Receiver function study in northern Sumatra and the

Malaysian peninsula
Katrin Kieling, Dirk Roessler, Frank Krueger

To cite this version:

Katrin Kieling, Dirk Roessler, Frank Krueger. Receiver function study in northern Sumatra
and the Malaysian peninsula. Journal of Seismology, Springer Verlag, 2010, 15 (2), pp.235-259.
�10.1007/s10950-010-9222-7�. �hal-00653063�

HAL Id: hal-00653063

https://hal.archives-ouvertes.fr/hal-00653063
Submitted on 17 Dec 2011

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
Journal of Seismology manuscript No.
(will be inserted by the editor)

Receiver function study in northern Sumatra and the

Malaysian peninsula
Katrin Kieling · Dirk Roessler · Frank Krueger

Received: date / Accepted: date

Abstract In this receiver function study we investigate the Moho is located at 37 and 40 km depth. At station
the structure of the crust beneath six seismic broad- GSI, situated closest to the trench, the subducting slab
band stations close to the Sunda Arc formed by sub- is imaged as a north-east dipping structure separated
duction of the Indo-Australian under the Sunda plate. from the sediment layer by a 10 km wide gradient in VS
We apply three different methods to analyse receiver between 10 and 20 km depth. Within the subducting
functions at single stations. A recently developed algo- slab VS ≈ 4.7 km. At station BSI the subducting slab is
rithm determines absolute shear wave velocities from found at depth between 90 and 110 km dipping 20◦ ± 8◦
observed frequency dependent apparent incidence an- in approximately N60◦ E. A velocity increase in similar
gles of P-waves. Using waveform inversion of receiver depth is indicated at station PSI, however no evidence
functions and a modified Zhu and Kanamori algorithm, for a dipping layer is found.
properties of discontinuities such as depth, velocity con-
Keywords receiver functions · absolute shear wave
trast, and sharpness are determined. The combination
velocity · Sumatra · subduction zone structure
of the methods leads to robust results. The approach is
validated by synthetic tests. PACS 91.30.Wx
Stations located on Malaysia show high shear wave
velocities (VS ) near the surface in the range of 3.4 to
3.6 kms−1 attributed to crystalline rocks and 3.6 to 1 Introduction
4.0 kms−1 in the lower crust. Upper and lower crust are
clearly separated, the Moho is found at normal depths The Sunda (volcanic) Arc (Fig. 1) is caused by subduc-
of 30 − 34 km where it forms a sharp discontinuity at tion of the oceanic Indo-Australian plate underneath
station KUM or a gradient at stations IPM and KOM. the Sunda plate, which started during the Cretaceous
For stations close to the subduction zone (BSI, GSI, i. e. ∼ 100 Ma ago (Hamilton, 1979). It extends about
PSI) complexity within the crust is high. Near the sur- 6000 km from the Andaman Islands in the north-west
face low VS of 2.6 − 2.9 kms−1 indicate sediment lay- to the Lesser Sunda Islands in the south-east. The re-
ers. High VS of 4.2 kms−1 are found at depth greater gion is a formidable natural geologic laboratories and
than 6 km and 2 km at BSI and PSI, respectively. There, permits to study the behaviour of the lithosphere in
subduction zones, not least due to such extreme events
Katrin Kieling as the tsunamigenic 2004 earthquake or the Toba erup-
Institute of Geosciences, University of Potsdam, now at GFZ Ger- tion ∼ 74,000 years ago.
man Research Centre for Geosciences, Telegrafenberg, D-14473 Along Sumatra Baroux et al (1998) found a conver-
Potsdam, Germany
Tel.: +49-331-2881288
gence with obliquity of 20◦ up to 30◦ . Subarya et al
E-mail: [email protected] (2006) report a convergence rate of 65 mm/yr. The Re-
Dirk Rößler E-mail: [email protected] · Frank Krüger E-mail:
gional Upper Mantle (RUM) model by Gudmundsson
[email protected] and Sambridge (1998) infers the geometry of the slab
Institute of Geosciences, University of Potsdam, K.-Liebknecht- from seismicity (see Fig. 1 for the western part of Suma-
Str. 24/25, 14471 Golm,Potsdam, Germany tra). It shows that the depth of the slab varies from
2

B 80
A
depth [km] 60
60
400
BSI KUM 300 40 40
5˚N
200
IPM

80
100
20
0
Simelue Toba Malaysia
PSI

80
60
Nias KOM
Ind

20
GSI

60
o−

Su
A

Su

40
us

40
nd

0˚

m
tra

ap

65 mm/yr

at
lia

ra
lat
np

e
lat

80
e

60
95˚E 100˚E 105˚E 80

Fig. 1 A Map of Sumatra with considered stations (squares, compare Tab. 1) and major faults. Colours of isolines indicate depth of
the subducted slab (Gudmundsson and Sambridge, 1998). B Earthquakes (stars) used for receiver function analysis. Red lines show
epicentral distance from station PSI in degree. The central blue rectangle outlines the region shown in A.

600 km in eastern Java to 300 km in northern Suma- unit on Sumatra, the volcanic rocks are directly exposed
tra. From tomographic studies of this and other sub- on the surface at the Malaysian peninsula.
duction zones and modelling (Widiyantoro and van der
The Malaysian peninsular is characterised by the
Hilst, 1996; Gorbatov et al, 1999) it is known that the
Betong-Raub suture zone (Metcalfe, 2000), which was
subducting plate forms a cold, high-density slab with
formed during the subduction of the Paleo-Thethys
P- and S-wave velocities 5-10% higher than in the sur-
Ocean beneath Indochina during the Late Paleozoic and
rounding normal mantle. However, only little is known
the collision of the Sibumasu Terrane with Indochina.
regarding the crustal and upper mantle shear wave ve-
locity structure of the whole region from global sur- Receiver functions are widely employed to detect
face wave studies with rather limited resolution (e.g. P-to-S-converted waves and are especially useful to im-
Ritzwoller et al, 2002). In recent years extensive reflec- age seismic discontinuities if dense networks of high-
tion seismic studies of the region were carried out, e. g. quality three-component broadband stations are in-
by Kopp et al (2001); Singh et al (2008); Franke et al stalled. However, in many regions of the Earth the only
(2008) mapping the uppermost 10-20 km of the crust available information regarding the shear wave velocity
offshore Sumatra. Furthermore Wu et al (2005) stud- depth profile is due to relatively isolated stations. Clas-
ied the regional S-velocity of the South China Sea and sical receiver function methods for single stations in-
provides S-velocity maps in several depths. clude picking of difference traveltimes for direct conver-
The subduction has significant impact on the upper sions and multiples and estimation of the corresponding
mantle and the crust of the overriding plate. Increasing discontinuity depth with an average background model
pressure at depth causes slab dehydration (Rüpke et al, (e.g. Krueger et al, 2002); estimation of discontinuity
2004) and the transition from gabbro to eclogite. Fluids depth and VP /VS ratio by stacking direct Ps conver-
are released in the lithosphere. They cause serpentinisa- sion and multiple waveforms in a gridsearch procedure
tion and reduce the melting temperatures of peridotite (Zhu and Kanamori, 2000); inversion of stacked receiver
rocks (Gill, 1981; Hacker et al, 2003). This leads to the functions for 1D shear wave models (Ammon, 1991).
generation of low-density magma which rises and in- The retrieval of absolute shear wave velocities from re-
trudes or extrudes on the upper lithosphere, forming a ceiver functions is still a major challenge. Several stud-
chain of volcanoes. The significance of the volcanic ac- ies applied combined inversions of receiver functions
tivity is demonstrated for example by the Toba caldera and phase-/group-velocity dispersion curves from sur-
(see Fig. 1). It is an ≈ 30×100 km large topographic de- face waves to stabilise the inversion (e.g. Julià et al,
pression in northern Sumatra which was formed during 1998; Chang et al, 2004). Recently Svenningsen and Ja-
several eruptions over the last 1.2 million years. Geo- cobsen (2007) proposed to use the frequency dependent
logic maps (e. g. Crow and Barber, 2005) show large apparent incidence angle of receiver functions to esti-
areas dominated by intrusive rocks such as granite or mate absolute shear wave velocities as well as depths of
gabbro as well as extrusive rocks such as andesite or discontinuities. The latter method has the advantage
basalt. While those are often covered by a sedimentary that local velocity models can be retrieved.
 3

