributions from mantle velocity heterogeneity based on a global tomography
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
model [Ritsema et al., 2002].
Additionally, certain types of focal mechanisms for available events may be
unfavorable for excitation of shear wave energy toward our stations which have a very
limited distance and azimuthal coverage. It can cause fluctuations in waveform
amplitudes between nearby stations and introduce large uncertainties in modeling the
changes of amplitude ratios and overall waveform shapes genuinely associated with
the D” structure. We thus exclude those data observed at station azimuths that would
yield large fluctuations of amplitude ratios estimated based on theoretically-predicted
radiation patterns [Lay and Wallance, 1995],
𝑅𝑆𝐻=𝑐𝑜𝑠𝜆𝑐𝑜𝑠𝛿𝑐𝑜𝑠𝑖ℎ𝑠𝑖𝑛𝜙+𝑐𝑜𝑠𝜆𝑠𝑖𝑛𝛿𝑠𝑖𝑛𝑖ℎ𝑐𝑜𝑠2𝜙+𝑠𝑖𝑛𝜆𝑐𝑜𝑠2𝛿𝑐𝑜𝑠𝑖ℎ𝑐𝑜𝑠𝜙−1
2 𝑠𝑖𝑛𝜆𝑠𝑖𝑛2𝛿𝑠𝑖𝑛𝑖ℎ𝑠𝑖𝑛2𝜙, (S.1)
where 𝜙=𝜙𝑓−𝜙𝑠, is the difference between the strike azimuth of the fault plane, 𝜙𝑓,
and the source-to-station azimuth, 𝜙𝑠, 𝛿 is the dip of the fault plane, 𝜆 is the slip rake
angle, and 𝑖ℎ is the takeoff angle for a given shear wave. We select 7 representative
focal mechanisms and calculate the theoretical amplitude ratios at the epicentral
distances from 75o to 80o and azimuthal range between 310o and 360o most available
in our data (Fig. S2). Only the stations with the predicted values that fall within the
standard deviation of all the calculated amplitude ratios for each event are used for the
following data measurement and grid search modeling.
Not only does 3D structural heterogeneity have a significant influence on
observed amplitude anomalies, but different radiation patterns of the earthquakes can
also produce intrinsic fluctuations in amplitude for seismic phases arriving at different
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
azimuths and takeoff angles. We thus correct the source-related amplitude-ratio
deviations from the raw measurements using eq. (S.1). It is worth mentioning that
amplitude information can provide critical constraints particularly on the sharpness of
the seismic discontinuity superior to differential traveltime data but they should be
treated with caution owing to numerous factors as discussed here which can
substantially alter the amplitudes of arriving phases. Further detailed investigation of
3D wave propagation effect on amplitude fluctuations is necessary for future studies.
3. Synthetic Experiments on Data Sensitivity to Model Variables
Rather than using numerous free variables to parameterize a radial velocity
model for the D” layer and fit every detail in the observed data with suspected
resolvability, we aim primarily at investigating the overall characteristics of the
discontinuity undulation and shear velocity structure of the underlying D” layer linked
to the “slab graveyard” feature as inferred from seismic tomography [Grand et al.,
1997]. We choose three model parameters including and impedance contrast (VD”) on
the D” discontinuity and radial velocity gradient (GD”) within the D” layer to depict
the first-order feature of D” in our study region.
It is well noted that there exist severe trade-offs between the constraining
discontinuity topography and neighboring velocity structure particularly relying on
differential traveltime observations. As each type of the datasets provide different
effective levels of constraints on a finite number of model variables used to
characterize the D” structure, we first construct a large set of ground-truth synthetic
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
seismograms for a thorough combination of different model parameters computed
with the DSM [Kawai et al., 2006] to elucidate the sensitivity of the observed
waveforms and measured data to the perturbation of each model parameter which
guides for the following search of optimal D” models.
