Download Supporting Online Material for

www.sciencemag.org/cgi/content/full/311/5758/198/DC1
Supporting Online Material for
The Nature of the 660-Kilometer Discontinuity in Earth’s Mantle
from Global Seismic Observations of PP Precursors
Arwen Deuss,* Simon A. T. Redfern, Kit Chambers, John H. Woodhouse
*To whom correspondence should be addressed; E-mail: [email protected]
Published 13 January 2006, Science 311, 198 (2006)
DOI: 10.1126/science.1120020
This PDF file includes:
SOM Text
Figs. S1 to S7
References
The nature of the 660 km discontinuity in Earth’s
mantle from global seismic observations of PP
precursors
Arwen Deuss, Simon A. T. Redfern, Kit Chambers, John H. Woodhouse
Supporting online material
Data selection and processing
We have used a large data set of SS and PP precursors to study the characteristics of the 660
km discontinuity. Fig. S1 shows the ray paths for the PP wave and its precursor PdP. PP is
a compressional wave that travels the mantle and reflects once at the surface halfway between
the source and receiver. The precursor PdP does not travel all the way up to the surface at
the reflection point, but reflects at a discontinuity at depth d below the surface. Therefore, the
precursors arrive before the major PP wave, and with much smaller amplitude. SS and SdS are
compressional waves with similar ray paths.
bouncepoint
PdP
PP
source
receiver
Figure S1. Ray paths of PP (reflected at the surface) and the precursors PdP (here reflected at d=660
km discontinuity). SS and the corresponding SdS precursors have similar ray paths.
1
We collected data from the Incorporated Research Institutions for Seismology (IRIS) and
International Deployment of Accelerometers (IDA) global networks, including 2425 events for
the PP and 1625 events for the SS data set. The data were initially filtered between 15 and 75
seconds to remove short period noise and make all data sensitive to the same structure. We
picked the PP and SS arrival times using cross-correlation with a reference pulse, resulting in
9460 PP seismograms and 8054 SS seismograms. The PP and SS seismograms have epicentral
distance range of 80◦ -140◦ and 100◦ -160◦ , respectively.
Fig. S2 shows the bouncepoints of the selected seismograms. The bouncepoint is halfway
between the source and receiver, marking the location where PP (or SS) reflects at the surface.
This is also the location where the precursors have sensitivity to the discontinuities below the
surface. It is interesting to note that for the PP precursors, the regions where we see reflections
from the 660 km discontinuity (Fig. 3A) correlate with regions where we have a high data
coverage, including the Mid Pacific and the Indian subcontinent.
A
B
PP precursors
SS precursors
Figure S2. Bouncepoints of (A) the PP and (B) SS precursor data sets.
The selected seismograms were normalised by the amplitude of PP (or SS), making the ratio
of P660P/PP (or S660S/SS) amplitude a measure of the jump in seismic velocity and density at
the discontinuity. The precursors have much smaller amplitude than the surface reflected PP (or
SS) waves and cannot be seen in a single seismogram. Thus we stack the seismograms in the
slowness-time domain to suppress incoherent noise and make the precursors visible. We use a
linear stacking technique which does not distort the amplitude of the peaks in the stacks. These
stacks are then converted into a trace in which the stacking slowness is time-dependent, the timedependence being chosen to maximise the stacked amplitude corresponding to reflections from
a continuous range of depths. This corresponds to following a curved trajectory in the more
2
typical time-slowness plots. The resulting stacks amplify and clarify the peaks in amplitude
associated with the transition zone discontinuities at 410, 520 and 660 km depths.