The aim of this study is twofold. First, crustal seis-

mic properties in the study region are investigated ex-
ploiting different single station receiver function tech-
niques. We use the method of Svenningsen and Jacob-
sen (2007) to retrieve initial absolute shear wave veloc-
ity models by a gridsearch technique. In a next step
this velocity model is used to invert for 1D models (see
e. g. Kind et al, 1995, Jacobsen and Svenningsen, 2008)
using receiver function waveforms and fixed VP /VS ra-
tios. We find the VP /VS ratio of the top layer and the
depths of discontinuities following the approach of Zhu
and Kanamori (2000) but fixing VS instead of VP , what
enables us to link the analysis of apparent incidence Fig. 2 Ray diagram of phases used for receiver function analysis
angles directly to the technique of Zhu and Kanamori. in a two-layer model. The incident P-wave converts at depth H
This modified algorithm is referred to as MZK from into the refracted P- and S-waves and their multiples within the
horizontal layer.
hereon. Since all methods give the depth of disconti-
nuities the consistency of the results can be checked.
Furthermore we test the effect of velocity gradients on shallow layers but not caused by distant deep struc-
the outcome of the different methods. tures. As we want to take these shallow layers into ac-
In a second step, based on the analysis of the crustal count during waveform inversion, we also want to con-
properties, structures related to the subduction zone in sider the phases at zero time. After rotation, a deconvo-
the upper mantle are investigated. With the help of lution in the time domain is performed by Wiener filter-
lateral and azimuthal variations of conversion depths ing (Kind et al, 1995). The deconvolution is stabilised
found from receiver functions, the dip of the subducting by adding a constant 0.1 to the diagonal elements of
plate is constrained beneath two stations. the filter matrix. This corresponds to intermediate sta-
bilisation (Svenningsen, 2007). The calculation of the
inverse filter depends on the length of the source-time
2 Methodology
function and the arrival time of later phases. In this
The receiver function method takes advantage of the study we chose window lengths between 80 and 140 s
fact that a seismic compressional wave (P-wave) im- following the direct P-wave. This inverse filter is ap-
pinging obliquely on a discontinuity is partly refracted plied to the Q- and T-components to obtain Q- and
as a P-wave and partly converted to a shear wave (S- T-receiver functions.
wave) (see Fig. 2). Besides the direct P-to-S converted
Ps-wave there are also the multiple reverberations PpPs
2.1 Investigation of apparent incidence angles
and PpSs (see Fig. 2) that have a comparable ampli-
tude to the Ps phase. Delay times and amplitudes of The apparent incidence angle of the incoming P-wave
direct and multiple phases provide information about results from the superposition of the incoming and the
the depth of a discontinuity and the impedance con- reflected wavefield. It can be used to estimate absolute
trast. For a 1D structure traveltime differences depend S-wave velocities from receiver functions (Svenningsen
on wave type and slowness only. and Jacobsen, 2007) based on the relation
The calculation of receiver functions mainly involves  
two steps. First, seismograms are rotated to the LQT- sin ī2P sin(iP )
coordinate system of the ray assuming theoretical back- = =p (1)
VS VP
azimuths and incidence angles (Kind et al, 1995). We
used incidence angles given by Hoang-Trong and Behe given by Wiechert (1907). Here, p, iP , and īP are the
(1972). By this rotation the P-wave signal should be slowness, the theoretical incidence angle, and the ap-
minimised on the Q-component. This improves visibil- parent incidence angle of the direct P-wave, respec-
ity of shallow layers. However, artificial phases appear tively. The latter can be measured. Hence, calculating
on Q-components when incorrect angles are used for ro- the slowness with the help of IASP91 reference model
tation. The true azimuth deviates from the theoretical (Kennett, 1991) and measuring īP , VS underneath a
value by less than 10◦ . Deviations in incidence angles station can be retrieved. Typically the apparent an-
are larger. Most likely, these deviations in incidence an- gle of incidence is determined by dividing the R- by
gle are due to erroneous assumptions on velocities of the Z-component of the receiver function (Svenningsen
4

and Jacobsen, 2007). In this study, we use the Q- and 2.2 Waveform inversion
L-component and add the theoretical incidence angle
which was used for rotation (see formula 2). Receiver Receiver functions can also be inverted for VS -depth
functions are convolved with a squared cosine of width models by linear inversion. This is non-unique (Ammon
T , resulting in low-pass filtering. The convolution is et al, 1990) and depends on the starting model. In this
evaluated at time zero which results in an apparent study we use the method by Kind et al (1995) and a
incidence angle dependent on the low-pass parameter large number of thin layers which allows quasi-gradient
T, models. Jacobsen and Svenningsen (2008) show that
R  fine layering and a high number of iterations result in
T 2 πτ more unique results.
(−T ) rfQ (τ ) cos ( 2T )dτ
īP (T ) = arctan  R T  + iROT , (2)
rf (τ ) cos2 ( πτ
(−T ) L 2T )dτ
2.3 Modified Zhu&Kanamori algorithm (MZK)
where rfQ and rfL are Q- and L-components of the re-
ceiver function and iROT is the incidence angle which A widely used method to extract the depth of disconti-
was used for rotation to the LQT-system. Inserting (2) nuities, H, and the VP /VS ratio from receiver functions
into (1) gives a T-dependent VS denoted as apparent was introduced by Zhu and Kanamori (2000). This in-
velocity VS,app volves stacking of receiver functions within time win-
h i dows according to predicted delay times of Ps, PpPs,
sin īP (T ) and PpSs and analysis of the stack amplitudes. We use
2
VS,app (T ) = . (3) semblance analysis (Neidell and Taner, 1971) to im-
p prove resolution. The semblance S within a time win-
With increasing T the calculation window increases dow is the energy of the stacked trace divided by the
and direct conversions and their multiples affect īP (T ). sum of the energies of all traces that make up the stack,
Thus, for higher values of T , the apparent velocity can P Pn 2
be associated to a higher depth (see Fig. 3B,C). For a τ | i=1 rfi (tPs ) + rfi (tPpPs ) + rfi (tPpSs )|
S= P P n 2 2 2
detailed description of the method see Svenningsen and n τ i=1 (|rfi (tPs )| + |rfi (tPpPs )| + |rfi (tPpSs )| )
Jacobsen (2007). For a model consisting of a layer over (5)
a halfspace the S-wave velocities can be directly read
from the VS,app (T ) curve. More complicated models re- where n equals the number of traces used, rfi is the re-
quire an inversion of the VS,app (T ) curve. ceiver function i and tPs , tPpPs and tPpSs are the delay-
Curves of VS,app (T ) are similar for different slow- times of the conversion and its multiples. τ is the time
nesses (Svenningsen and Jacobsen, 2007). In the pres- window of 2 s length arranged symmetrically around
ence of noise, mean curves, V̄S,app (T ), can therefore the considered times. Delay times and semblances are
be formed by stacking VS,app (T ) curves for all receiver usually computed in a grid search over H and VP /VS
functions at one station. They are modelled by the by fixing VP . Where semblance is at a maximum the
VS,app (T ) curve for a mean slowness which is 6.4 s/◦ correct H and VP /VS ratio are assumed to be found.
in this study. Models for VS can be retrieved in a grid Here, we make two modifications. The resulting al-
search by forward modelling of VS,app (T ) from synthetic gorithm is referred to as MZK. First, we assume that VS
receiver functions calculated by the Thomson-Haskell is given instead of VP , which permits the usage of VS -
method (Kosarev et al, 1987). The best fitting curve values as obtained from the analysis of apparent inci-
was determined by minimising the misfit esd dence angle and the waveform inversion. Formulas used
for calculating delay times are:
T =Tmax
r 2 r !
syn
X q −2
esd = V̄S,app (T ) − VS,app (T ) , (4) tPs = H −2
VS − p − 2 VP 2
VS VS −p
T =0.5
r !
(6)
q −2
syn V
where VS,app is modelled from synthetic receiver func- tPpPs = H VS − p2 +
−2
VS VS
P
− p2
tions. In our study Tmax is the largest T for which q
V̄S,app (T ) is below 5 kms−1 . This limit was chosen, as tPpSs = 2H VS−2 − p2 .
VS,app (T ) tends to diverge for higher values of T , and
5 kms−1 is a reasonable upper limit for VS of the up- Reformulation of Eqs. (6) gives the general form H =
per mantle. To obtain receiver functions, the theoretical H(t, VVPS ). For Ps the slope of the H(tPs , VVPS ) is positive,
seismograms are treated in the same way as the data. while for the first multiple the slope of H(tPpPs , VVPS )
 5