Figs. S3-S5 show the testing ranges of the perturbations for individual model
parameters and corresponding shear wave velocity variations with depth in the
lowermost 600 km mantle. The figures also exemplify the comparison of DSM-
computed synthetic SH triplication waveforms and predicted differential traveltimes (
∆ T ScS−S and ∆ T SdS−S) and amplitude ratios (∆ AScS /S and ∆ ASdS/S) between the testing
models at epicentral distances of 70 and 75 degrees. In Fig. S3, starting with the
PREM model, we fix the thickness of the D” layer to be 264 km and vary the velocity
jump across the D” discontinuity from 1% to 3%. It is clear that the SdS arrival times
are barely changed but their amplitudes increase significantly with the magnitude of
the velocity contrast. On the other hand, both the ScS-S times and ScS/S amplitudes
decrease gradually with the velocity contrast. In Fig. S4, we instead fix the velocity
contrast of 3% across the D” discontinuity and vary the velocity gradient from -6% to
+3% within a 264-km thick D” layer. Similar to those shown in Fig. S3, the decrease
of the velocity gradient in the D” layer has a negligible influence on the SdS arrival
times but substantially reduces the ScS-S differential times. Moreover, it has distance-
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
dependent but reverse influence on SdS/S and ScS/S amplitude ratios being enlarged
and reduced, respectively. In Fig. S5, we vary the thickness of the D” layer from 150
km to 350 km and implement a fixed velocity jump of +3% across the discontinuity in
PREM model with the uniform velocity in the D” layer. The results show the SdS-S
differential times are most sensitive to the topographic variation of the D”
discontinuity.
According to the above synthetic experiments, we find that all three model
parameters exert strong effects on amplitude ratios, ∆ ASdS/S but differential
traveltimes, ∆ T SdS−S, are only affected by the thickness of the D” layer. That is, we
are capable of mapping the topography of D” discontinuity based on ∆ T SdS−S alone.
Furthermore, for the ScS related data measurements, only the change of the velocity
near the CMB can induce noticeable variations in differential traveltimes and
amplitude ratios, providing a critical constraint on the gradient within the D” layer.
See Table S1 for a brief summary of the extent of influence of each model variable on
observed differential traveltime and amplitude ratio data.
4. Synthetic Example to Illustrate the Waveform Modeling Procedure
In Fig. S6, we present a synthetic example to illustrate our modeling procedure.
First, we construct DSM-computed, SH-component seismograms for a presumed 1-D
model in D” using the same source-receiver geometry as that of real data (right panel
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
of Fig. S6(a)). Second, various levels of Gaussian random noise are added to mimic
the observed waveforms with signal-to-noise ratios (SNR) on the order of 16 and 5.
The procedure applied to real data measures is then employed to generate synthetic
datasets which comprise differential traveltimes, amplitude ratios and waveform
decorrelation coefficients measured directly from the simulated waveforms. Finally,
we determine the topographic height of the D” discontinuity based on the measured
SdS-S differential times and then search for the optimal solutions of the other two
model variables, VD” and GD”, by minimizing the defined cost function.
Fig. S6(b) illustrates how the changes of VD” and GD” would affect the fitting of
different types or combinations of data measures and how severe the tradeoff between
them would be provided that only specific datasets are included as constraints. It also
demonstrates that the model solution, if considering only the phase information, i.e.,
differential traveltime and waveform decorrelation, in the cost function, is possibly
trapped into multiple local minima, especially for the low S/N data. Further including
the amplitude-ratio misfits in the cost function, on the other hand, can largely
eliminate the tradeoff between VD” and GD” and drive the search toward the global
minimum of the model space.
5. Data Misfits and Lateral Variation of the Resulting 1-D Models
In Fig. S7, we plot a histogram showing the distribution of the total misfit errors
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
and proportions of three misfit terms in the defined cost function as a function of the
topographic height of the D” discontinuity (HD”) for all the resulting 1-D models. As
the 3D structural variations have more pronounced effects on waveform amplitudes,
the amplitude-ratio data account for a larger proportion of the errors of about 50%.
In Fig. S8(a), we show the lateral variations in the topographic height of the D”
discontinuity (HD”) and shear velocity perturbation (lnVs) at 2800 km depth from the
individual, unaveraged 1-D models. They in general display similar E-W variations as
those shown in Fig. 3(b), which have the lowest topographic relief and strongest shear
velocity reduction in the central part of our study region. Fig. S8(b) displays the
standard deviations HD” and lnVs within each cap, while Fig. 8(c) shows the cap size
used for lateral averaging of HD” and lnVs shown in Fig. 3(b). The larger standard
deviations may roughly indicate the abrupt, unphysical structural variations between
the nearby 1-D models within the cap resulting from the unmodelled 3D effects. They
are usually much smaller in the densely-sampled region compared to the variational
ranges of HD” (over 150 km) and lnVs (~5%) for the large-scale structure over 600
km along the E-W direction.