When looking for weak reflections from the 660 km discontinuity, it is important not to
misinterpret noise as reflections from the 660 km discontinuity. Therefore we also applied a
bootstrap resampling technique to estimate the error in the stacks and obtain the 95% confidence
levels (1) for each stack. The 95% confidence levels are given by taking twice the standard error
of the stack and we only interpret a peak in the stack for which the lower confidence level is
still significantly larger than zero. In addition, we employed a deconvolution technique (2)
and we only interpret precursors picked up by the deconvolution. These are called ‘robust’
observations. Fig. 1 shows examples of stacks with their corresponding 95% confidence levels
with dashed lines. The grey coloured areas denote robust peaks for which the 95% confidence
is above zero. The regions used to generate the stacks in Fig. 1 are given in Fig. S3.
90˚
60˚
N America
30˚
0˚
MW
Pacific
India
-30˚
180˚
240˚
300˚
0˚
60˚
120˚
180˚
Figure S3. Regions used to generate the stacks in Fig.1.
We also put constraints on areas where P660P cannot be observed, so called ‘null observations’. This is an important issue, as we interpret our null observations using mineral physical
models. However, it is difficult to determine if a null observation is caused by the absence of
a sharp discontinuity at 660 km depth (i.e. a robust null observation), or because there is not
enough data available to make the 660 km discontinuity visible (not a robust null observation).
We applied three criterions in finding null observations (i) a minimum of 50 traces were stacked
in order to have enough data available, (ii) the 410 km discontinuity needs to be clearly visi-
3
ble and (iii) the 660 km discontinuity has to be below the noise level of the stack. These null
observations are shown as open circles in Fig. 3A.
Robustness of observations
In particular for the PP precursors, there is a possible interference with other major phases such
as Pp410p, Pp660p (near receiver topside reflections of the 410 and 660 km discontinuity) and
PKPdf (travelling the Earth’s core). We have to be careful not to interpret those as reflections
from the 660 km discontinuity. This was prevented by carefully checking the signal to noise
ratio in the time window before the PP and SS arrival up to 150 sec and 300 sec, respectively,
when the transition zone discontinuities are expected to arrive. We only included data for which
the signal to noise ratio was larger than 3.0. The signal value was taken as the peak amplitude
of the 30 second PP (or 60 second SS) arrival window, and the noise value was the maximum
amplitude in the time window up to 150 (or 300 seconds) before the PP (or SS) arrival. This
successfully removed seismograms with other major phases arriving in the precursor time window.
In addition, we also stacked our data for subsets of the epicentral distance range, where
no other major phases were expected to arrive, to check that the reflections from the 660 km
discontinuity were still present. The phases Pp410p and Pp660p could potentially interfere at
epicentral distance ranges of 80-100◦ , and PKPdf could be a problem at 120-140◦ . Figs. S4
and S5 show recomputed PP precursor stacks for Figs. 1B and 2C at epicentral distance ranges
of 80-120◦ (no PKPdf interference) and 100-140◦ (no Pp410p and Pp660p interference). This
excludes potentially interfering phases, but the data set is still large enough to make many of
the stacks robust. Figs. S4 and S5 show that reflections from the 660 km discontinuity are still
visible in the resampled stacks. This is additional evidence for the robustness of our 660 km
discontinuity reflections in PP precursors.
A
N America MW Pacific
B
India
India
Depth (km)
400
Depth (km)
400
N America MW Pacific
500
500
600
600
700
700
800
800
Figure S4. Recomputed PP precursor stacks for the same configuration as Fig. 1B for epicentral
distance ranges of (A) 80-120◦ and (B) 100-140◦ .
4
A
1
Alaska
Depth (km)
400
500
2 Pakistan
410
400
520
500
600
700
3 Indian Ocean
410
400
410
520
500
520
600
660
800
600
660
700
800
700
690
800
B
Depth (km)
400
500
410
400
520
500
600
700
800
410
400
410
520
500
520
600
660
600
660
700
800
700
690
800
Figure S5. Recomputed PP precursor stacks for the same configuration as Fig. 2C for epicentral
distance ranges of (A) 80-120◦ and (B) 100-140◦ .
Global observations
We have studied the characteristics of the 660 km discontinuity in two different frequency
bands. Fig. 3A shows the global observations of the 660 km discontinuity in the short period band of 8-75 sec. For comparison, we have also included the global observations in the
long period 15-75 sec. frequency band in Fig. S6.