is negative and for the second multiple H(tPpSs , VVPS ) for a slowness of 7.08 s/◦ (57◦ epicentral distance), since
is constant (see e.g. Fig. 3G). This results in a more this is the mean slowness of the stacked receiver func-
distinct semblance maximum in the VP /VS -H-plot and tions. We invert receiver functions for VS -depth mod-
higher resolution of VP /VS . Second, we replace the orig- els with VS /VP fixed at 1.73 in the crust and 1.79 in
inal layer-over-halfspace model by a two-layers-over- the mantle. Densities in the crust are calculated using
halfspace model. In our study this turns out to be more Birch’s law (Birch, 1961), but fixed to 3.3 g/cm3 in the
appropriate because we can often constrain the upper mantle. Because of their non-uniqueness, 200 inversions
layer from the incidence angle analyses and waveform with different starting models are performed. They are
inversion. Formulas for calculating delay times are given randomly selected from a uniform distribution within
in the appendix. In this study time windows of 2 s ±0.75 km−1 around the initial reference model which is
length centred around the predicted times tPs , tPpPs , obtained from VS,app (T ) analysis (see Figs. 3D-F and
and tPpSs are used for stacking. All three phases are 4D-F). The distribution of the results is analysed. Out
equally weighted during the stacking. of those, the model with the best misfit is searched.
For the calculation of a mean model only those models
with a misfit of maximal 4 times the global best misfit
3 Synthetic Examples
are used. By this we correctly reproduce the position
of the discontinuity and of the gradient, the velocities
First we study the different effects on receiver functions
of both layers (see Figs. 3D, 4D), and the waveforms
that can be expected due to sharp seismic discontinu-
(Figs. 3E,F and 4E,F).
ities compared to velocity gradients (Figs. 3C, 4C). For
Major differences due to the model differences are
this purpose synthetic seismograms are computed for
found by the MZK (Figs. 3G and 4G). As mentioned
two simple layer-over-halfspace models. We chose for
before, the slopes H(t, VVPS ) (coloured curves in Figs. 3G
the upper layer VS = 3.2 kms−1 and VP /VS = 1.73 over
and 4G) differ in sign depending on phase type. This
a halfspace with VS = 4.6 kms−1 , VP /VS = 1.79. The
leads to a distinct intersection of the three functions
discontinuity and the gradient are located at depths of
at the expected depth and VP /VS ratio and reduced
29 km and between 25 and 33 km, respectively. Based
smearing along the curves (Fig. 3G). For the sharp dis-
on these 2 models, receiver functions with a source
continuity the depth√can be found in the grid search at
wavelet of 1 s duration (Küpper, 1958) are calculated
29 km and VP /VS = 3. Both equal the original model.
and stacked assuming epicentral distances between 27◦
In the case of the gradient, however, the method is al-
and 92◦ (Figs. 3A, 4A). For stacking, receiver functions
most insensitive to VP /VS . No single depth of conver-
were moveout corrected assuming Ps, PpPs or PpSs
sion can be found but the broad semblance maximum
phases and a slowness of 6.4 s/◦ (Yuan et al, 1997). In
resembles the location of the gradient. Therefore, fix-
the single traces we find that the converted phases are
ing shear wave velocities to values obtained in prior
broadened in the gradient model compared to conver-
inversions we expect to distinguish between sharp seis-
sion at the sharp discontinuity. The effect is most pro-
mic contrasts and gradient. The finding of VP /VS seems
nounced for the PpPs and PpSs multiples. In the move-
promising for first order seismic discontinuities but not
out stacked receiver functions the waveform of the refer-
in case of velocity gradients.
ence phase retains its shape while the other phases are
artificially broadened. This effect is most pronounced
in the discontinuity model for the multiples when Ps is 4 Application to stations along Sunda Arc and
assumed (Figs. 3A, top traces). Ignoring this distortion, Malaysia
discontinuities may be misinterpreted as gradients. Re-
sulting VS,app (T ) curves are only marginal different for 4.1 Data and processing
both models (Figs. 3B,C and 4B,C). Therefore, we can-
not expect to discriminate between a sharp discontinu- We study crustal and upper mantle seismic structures
ity and a gradient from VS,app (T ). It turns out that on northern Sumatra and the Malaysian Peninsula. Re-
the short-period part of the VS,app (T ) curve is usually ceiver functions were calculated at 6 stations that had
stable and equals VS of the upper layer. been in operation for several years (Fig. 1A, Tab. 1).
During waveform inversions mismodelling from 112 earthquakes within epicentral distances of 30◦ −96◦
stacking was avoided by limiting the considered dis- and magnitudes mB ≥ 5.8 were chosen (Fig. 1B).
tance range to 47◦ − 67◦ , what corresponds to a slow- The resulting backazimuthal event distribution is un-
ness range of 6.4−7.8 s/◦ and allows to stack a sufficient even with a majority in north-east (Japanese and Kuril
number of receiver functions at each station. For the islands) and south-east (Kermadec-Tonga subduction
inversions synthetic receiver functions were calculated zone). The number of calculated receiver functions per
6

Sharp Discontinuity
0
SUM−PpSs (A) (B) delay time [s] (C)
SUM−PpPs 1 10 100
SUM−Ps
10
92°
87°
82°

depth [km]
77° 20
72°
67°
62° 5 30

VS,app [km/s]
57°
52°
47° 4
42° 40
37°
32° 3
27° 50
0 10 20 1 10 100 2 3 4 5
time [s] low−pass parameter T [s] VS [km/s]

0 0 0
(D) (E) (G)

delay time [s] for the maximal vP/vS

10
20 10
30
2
depth [km]

depth [km]
40 20
50
60 (F) 30
4
70
80 40
90
100 50 6
2 3 4 5 6 −5 0 5 10 15 20 25 30 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3
VS [km/s] time [s] vP/vS with vS=3.2

0.0 0.1 0.2 0.3 0.4 0.5 0.2 0.4 0.6 0.8
relative frequency semblance
Fig. 3 Analysis of synthetic receiver functions computed for a model consisting of a layer (thickness: 29 km, VS = 3.2 kms−1 ,
VP /VS = 1.73) over a halfspace (VS = 4.6 kms−1 , VP /VS = 1.79). A Synthetic receiver functions for events in different epicentral
distances (annotated on the left). Important phases also discussed in the text are highlighted. Blue, black, and green ticks mark
arrival of the primary Ps conversion and associated PpPs and PpSs multiples (compare G). Top three traces: summation with moveout
corrections for the Ps, PpPs, and PpSs phases assuming a slowness of 6.4 s/◦ (67 ◦ epicentral distance). B Synthetic receiver function
(6.4 s/◦ (black), top) and derived VS,app (T ) curve (bottom) with logarithmic time axes. For comparison, receiver functions and derived
VS,app (T ) curve for a slowness of 5.0 s/ are shown in green. C VS -depth model corresponding to the VS,app (T ) curve in B. D Density
function of the VS -depth models from 200 inversions. For inversion, synthetic receiver functions in A were stacked within 47-67◦
epicentral distance. White dashed lines: range limit of starting models around the VS derived from VS,app (T ) in B. Red solid and black
dashed lines: resulting medium velocities and standard deviations. E Receiver functions predicted by the models obtained from 200
inversions in D. Blue line: original receiver function, low-pass filtered at 1 Hz. It results from stacking receiver functions for epicentral
distances between 47 ◦ and 67 ◦ (without moveout correction). F Same as E but for receiver functions corresponding to the starting
models. G Results of MZK inversions assuming one layer (VS = 3.2 kms−1 ) above a halfspace. Thin black line indicates the area
with values greater than 75% of the semblance maximum. For comparison with receiver functions in A, delay times (right y-axis) are
calculated with VP /VS given at semblance maximum. Blue, black, and green lines show conversion depth curves H (Eqs. 6) for given
VP /VS ratios and travel time differences. The latter correspond to the ticks in A (upper three traces) with the same colour.
 7