6. Modelled 1-D Structure for Waveform with no SdS Arrival
The synthetic experiments shown above indicate the differential SdS-S times
provide robust constraints on HD”. However, many shear wave traces particularly
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
traversing D” in the south-central part of our study region as shown in Fig. 3(b) do not
have noticeable SdS arrivals for HD” estimates (Fig. S9). We thus assume 150 km for
HD” in the waveform modeling, which essentially results in a 1-D structure with very
small and negligible velocity contrast on the D” discontinuity (Fig. S9)
7. Detectability of the D” Discontinuity
In Fig. S10, we show 1-D DSM synthetics for the D” models with a 0.2 km/s
shear velocity increase within a transition zone whose thickness (W) varies from 0 to
120 km near 300 km above the CMB. The results show the SdS amplitudes decrease
with increasing W more drastically at shorter distances but still are visible at >78o
distances for W=120 km. The peak arrival times of SdS relative to S used to estimate
HD” are almost unchanged.
8. Large SdS Amplitudes Observed in the Central North American Continent
Fig. S11 shows the anomalously large SdS amplitudes recorded at the stations
located in the central part of the USArray and the distance of ~80o away from the
event. These triplication arrivals traverse the middle of our study D” region with the
depressed D” topography and large negative shear velocity gradients in the D” layer
as constrained by the observed shear wave data. The resulting models properly predict
the observed ScS/S amplitude ratios but apparently underestimate the SdS/S amplitude
ratios.
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
Supplementary Reference
Grand, S.P., van der Hilst, R.D., Widiyantoro, S., 1997. Global seismic tomography: a
snapshot of convection in the Earth, GSA Today 7(4), 1–7.
Kawai, K., Takeuchi, N., Geller, R.J., 2006. Complete synthetic seismograms up to 2
Hz for transversely isotropic spherically symmetric media, Geophys. J. Int. 164,
411-424.
Lay, T., Wallace, T.C., 1995. Modern Global Seismology, Academic Press, San Diego,
521.
Ligorría, J.P., Ammon, C.J., 1999. Iterative deconvolution and receiver function
estimation, Bull. Seismol. Soc. Am. 89, 1395-1400.
Rawlinson, N., Kennett, B.L.N, 2004. Rapid estimation of relative and absolute delay
times across a network by adaptive stacking, Geophys. J. Int. 157: 332–340.
Ritsema, J., Deuss, A., van Heijst, H.J., Woodhouse, J.H., 2011. S40RTS: a degree-40
shear-velocity model for the mantle from new Rayleigh wave dispersion,
teleseismic traveltime and normal-mode splitting function measurements.
Geophys. J. Int. 184, 1223–1236.
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
Supplementary Table
Table S1. A brief summary of the results of data sensitivity tests.
Data Increase in Impedance Contrast (VD”)
Increase in Velocity Gradient (GD”)
Increase in D” Thickness (H)
SdS/S amp increase increase increaseSdS-S time unchanged unchanged decreaseScS/S amp decrease decrease unchangedScS-S time decrease decrease unchanged
200
201
202
Supplementary Figures
Figure S1. Illustration of processing of triplication shear waves recorded by the
USArray. (a) SH-component displacement traces first aligned along predicted S
arrival times based on PREM. (b) Further alignment of the S phase arrivals through
10 iterations of the adaptive stacking procedure [Rawlinson and Kennett, 2004]. The
top red trace is considered as a reference wavelet or source wavelet obtained from the
linear stack of the waveforms recorded at distance less than 70o. (c) Impulsive-like S-
wave signals after the source wavelet being removed from the traces in (b) through an
iterative deconvolution [Ligorría and Ammon, 1999]. For the following data measure
and waveform modeling, we restrict our attention only to S and ScS phases and
triplication arrivals in between them at the distance range of about 65o-85o (blue
traces), where these waves all traverse the D" layer within the region of our primary
interest.
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
Fig. S2. The investigation of the influence of earthquake radiation patterns on
relative waveform amplitudes. Theoretical amplitude ratios of ScS/S (top) and SdS/S
(bottom) plotted as a function of azimuth at the epicentral distances of 75o and 80o for
seven earthquakes used in our study. The corresponding double-couple focal
mechanisms are shown in the upper-right corner of each panel. The dashed lines
indicate the azimuths at which the predicted amplitude ratios are larger than the
standard deviation of those at the azimuths between 310o and 360o and excluded from
the cost function estimation in the waveform modeling.
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
Fig. S3. Data sensitivity to the velocity contrast (VD”) across the D” discontinuity.
Synthetic experiments illustrating how the sudden velocity increase across the D”
discontinuity affects shear wave triplication waveforms and measured differential
ScS-S and SdS-S traveltimes (∆ T ScS−S and ∆ T SdS−S) and ScS/S and SdS/S amplitude
ratios (∆ AScS /S and ∆ ASdS/S), shown on the middle and right of the figure,
respectively. The synthetic SH waveforms are calculated by the DSM [Kawai et al.,
2006] in PREM and five trial models with VD” ranging from +1 to +3% shown on the
left.