PdP < 700km
PdP > 700km
no PdP
Figure S6. Global observations of the 660 km discontinuity in the 15-75 sec. frequency band. Similar
to Fig. 3A, which shows the global observations in the 8-75 sec. frequency band.
5
Reflection amplitudes
Reflections from the 660 km discontinuity cannot be seen in global stacks of PP precursors.
We interpret this as being due to the high variability in discontinuity topography and amplitude
from region to region. To quantify this effect more precisely, we investigated the effect of
reasonable variations in discontinuity topography and amplitude for global stacks by looking at
the distribution of our robust observations (i.e. Fig. 3A). We find that 30% of reflections has
a single discontinuity depth less than 700 km, another 30% has a depth larger than 700 km,
10% shows double peaks and the remaining 30% are null observations. There is no Guassian
distribution with a peak around 700 km depth. This distribution indeed leads to a global stack
with no observable and robust P660P reflection.
Then we investigated the range in amplitudes that we observe for the robust observations
of the 660 km discontinuity using the deconvolution technique. We find that S660S is seen in
almost every stack with amplitudes of S660S/SS varying between approximately 3 and 5%. In
stacks where we observe P660P, the amplitudes of P660P/PP vary between 1.5 and 3%. The
robust null observations have P660P/PP amplitudes of less than 1%. These ranges are shown in
Figs. 5 and S7 using errors bars for the corresponding amplitude range at the average epicentral
distance range.
We computed reflection amplitudes of P660P (and S660S) as a percentage of the transmitted
PP (and SS) amplitudes for different velocity and density models using equations given by Aki
& Richards (3). These equations are only valid for first order discontinuities. The velocity and
density profiles given by most mineral physical models also contain gradients. These gradients
are second order discontinuities and they will act as a low pass filter to reflected waves. Thus
we have to investigate the effect of these gradients on reflection amplitudes in our frequency
range. As a rule of thumb (4, 5), a seismic gradient can only be seen by reflected waves if
the thickness h of the gradient is significantly less than 0.5λ, where λ is the wavelength of the
reflected waves. The reflection coefficient becomes very small as the discontinuity thickness h
approaches half the wavelength λ.
The amplitudes are computed as a function of epicentral distance range. We stack our PP
data for epicentral distance ranges between 80◦ and 140◦ and SS between 100◦ and 160◦ . Our
observed amplitudes depend on the epicentral distance range distribution for that specific stack.
In addition to the impedance contrast at the 660 km discontinuity, other factors such as incoherent stacking and attenuation variations influence the amplitude of the precursors. There are
also uncertainties in the precise details of the mineral physical models and their corresponding
6
velocity and density profiles. Thus, we only compare the spread in amplitudes as seen in our
data with the range of amplitudes that can be obtained from the mineral physical models. We
have plotted the range of our observed amplitudes at the average epicentral distance for each
data type and then compared these with the computed amplitudes (Figs. 5 and S7).
The amplitudes predicted by PREM are much larger than our observations for S660S and
P660P, confirming the results of older studies (6, 7). Then we computed reflection amplitudes
for pyrolite models of Weidner & Wang (8) and piclogite models of Vacher et al. (9) for the
660 km discontinuity. These two studies computed seismic velocity and density profiles for
characteristic temperatures in the mantle around 660 km depth. Both computed seismic profiles
for mantle temperatures of 1700-1900 K at 660 km depth, which is generally thought to be the
average mantle temperature and thus most representative for the interpretation of our global
observations.
Our observations agree best with a pyrolite model with 5% Al (Fig. 5). For this model, both
the 1700 and 1900 K geotherms predict amplitudes that fall within our S660S/SS observations.