Gradient
0
SUM−PpSs (A) (B) delay time [s] (C)
SUM−PpPs 1 10 100
SUM−Ps
10
92°
87°
82°

depth [km]
77° 20
72° 6
67°
62° 30

VS,app [km/s]
57° 5
52°
47° 4
42° 40
37° 3
32°
27° 2 50
0 10 20 1 10 100 2 3 4 5
time [s] low−pass parameter T [s] VS [km/s]

0 0 0
(D) (E) (G)

delay time [s] for the maximal vP/vS

10
20 10 2
30
depth [km]

depth [km]
40 20 4
50
60 (F) 30 6
70
80 40 8
90
100 50 10
2 3 4 5 6 −5 0 5 10 15 20 25 30 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3
VS [km/s] time [s] vP/vS with vS=3.2

0.0 0.1 0.2 0.3 0.4 0.5 0.2 0.4 0.6 0.8
relative frequency semblance
Fig. 4 Same as Fig. 3 but for a synthetic model containing a velocity gradient at 25 to 33 km depth. PpPs and PpSs multiples (A,
B, E, F) and the band of high semblance (G) are broadened. Semblance by the MZK (G) becomes almost insensitive to VP /VS .

station varies with station operation time and due to 1 s ≤ T ≤ 7 s. This selection of high-quality receiver
different data quality (Tab. 1). Data processing was car- functions finally leads to stable mean curves. They are
ried out manually. Erroneous and low-quality data were used to find a VS -depth model in a grid search. Stability
rejected. Seismograms were processed in the same way is tested by jack-knifing i.e. the grid search was repeated
as the synthetics in section 3. Table 2 summarises the 160 times including only 60% of all available VS,app (T )
results for the considered stations and methods. curves that were randomly selected to form V̄S,app (T ).
Long period noise may obscure the VS,app (T ) curves For stations BSI and PSI observed V̄S,app (T ) curves
(Svenningsen and Jacobsen, 2007). The modelling of cannot be explained by three-layer models. Therefore,
VS,app (T ) was therefore refined to include only high a fourth layer was added. The velocity of the upper-
quality receiver functions. First, only receiver functions most layer was always read directly from V̄S,app (T )
with signal-to-noise ratio > 2 before and after low- within the interval 0.2 s < T < 0.5 s. The velocities
pass filtering at 0.2 Hz are accepted. VS,app (T ) curves of the following layers where allowed to vary between
are calculated from the remaining receiver functions. 3.0 kms−1 < VS < 5.6 kms−1 in steps of 0.2 kms−1 .
The mean and the standard deviation are computed Depths were searched from 0 to 60 km in steps of 1 km.
at every point of the curves. An improved mean curve Density plots show the relative frequency of the ob-
(V̄S,app (T )) is found by rejecting again all curves, which tained models as well as the median model and the
exceed the limit of one standard deviation in the range standard deviation (e. g. Fig. 5c). Waveform inversion
8

station RF RF(47-67◦ ) network seismometer latitude [◦ N] longitude [◦ E] starting date

BSI 64 29 JISNET CMG-3T/100 5.49 95.29 01-Jan-2000
GSI 65 23 GEOFON STS-2 1.30 97.57 02-Oct-2005
IPM 65 22 MY STS-2 4.48 101.03 29-Mar-2006
KOM 61 18 MY STS-2 1.79 103.84 29-Mar-2006
KUM 83 25 MY STS-2 5.29 100.65 17-Feb-2005
PSI 75 31 PS STS-2 2.69 98.92 29-Mar-1993

Table 1 Station specifications and total number of calculated receiver functions (RF) and for earthquakes within 47-67◦ epicentral dis-
tance (compare Fig. 1). Networks are: MY – Malaysian National Seismic Network, PS – Pacific21, GEOFON – GeoForschungsZentrum
Potsdam, and JISNET – Japan-Indonesia Seismic Network.

is limited to receiver functions within a slowness range contrast we find very high semblance for shallow depths
of 6.4 − 7.8 s/◦ . Initial models are based on the out- between 2 and 4 km which is consistent with the thin
come of the VS,app (T ) modelling. A mean model is cal- shallow layer found from the modelling of the VS,app (T )
culated based on the models with misfits smaller than curves (Fig. 5B,C).
four times the best misfit in case of Malaysian stations
and twice the best misfit in case of Sumatran stations.
Observed V̄S,app (T ) curves are compared with VS,app (T )
predicted from the outcome of the waveform inversions
(red lines in Figs. 5B-10B). Finally, the MZK is applied
by fixing VS which was retrieved before. The thickness 4.3 IPM
of the uppermost layer H1 is constrained by the pre-
ceding waveform inversions. Receiver functions at station IPM show a positive con-
version close to zero time (Fig. 6A), indicating high
near-surface velocities. This is indicative of imperfect
4.2 KUM seismogram rotation. Furthermore, converted phases
are found at 2 s and 3.5 s corresponding to a conver-
Receiver functions at station KUM (Fig. 5A) show pos- sion at a crustal discontinuity and at the Moho, respec-
itive signals at zero time as well as conversions from a tively. The phases at 10 and 13.5 s may be associated
shallow interface at approximately 1 s. The deviations to multiples, although arrivals are rather early if they
of the observed from the theoretical incidence angle correspond to the direct conversion at 3.5 s.
leading to phases at zero time are mainly due to shal- The VS -depth models obtained from apparent inci-
low layers near the surface. The Moho conversion at 4 s dence angles (Fig. 6B) support these observations. Up-
appears relatively weak. However, corresponding PpPs per crust velocities (VS = 3.6 kms−1 ) seem to be sep-
and PpSs multiples are clear at 13.5 s and 17.5 s. Mod- arated by a discontinuity near 10 km from the lower
elling of VS,app (T ) (Fig. 5B,C) results in a shallow, 3 km crust where VS = 4.2 kms−1 . The Moho is found at
thick layer with VS = 3.3 kms−1 followed by increased 24 km depth.
velocities of VS = 3.6 kms−1 down to a depth of 25 km.
Velocities are well defined by this method. Depth reso- Waveform inversion (Fig. 6C) confirms the VS =
lution turns out to be weaker and the variability of the 3.6 kms−1 but shows a thickness of about 14 km for
lower depth limit is high. the top layer and VS = 4.1 kms−1 for the lower crust.
By waveform inversion the uppermost layer is unre- The Moho is apparently formed by a velocity gradi-
solved (Fig. 5D) but a discontinuity at approximately ent between 25 to 35 km depth where VS increase by
19 km is indicated. For this upper layer VS = 3.5 kms−1 . 0.06 kms−1 per km. However, scattering of the models
Below VS = 4.0 kms−1 down to the Moho which is found is rather high in this region. Within the upper mantle
as a sharp discontinuity at 34 km. In the uppermost VS = 4.7 kms−1 is found.
mantle VS = 4.7 kms−1 is found. The MZK (Fig. 6G) shows increased semblance at
An abrupt Moho discontinuity is also confirmed by 12 − 16 km depth corresponding to the crustal disconti-
MZK (Fig. 5G). Semblance forms a pronounced and nuity found before. Semblance is high within a broader
distinct maximum indicating the Moho at a depth of depth range from 24 to 33 km. Similar to the observa-
33 km and a rather high VP /VS ratio of VP /VS = 1.84. tions from synthetic examples sensitivity to VP /VS is
The discontinuity at 19 km is not resolved but increased low (compare Fig. 4) confirming the velocity gradient
semblance is found at depths between 15 and 20 km. In at Moho depth found by waveform inversion.
 9