242
243
244
245
246
247
248
249
250
251
252
Fig. S4. Data sensitivity to the shear velocity gradient (GD”) in the D” layer.
Synthetic experiments illustrating how GD” influences shear wave triplication
waveforms and measured differential ScS-S and SdS-S traveltimes (∆ T ScS−S and
∆ T SdS−S) and ScS/S and SdS/S amplitude ratios (∆ AScS /S and ∆ ASdS/S) in PREM and
six trial models with GD” varying from -6% to +3%. The figure layout remains the
same as Fig. S3.
253
254
255
256
257
258
259
260
261
262
263
Fig. S5. Data sensitivity to the topographic height of (HD”) of the D”
discontinuity. Synthetic experiments illustrating how HD” or the thickness of the D”
layer influences shear wave triplication waveforms and measured differential ScS-S
and SdS-S traveltimes (∆ T ScS−S and ∆ T SdS−S) and ScS/S and SdS/S amplitude ratios (
∆ AScS /S and ∆ ASdS/S) in PREM and five trial models with HD” varying from 150 km to
350 km above the CMB. The figure layout remains the same as Fig. S3.
264
265
266
267
268
269
270
271
272
273
274
Fig. S6. A synthetic test to illustrate and verify our waveform modeling
275
276
277
278
procedure. (a) (Left) Presumed 1-D model with a 260-km thick D” layer (red line)
used to generate synthetic shear waveforms shown on the right. Black lines represent
trial models in grid search. (Right) The top trace shows a synthetic SH waveform at
75∘for the presumed model computed by the DSM [Kawai et al., 2006]. The middle
and bottom traces simulate the observed waveforms by adding Gaussian noise to the
top trace with high and low signal-to-noise ratios (SNR) of 16 and 5, respectively.
The numbers next to the three synthetics are the thickness of the D” layer determined
from the observed SdS-S times marked by the dashed vertical lines. (b) Image maps
with contours showing the cost function estimated over a wide range of model
variables, VD” (velocity contrast on the D” discontinuity) and GD” (velocity gradient in
D”). Each column shows the grid search results constrained by the combined dataset
of differential traveltimes and waveform decorrelation coefficients (left), amplitude-
ratio data only (middle), and all of them simultaneously (right). The rows from top to
bottom correspond to the results obtained with noise-free, high and low SNR
waveforms shown in (a). The presumed “true” model and optimal solutions of VD” and
GD” at the global minima of the cost function are marked by magenta circles and
yellow stars, respectively.
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
Fig. S7. Histogram showing the total misfit error and proportion of the errors from
each type of data varying with the determined topographic height (HD”) of the D”
discontinuity for all the resulting 1-D models. There is no obvious correlation for the
depressed topography which yields a larger total misfit error and amplitude error.
298
299
300
301
302
303
Fig S8. (a) Relative shear velocity perturbations (δlnVs) with their mean removed at
2800 km depth (top) and topographic heights (HD”) of the D” discontinuity (bottom)
plotted at the ScS bounce points, obtained from the individual resulting 1-D models.
(b) The standard deviations of the δlnVs and HD” for the resulting 1-D models within
each cap. (c) The cap size used for lateral averaging of δlnVs and HD”.
304
305
306
307
308
309
310
311
Figure S9. Example of a shear waveform showing no noticeable SdS arrival and
the resulting best-fit 1D structure. For the trace with no available Scd-S differential
time to estimate the topographic height of the D” discontinuity, we assume 150 km in
grid search modeling of the 1-D velocity structure. The resulting best-fit model yields
a negligible impedance contrast on the presumed D” discontinuity.
312
313
314
315
316
317
318
319
Fig. S10. 1-D DSM synthetics for the D” discontinuity occurring over a depth
interval (W) from 0 to 120 km. (Left) 1-D shear wave velocity models in D” used in
the calculation of synthetic waveforms. (Middle) The resulting triplication shear
waveforms plotted as a function of distance between 66-87o. (Right) The differential
SdS-S and ScS-S times measured from the synthetic records. Color dots shown on the
left side of the plot indicate the minimum distance for each W at which the SdS starts
to be detectable.
320
321
322
323
324
325
326
327
328
329
330
331
332
Fig. S11. Abnormally large amplitudes of SdS phases observed in the central
North American continent. The optimal 1-D models for individual station-event
pairs indicate that the densely sampled region beneath northern South America has a
thinner D” layer with relatively slow shear velocities. The observed SdS amplitudes
are clearly underpredicted by the synthetics, particularly at the distance of ~80o.
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350