It can also explain the robust P660P reflections and the null observations. The 1700 K geotherm
corresponds to a first order discontinuity in seismic properties and predicts the correct amplitudes for areas where P660P is seen (dashed line in Fig. 5). The 1900 K geotherm corresponds
to a much smaller first order seismic discontinuity followed by a strong gradient, or second
order discontinuity, in seismic properties. Our reflected P660P waves have a typical wavelength
λ of 80-100 km and the gradient in the velocity profile is spread out over a thickness h of at
least 50 km. As h is more than 0.5λ, our seismic waves are not sensitive to the gradient and
will only be reflected off the first order discontinuity. This results in much smaller amplitudes
for the 1900 K geotherm than for the 1700 K geotherm, which is in agreement with areas where
P660P is not seen (dotted line in Fig. 5). A pyrolite model with 3% Al overpredicts the amplitudes of both S660S and P660P for the 1700 and 1900 K geotherms, and is very similar to the
amplitudes predicted by PREM (Fig. S7A).
Piclogite models are not successful in explaining the range in our observations (Fig. S7B).
The piclogite models (9) are computed for mantle adiabats of 1500 and 1600 K, rather than
geotherms. These corresponds to 1700 K and 1850 K geotherms, respectively, at 660 km depth.
Geotherms of 1700 and 1850-1900 K are representative of average mantle temperatures. Only
the 1700 K geotherm of the piclogite model is within the range of S660S/SS amplitudes, and the
1850 K geotherm underpredicts the amplitudes. In addition, while piclogite geotherms of 1700
and 1850 K are able to predict the range of amplitudes seen for the robust P660P observations
7
(dashed and dotted lines), it does not give small enough amplitudes for the null observations.
Thus, piclogite is not able to explain S660S amplitudes for a range of expected average mantle
temperatures.
S660S/SS
A
P660P/PP
Pyrolite, 3% Al
Pyrolite, 3% Al
10
10
6
8
1700K
PREM
amplitude (%)
amplitude (%)
8
1900K
obs
4
2
0
100
6
1700K
4
2
1900K
null obs
0
120
140
160
80
distance (deg)
B
Piclogite
120
140
Piclogite
10
8
8
PREM
amplitude (%)
amplitude (%)
100
distance (deg)
10
6
obs
4
2
0
100
obs
PREM
1700K
6
PREM
4
obs
2
1850K
1850K
null obs
0
120
140
160
80
distance (deg)
1700K
100
120
140
distance (deg)
Figure S7. Amplitudes of S660S/SS (left panels) and P660P/PP (right panels) for different mineral
physical models, similar to Fig.5. The pyrolite models are taken from Weidner & Wang (8) and the
piclogite models are from Vacher et al. (9). The amplitudes for PREM are given for comparison (solid
lines). The range of amplitudes found in our data are given by error bars at the average epicentral distance
range. The error bars are labelled ‘obs’ for robust ‘S660S and P660P reflections, and ‘null obs’ for null
observations of P660P. (A) Pyrolite model with 3% Al for a geotherm of 1700 K (dashed line) and 1900
K (dotted line). (B) Piclogite model for geotherms of 1700 K (dashed line) and 1850 K (dotted line).
8
References and Notes
1. B. Efron, R. Tibshirani, Science 253, 390 (1991).
2. K. Chambers, J. H. Woodhouse, A. Deuss, Earth and Planetary Science Letters 235, 610
(2005).
3. K. Aki, P. Richards, Quantitative Seismology, vol. 1 (W.H. Freeman and Company, San
Francisco, 1980a).
4. P. Richards, Zeitschrift für Geophysik 38, 517 (1972).
5. P. M. Shearer, Geophysical Monograph 117, 115 (2000).
6. P. M. Shearer, M. P. Flanagan, Science 285, 1545 (1999).
7. C. H. Estabrook, R. Kind, Science 274, 1179 (1996).
8. D. J. Weidner, Y. Wang, J. Geophys. Res. 103, 7431 (1998).
9. P. Vacher, A. Mocquet, C. Sotin, Phys. Earth Planet. Inter. 106, 275 (1998).
9