1 4 13.5 17.5
KUM
7 0
SUM−PpSs (A) (B) (C)
SUM−PpPs
SUM−Ps 6 10

85−90°,1

VS,app [km/s]
5 1.0

depth [km]
80−85°,8 20

relative frequency
75−80°,5
70−75°,7 0.8
65−70°,4 4
60−65°,6 30 0.6
55−60°,4
50−55°,3 3 0.4
45−50°,8 40
40−45°,5 0.2
35−40°,7 2
30−35°,6 0.0
50
0 10 20 0.5 1 2 5 2 3 4 5
time [s] low−pass parameter T[s] VS [km/s]

0 0 0
(D) (E) (G)

delay time [s] for the maximal vP/vS

10
20 10 2

30
4
depth [km]

depth [km]

40 20
50
60 (F) 30 6

70
8
80 40
90
10
100 50
2 3 4 5 6 −5 0 5 10 15 20 25 30 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3
VS [km/s] time [s] vP/vS with vS1=3.5, vS2=4.0

0.0 0.1 0.2 0.3 0.4 0.5 0.1 0.2 0.3 0.4 0.5 0.6 0.7
relative frequency semblance

Fig. 5 Analysis of receiver functions at station KUM (see with Fig. 3 for more details). A Receiver functions, stacked within bins of 5◦
according to epicentral distance and low-pass filtered at 1 Hz. Distance ranges and number of stacked receiver functions are annotated
on the left. Notice complex waveforms also for bins with larger number of observations. Arrows mark delay times of important phases.
B VS,app (T ) curves. Grey lines: 160 mean curves (V̄S,app (T )) from jackknifing. Black and red curves: synthetic VS,appsyn (T ) curves
predicted from retrieved VS models and from the medium model resulting from waveform inversion in D, respectively. C Density
function of the VS -depth models corresponding to the mean V̄S,app (T ) curves in B. Black solid and dashed lines show the mean
and one standard deviation. D red solid shows mean and black dashed lines show one standard deviation. G Semblance from MZK
assuming 2 layers above a halfspace. Velocities of these 2 layers are VS1 = 3.5 kms−1 and VS2 = 4.0 kms−1 . The thickness of the
uppermost layer is fixed at 19 km.
10

0.52 3.5 10 13.5

IPM
7 0
SUM−PpSs (A) (B) (C)
SUM−PpPs
SUM−Ps 6
95−100°,1
10

85−90°,2

VS,app [km/s]
5 1.0

depth [km]
80−85°,8 20

relative frequency
75−80°,6
70−75°,5 0.8
65−70°,6 4
60−65°,5 30 0.6
55−60°,3
50−55°,3 3 0.4
45−50°,6 40
40−45°,5 0.2
35−40°,2 2
30−35°,4 0.0
50
0 10 20 0.5 1 2 5 2 3 4 5
time [s] low−pass parameter T[s] VS [km/s]

0 0 0
(D) (E) (G)

delay time [s] for the maximal vP/vS

10
20 10
2
30
depth [km]

depth [km]

40 20
50 4
60 (F) 30
70
6
80 40
90
100 50 8
2 3 4 5 6 −5 0 5 10 15 20 25 30 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3
VS [km/s] time [s] vP/vS with vS1=3.6, vS2=4.1

0.0 0.1 0.2 0.3 0.4 0.5 0.1 0.2 0.3 0.4 0.5 0.6 0.7
relative frequency semblance

Fig. 6 Analysis of receiver functions computed from data recorded at station IPM. For MZK the thickness of the uppermost layer is
fixed at 14 km. See with Figs. 3 and 5 for more details.
 11

4 7 14 17
KOM
7 0
SUM−PpSs (A) (B) (C)
SUM−PpPs
SUM−Ps 6 10

85−90°,1

VS,app [km/s]
5 1.0

depth [km]
80−85°,5 20

relative frequency
75−80°,8
70−75°,4 0.8
65−70°,4 4
60−65°,3 30 0.6
55−60°,4
50−55°,4 3 0.4
45−50°,6 40
40−45°,7 0.2
35−40°,3 2
30−35°,1 0.0
50
0 10 20 0.5 1 2 5 2 3 4 5
time [s] low−pass parameter T[s] VS [km/s]

0 0 0
(D) (E) (G)

delay time [s] for the maximal vP/vS

10
20 10
30
2
depth [km]

depth [km]

40 20
50
60 (F) 30
70 4

80 40
90
100 50
2 3 4 5 6 −5 0 5 10 15 20 25 30 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3
VS [km/s] time [s] vP/vS with vS1=3.5, vS2=3.9

0.0 0.1 0.2 0.3 0.4 0.5 0.1 0.2 0.3 0.4 0.5 0.6 0.7
relative frequency semblance

Fig. 7 Analysis of receiver functions computed at station KOM. For MZK the thickness of the uppermost layer is fixed at 12 km. See
with Figs. 3 and 5 for more details.
12

BSI GSI IPM KOM KUM PSI

1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3

VS1 2.8 2.9 2.6 2.7 3.6 3.6 3.5 3.5 3.3 3.5 2.9 3.1
0.1 0.3 0.2 0.3 0.1 0.3 0.1 0.4 0.1 0.4 0.1 0.5
VS2 3.4 0.17∗ 0.16∗ 0.13∗ 4.2 4.1 4.2 3.9 3.6 4.0 4.2 4.1
0.2 0.3 0.3 0.2 0.4 0.2 0.4 0.3 0.4
VS3 4.2 4.2 - 0.06∗ 0.14∗ 4.0 3.9
0.03∗
0.3 0.5 0.2 0.5
VS4 (5.6) 4.8 (5.3) 4.7 (5.6) 4.7 (5.6) 4.8 (4.8) 4.7 (5.6) 4.7
(0.3) 0.4 (0.3) 0.4 (0.1) 0.4 (0.2) 0.4 (0.3) 0.4 (0.3) 0.5
H1 1 4† 5 7 10 14 14† 10 12 12 3 4 1 10†
1 2 4 3 1 2 2 0
H2 6 14 14 13† 15† 10† 19 6 25 24
3 3 8 4 2
H3 (28) 37 36† 31† (24) 30† 26† (27) 34† 35† (25) 34 33 (33) 40†
(4) (3) (3) 3 (4) 3 (3)

Table 2 Summary of parameters (bold numbers) and their standard deviation (values underneath) retrieved by three different
methods. 1=Results from modelling of the apparent incidence angle (Note that for stations BSI and PSI a fourth layer was introduced
during the grid search). 2=Results from waveform inversion. 3=Results from MZK. VS is given in [km/s], depths H in [km]. For
gradients, ∗ denote their slope [s−1 ] and † denote their mean depth. VS4 denotes mantle VS . H3 denotes Moho depth. In brackets:
values obtained from modelling of īP which are insufficiently resolved (compare section 6).

4.4 KOM 4.5 PSI

For all epicentral distances receiver functions for station Receiver functions at station PSI are complex although
KOM show coherent signals between 0 and 10 s (Fig. seismograms from 75 events could be included (Fig. 8A,
7A). A positive amplitude at 0 s indicates again imper- Tab. 1). A strong conversion from a very shallow inter-
fect seismogram rotation. The following two small posi- face indicated at 0.5 s is followed by a negative signal
tive signals probably originate from crustal discontinu- at ≈ 1 s. The positive phase at 4.5 s may correspond
ities. The large positive signal near 4 s indicates a strong to the conversion at the Moho. A clear identification of
discontinuity in the typical depth region of the Moho. its multiples is difficult. Thereafter, coherently signals
Multiples are found at 13 s (PpPs) and 17 s (PpSs). The appear at 9 and 10.5 s (positive) and at 12 s (negative).
pronounced negative signal at 7 s is probably the PpSs- Either the phase at 9 s or 10 s may be indicative of the
multiple of the phase at ≈ 2 s, although it cannot be slab, as the top of the subducted plate is found at ap-
excluded that it is a direct phase from a deep struc- proximately 100 to 150 km.
ture. The PpPs-multiple of this phase interferes with Three layers are insufficient to model V̄S,app (T ) (Fig.
the direct Moho conversion. Similar to station IPM, 8B,C) because a minimum is observed at T = 4 s indi-
the model obtained from VS,app (T ) (Fig. 7B,C) starts cating a zone of decreased VS . Therefore a fourth layer
at high VS = 3.5 kms−1 for the upper 10 km followed was introduced. The resulting density-function shows
by VS = 4.2 kms−1 within the lower crust. The Moho is a 1 km thick layer with VS =2.9 kms−1 . Then, the VS
indicated at a depth of 27 km. increases at a sharp discontinuity to the high value of
Waveform inversion (Fig. 7D) confirms VS = VS = 4.2 kms−1 followed by a decrease at 6 km depth to
3.5 kms−1 at the surface but shows the depth of the VS = 4.0 kms−1 . The third discontinuity indicates the
intercrustal discontinuity slightly deeper at 12 km. The Moho at a depth of ≈ 33 km.
second layer with VS = 3.9 kms−1 is imaged down to The VS -depth model obtained from waveform inver-
a depth of ≈ 30 km. The Moho is modelled as a gra- sion (Fig. 8D) starts at the surface with VS =3.1 kms−1 .
dient between 30 and 35 km with a velocity increase The transition to high VS of the next layer seems also
of 0.14 kms−1 per km and an additional velocity jump rapid. The mean velocity between 2 and 10 km depth is
at 35 km. Below, VS = 4.8 kms−1 indicates rather high 4.1 kms−1 . Again, a velocity inversion is found between
S-wave-velocity in the mantle. 10 and 25 km depth, where VS ≈ 3.9 kms−1 . Velocities
The MZK (Fig. 7G) confirms depth of the crustal increase again at 25 km depth to 4.2 kms−1 and at a
discontinuity and of the Moho at 12 − 13 km and at depth of 40 km by 0.1 kms−1 per km. Mantle velocities
33 − 40 km, respectively. In both cases, VP /VS ≈ 1.8 of VS =4.7 kms−1 are reached at 45 km. However, a low
although resolution is low. velocity zone is again indicated in the mantle for depth
 13

0.5 1 3 4.5 9 10.5 12

PSI
7 0
SUM−PpSs (A) (B) (C)
SUM−PpPs
SUM−Ps 6
95−100°,1
10
90−95°,1
85−90°,2

VS,app [km/s]
5 1.0

depth [km]
80−85°,14 20

relative frequency
75−80°,3
70−75°,3 0.8
65−70°,3 4
60−65°,7 30 0.6
55−60°,4
50−55°,8 3 0.4
45−50°,9 40
40−45°,3 0.2
35−40°,8 2
30−35°,2 0.0
50
0 10 20 0.5 1 2 5 2 3 4 5
time [s] low−pass parameter T[s] VS [km/s]

0 0 0
(D) (E) (G)

delay time [s] for the maximal vP/vS

10
20 10
30
2
depth [km]

depth [km]

40 20
50
60 (F) 30
4
70
80 40
90
100 50 6
2 3 4 5 6 −5 0 5 10 15 20 25 30 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3
VS [km/s] time [s] vP/vS with vS1=3.9, vS2=4.2

0.0 0.1 0.2 0.3 0.4 0.5 0.1 0.2 0.3 0.4 0.5 0.6 0.7
relative frequency semblance

Fig. 8 Analysis of receiver functions computed from data recorded at station PSI. For MZK the thickness of the uppermost layer is
fixed at 6 km. See with Figs. 3 and 5 for more details.
14

around 55 km. It should be noticed that standard de- The VS,app (T ) curves could not be sufficiently mod-
viations are rather high for most part of the obtained elled with only three layers. Therefore, a fourth layer
model. Therefore reliability of results from the wave- was introduced. The results (Fig. 10C) indicate a thin
form inversion is limited. top layer of 1 km thickness with VS =2.8 kms−1 . In the
Due to the complexity of the receiver functions, the following layer VS = 3.4 kms−1 down to 6 km depth.
MZK algorithm is not able to resolve all of these details Below this we find VS = 4.2 kms−1 down to a depth of
(Fig. 8G). Essentially, only the discontinuity observed 28 km.
by waveform inversion at 25 km depth is indicated in the Due to the lack of clear multiples, waveform mod-
semblance map. The more shallow discontinuity may be elling at this station is difficult (Fig. 10D). The density-
weakly suggested by the high semblance at 2 km depth. function of VS -depth models confirm the low velocity of
The velocity increase at great depth is not visible. 2.9 kms−1 near the surface and a rapid increase in VS
to VS =3.9 kms−1 on average at a depth of 6 km. Within
the following layer between 14 and 37 km depth the
4.6 GSI mean velocity is 4.1 kms−1 , but rather high uncertain-
ties are observed. The mantle velocity of 4.8 kms−1 is
The receiver functions of station GSI show pronounced reached below a discontinuity at 37 km depth. The pro-
conversions at 2 s but also conversions with smaller am- nounced phase at 10 s is modelled by a clear velocity in-
plitude at 4, 5, 6.5, and 14 s. The signal at 6.5 s could crease at 95 km depth. Although this phase, which can
be the PpPs multiple of Ps at 2 s. The signals at 14 and probably be associated with a direct conversion from
18 s could be the PpPs and PpSs multiples of Ps at 4 s. the slab (see next section), shows azimuthal variations,
Resulting VS,app (T ) curves (Fig. 9B,C) reveal VS = we decided to include receiver functions of all azimuths
2.6 kms−1 for the uppermost 5 km. Velocity and thick- in the stack, as the slab is found in great depth and
ness of the second layer are not well defined but a does not have influence on the results for the shallower
velocity gradient seems to exist between 5 and 20 km layers.
with 0.16 kms−1 per km. During waveform inversion The MZK shows regions of high semblance between
(Fig. 9D) stacking was limited to receiver functions with 2 and 4 km and between 6 and 8 km depth confirm-
backazimuth between 30◦ and 60◦ , because azimuthal ing the increases in velocity found before. Highest sem-
variations of receiver functions are observed (see section blance are attained between 12 and 17 km depth. How-
5). It confirms the velocity of the upper layer as well as ever, this zone of high semblance is probably caused by
the velocity gradient that leads to VS = 4.7 kms−1 at a misinterpretation of the phase at 10 s as a multiple.
a depth of only 20 km. However, the uppermost layer We have reasonable evidence that this phase is a direct
appears thicker than observed from VS,app (T ). Between phase. This will be discussed in the next section. Here,
20 and 40 km depth the model shows a velocity inver- the location of the Moho is not resolved by MZK.
sion. On average VS decreases from 4.7 kms−1 at 20 km
VS to only 4.3 kms−1 at 30 km depth. However uncer-
tainties are rather high in this part of the model (up to 5 Structures related to the subduction
0.75 km/s). Upper mantle velocities of VS = 4.7 kms−1
are reached again at 40 km depth. Due to their proximity to the Sunda trench we pay spe-
The MZK shows high semblance within a broad cific attention to inclined structures possibly related to
band between 2 and 18 km depth without resolution subduction at the stations BSI, PSI and GSI (Fig. 1A).
for the VP /VS . This is in accordance with the veloc- There, the slab is subducted at a low angle and still
ity gradient at this depth found before. Additionally, rests at relatively shallow depth of less than 150 km
a semblance maximum is found at 31 km depth where at PSI, 100 km at BSI and less than 50 km at GSI
VP /VS =1.84. This corresponds to the depth of the low- (Gudmundsson and Sambridge, 1998). Synthetic re-
velocity layer found by the waveform inversion. ceiver functions show that amplitudes and delay times
of phases generated by conversion at inclined struc-
tures are sinusoidal and 360◦ periodic with backaz-
4.7 BSI imuth (e.g. Cassidy, 1992). On the Q-component small-
est and largest amplitudes and delay times are expected
At station BSI the most prominent phases are the pos- in up-dip and in down-dip direction, respectively. Al-
itive signals at approximately 1, 5, and 10 s (Fig. 10A). though delay times are equal on the Q- and on the
However, their corresponding multiples are difficult to T-component, amplitudes on the T-component vanish
identify. and reverse in up-dip and down-dip directions. In down-
 15

2 4 5 6.5 14 18
GSI
7 0
SUM−PpSs (A) (B) (C)
SUM−PpPs
SUM−Ps 6 10
90−95°,1
85−90°,3
5

Vs,app [km/s]
1.0

depth [km]
80−85°,5 20

relative frequency
75−80°,5
70−75°,5 0.8
65−70°,6 4
60−65°,2 30 0.6
55−60°,2
50−55°,4 3 0.4
45−50°,8 40
40−45°,4 0.2
35−40°,4 2
30−35°,3 0.0
50
0 10 20 0.5 1 2 5 2 3 4 5
low−pass parameter T[s] Vs [km/s]

0 0 0
(D) (E) (G)

delay time [s] for the maximal vP/vS

10
20 10
30 2
depth [km]

depth [km]

40 20
50
4
60 (F) 30
70
80 40
6
90
100 50
2 3 4 5 6 −5 0 5 10 15 20 25 30 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3
Vs [km/s] time [s] vP/vS with vS1=2.64, vS2=3.6

0.0 0.1 0.2 0.3 0.4 0.5 0.1 0.2 0.3 0.4 0.5 0.6 0.7
relative frequency semblance

Fig. 9 Analysis of receiver functions computed from data recorded at station GSI. In D, stacking of receiver functions was limited
to those with backazimuths between 30◦ and 60◦ . In C and D a clear velocity gradient above 20 km depth followed by a pronounced
low-velocity zone (D) is visible. For MZK the thickness of the uppermost layer is fixed at 10 km. Notice the broad band of high
semblance near 30 km supporting the observations of the velocity gradient. See with Figs. 3 and 5 for more details.
16

1 5 10
BSI
7 0
SUM−PpSs (A) (B) (C)
SUM−PpPs
SUM−Ps 6 10
90−95°,1
85−90°,2

VS,app [km/s]
5 1.0

depth [km]
80−85°,3 20

relative frequency
75−80°,2
70−75°,4 0.8
65−70°,3 4
60−65°,6 30 0.6
55−60°,2
50−55°,9 3 0.4
45−50°,2 40
40−45°,1 0.2
35−40°,2 2
30−35°,1 0.0
50
0 10 20 0.5 1 2 5 2 3 4 5
time [s] low−pass parameter T[s] VS [km/s]

0 0 0
(D) (E) (G)
10

delay time [s] for the maximal vP/vS

20
10
30
2
40
50
depth [km]

depth [km]

20
60
70 4
(F) 30
80
90
100
40 6
110
120
130 50
2 3 4 5 6 −5 0 5 10 15 20 25 30 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3
VS [km/s] time [s] vP/vS with vS1=3.4, vS2=4.2

0.0 0.1 0.2 0.3 0.4 0.5 0.1 0.2 0.3 0.4 0.5 0.6 0.7
relative frequency semblance

Fig. 10 Analysis of receiver functions computed from data recorded at station BSI. For MZK the thickness of the uppermost layer is
fixed at 6 km. See with Figs. 3 and 5 for more details.
 17

ducting oceanic plate at ≈ 16 km beneath the outer-arc

high on which station GSI located.
On the T-components, an additional phase with a
similar behaviour is found at 4 s. There, amplitudes re-
verse at 60◦ backazimuth from positive in northern di-
rection to negative in south-east. This indicates an in-
clined interface dipping N60◦ E. On the Q-component,
this phase is generally positive. Phase identification is
however difficult in up-dip direction which might be
attributed to the low number of observations. This
phase can be associated to the second velocity increase
found by waveform inversion between depths of 35 and
40 km (Fig. 9D) and may correspond inclined structures
within the subducted slab.
At station BSI clear evidences for the subducting
slab can be found. A pronounced positive phase was
A Q−component B T−component
found before at 10 s (Fig. 10A). In north-east direc-
300−315°,2
285−300°,1
270−285°,1
255−270°,1
tions the delay times of this phase depend clearly on
240−255°,1
the distance of the trench to their piercing points at
180−195°,2
150−165°,1
135−150°,2
120−135°,1
100 km (Fig. 12A,B). This again indicates an inclined
105−120°,7
90−105°,2
75−90°,1
60−75°,2
interface with dip to the north-east. Fig. 12C,D shows
45−60°,3
30−45°,23
15−30°,2
0−15°,2
that the delay times of this phase vary strongly with
345−360°,1
330−345°,5
315−330°,4
300−315°,2
backazimuth. The highest delay time is found for the
0 10 0 10 60 − 75◦ backazimuth bin. The change in polarity is
time[s] time[s]
found on the T-component at N60◦ E constraining the
dip direction of the discontinuity. The negative phase
Fig. 11 Receiver functions at station GSI as a function of back- at 8 − 9 s preceding the positive phase indicates a low
azimuth. A Q-component, B T-component. Traces are corrected
velocity zone related to the slab. Yuan et al (2008)
for moveout, low-pass filtered at 1 Hz and stacked within bins of
15◦ backazimuth. Blue and red ticks mark phases generated at and Tibi et al (2008) suggested that such a low ve-
inclined structures. locity zone may correspond to the subducted oceanic
crust that has not yet eclogised. The following velocity
increase represents the Moho of the subducted plate.
On the other hand, Peacock (1993b) states that the
dip direction, amplitudes on the T-component change slab can transport large amounts of water to depths
clockwise from negative to positive polarity. of ∼ 100 km. Released water will migrate upward into
Fig. 11 shows the receiver functions of station GSI as the mantle wedge causing hydration reactions or partial
a function of backazimuth. Two important conversions melting (e. g. Peacock, 1993a,Kawakatsu and Watada,
in the receiver functions, possibly related to inclined 2007), causing decreased seismic velocities.
structures, are highlighted. For backazimuths between To estimate the dip of the interface, the delay
0◦ and 180◦ the first conversion is found with positive times of the signal near 10 s on moveout-corrected re-
amplitude on the Q-component at ≈ 2 s (Fig. 11A). At ceiver functions are converted to location of the pierc-
the same time on the T-component (Fig. 11B), phases ing points and their shortest distances to the trench
with positive amplitudes are found for north-eastern di- x. An upper mantle velocity of VP = 8.1 kms−1 and
rections and negative phases are found for south-eastern VP /VS = 1.8 is used to calculate relative depths of
directions. The transition from positive to negative am- conversion H. Hereby it is assumed that the slab is at
plitudes is found near 90◦ backazimuth indicating an 100 km depth at station BSI (see Fig. 1). The dip α of
inclined interface with dip in N90◦ E direction. This in- the inclined interfaces is retrieved assuming the linear
terface corresponds to the apparently gradual velocity relation
increase between 10 and 20 km inferred from waveform
inversion and can be associated to the top of the sub- H(x) = tan(α) x + H0 . (7)
ducting slab. The depth location is in agreement with
findings from reflection seismic surveys (Kopp et al, In (7) H0 is the depth of the interface at the piercing
2001). Franke et al (2008) image the top of the sub- point closest to the trench. The solutions and the re-
18

B
2.50 90
data A (Fig. 12 B)
A 2.48 LSQ−fit A
data B (Fig. 12 C/D)
2.46 95 LSQ−fit B

20

distance from trench [°]

6˚N 2.44

depth [km]
15 2.42
0 100
2.40
5.5˚N BSI
2.38
105
2.36
10
0

2.34
5˚N
2.32 110
240 250 260 270 280 290
2.30 distance from the trench [km]
95.5˚E
2.28 Fig. 13 Depth of conversions derived from the delay times ob-
served at station BSI. Delay times corresponding to data A and
0 10 20
B are marked in blue and red, respectively, in Fig. 12. For depth
time [s]
calculations VP = 8.1 kms−1 and VP /VS = 1.8 were assumed for
C Q−component D T−component the upper mantle. Regression lines correspond to a dip of 20◦
300−315°,3
(blue, data A) or 21◦ (red, data B).
255−270°,1

wave velocities the variations are negligible (variations

180−195°,1 smaller than the accuracy).
165−180°,3
150−165°,1 From the distance of station PSI to the trench, we
135−150°,3
120−135°,1 would expect to see a slab phase at this station as well.
105−120°,5
90−105°,5 There are strong signals in the receiver functions at
75−90°,1
60−75°,3 9 s and 10.5 s which may correspond to direct phases
45−60°,8
30−45°,19 from a structure as deep as the slab (≈ 100 km) at this
15−30°,2
station. However processing the receiver functions in a
similar way as for station GSI and BSI did not reveal
330−345°,2 any evidence for inclined interfaces. This may point to
315−330°,3
300−315°,3 a locally flat slab surface.

0 10 20 0 10 20
time[s] time[s]
6 Discussion
Fig. 12 Receiver functions at station BSI. A Piercing points
(red points) corresponding to the conversions near 10 s in receiver 6.1 Potential and impact of combining the methods
functions (B). B Receiver functions as a function of distance to
the trench. Traces are bandpass filtered (2 - 30 s) and corrected P-to-S converted phases show complexity of the
for moveout. Delay times of the signal near 10 s (blue ticks) in-
lithosphere underneath stations located on the Sunda
crease with distance. C, D Q- and T-components of the receiver
functions in Fig. 10 stacked within 15◦ wide backazimuth bins. plate. To analyse receiver functions we combined three
The azimuthal dependence of arrival times of the phase near 10 s different methods. They include the modelling of
(red ticks) and polarity reversal on the T-components suggest a VS,app (T ), waveform inversion, and MZK which is a
N60◦ E dipping interface.
modification of the algorithm after Zhu and Kanamori
(2000). Their combination results in an improved de-
finition of VS vs. depth. Studies of synthetic receiver
functions show that VS of the uppermost layer equals
sulting uncertainties are found in a linear least square
VS,app (T ) for low T where VS,app (T ) is constant. Depth
sense (Menke, 1989).
resolution depends on data quality at long periods.
From the single receiver functions in Fig. 12B a dip The modelling of VS,app (T ) provides stable velocities
of α = 20◦ ± 9◦ at a depth of H0 = 99 km is found (see and depth estimates for the upper layers of the model.
data A in Fig. 13). Using the stacked receiver func- The resolution for discontinuities deeper than 20 km de-
tions in Fig. 12C, a similar dip angle α = 21◦ ± 8◦ is creases because VS,app (T ) tend to diverge for high val-
obtained (see data B in Fig. 13). Assuming other P- ues of T due to limited data quality at long periods.
 19

This reduces the upper limit of T for the calculation of VS,app (T ) needs high quality data even at long periods.
the misfit down to 7 s at some stations. As a result the Furthermore complex structures are difficult to model,
depth of the Moho is often underestimated compared to as a high number of layers is needed and the results
the results of the other methods used (compare Tab. 2). may even become non-unique. The waveform inversion
Additionally, there is little resolution for upper mantle turned out to be meaningless for only one single in-
velocities which requires data with high signal-to-noise version. Only a statistic over a lot of trials with dif-
at long periods. Therefore, we base the discussion of ferent initial models provides a sufficiently stable mean
Moho depth and mantle velocities mainly on the re- model. Its standard deviation serves as an estimate for
sults obtained by waveform inversion and MZK. For uncertainties. However this method still makes some
synthetic data, VS,app (T ) is stable even for larger T . approximations which are not correct in reality. First
The solutions are also affected by the parameterisation we fixed the densities in the crust according to Birch’s
of the model in the grid search approach. The data law. Mantle densities are fixed to 3.3 g/cm3 . Second the
sometimes demand up to four layers to increase the VP /VS -ratio is fixed to 1.73 in the crust and 1.79 in the
resolution in VS and in depth. mantle. For the MZK we tried to visualise uncertainties
The obtained VS -depth models provide reference by the 75% contour line of the semblance maximum. It
models for subsequent waveform inversions. During turned out that plotting curves correspondent to delay
processing, we note that single waveform inversions are times of the direct conversion and the multiples facili-
non-unique. However, a large number of inversions with tates understanding of the plot.
different initial models results in convergence to sta-
ble final VS -depth models at greater depths. They con-
6.3 Crust and upper mantle structures in Indonesia
firm the results of the VS,app (T ) modelling for shallow
and Malaysia
depths.
Retrieved shear wave velocities are used as input Stations on the Malaysian peninsula (KUM, IPM,
to MZK. For simple receiver functions, MZK provides KOM) all show relatively high near-surface VS of 3.3 −
good definitions of the depth limits of discontinuities. 3.6 kms−1 which are typical for plutonic rocks (Schön,
For sharp discontinuities also VP /VS can be retrieved. 1996) exposed in the region (Pubellier et al, 2005).
However, VP /VS cannot be resolved in the case of veloc- These three stations show one major intracrustal dis-
ity gradients. This poses problems to the application of continuity where VS increases to 4.0 − 4.2 kms−1 at
the algorithm in general, as natural discontinuities are depths of 12 km (KOM, ≈ 2 s arrival time), 14 km (IPM,
often formed by gradients. Additionally, we find that ≈ 3 s), and 19 km (KUM, ≈ 3 s). It seems to correspond
for complex receiver functions, individual phases may to a Conrad discontinuity separating the upper from the
be misinterpreted which leads to obscured discontinu- lower crust. The Moho discontinuity is a sharp disconti-
ities. nuity only at station KUM at 34 km depth, represented
In this study receiver functions are analysed for 6 by a phase at ≈ 4 s. At station IPM and KOM it forms
stations on the Sunda plate (compare Tab. 2). The com- a gradient between 25 − 35 km (≈ 3 s arrival time) and
bination of modelling of VS,app (T ) with waveform in- at 30 − 38 km (≈ 4 s) depth. Obtained Moho depths are
version provides reasonable constraints for the crustal in accordance with the crustal thickness of 30 − 35 km
VS -depth models. The application of MZK reveals the found by Wu et al (2005). VS in the crust is on aver-
depth of the Moho. Waveform modelling was also ap- age near to the values given by Wu et al (2005). The
plied to constrain VS for the upper mantle. Caused by VS velocities found for the uppermost mantle in this
the presence of gradient zones instead of sharp discon- study are higher by 0.1 − 0.3 kms−1 than the velocities
tinuities or missing of clear multiple conversions, VP /VS given by Wu et al (2005). It should be noted however
is generally not well resolved. that in the surface wave tomography of Wu et al (2005)
It should be emphasised, that all methods used in the typical resolution length is on the order of several
this study are local methods. They can be applied to 100 km, while it is on the order of a few 10 km for the
single stations where dense networks are missing. receiver function study presented here.
The stations on Sumatra (BSI, GSI, PSI) generally
indicate more complex structures. They all show signif-
6.2 Methodological limits and approximations icantly lower near-surface VS in the range of 2.6 kms−1
at GSI and 2.9 kms−1 at PSI. At station PSI the low
The combination of the different methods balances velocities at the surface are immediately followed by
weaknesses of the single methods and misinterpreta- high VS velocity of about 4.1 kms−1 down to 10 km
tions can be corrected. We found that modelling of depth and a complex layer with decreasing velocities
20

down to about 3.5 kms−1 in 25 km depth where veloc- pare Crow and Barber, 2005). Below, VS increases to
ity increases abruptly. This unusual structure may be 3.8 kms−1 at a depth of 6 km, indicating basement
explained by the proximity of the station to the Toba rocks. VS increases to 4.1 kms−1 at 14 km depth. Similar
caldera. Partial melting underneath the caldera due to to the stations on the Malaysian peninsula this corre-
active volcanism (Masturyono et al, 2001) seems to re- sponds to the transition to the lower crust. This obser-
sult in reduction of shear wave velocities, explaining vation is mainly supported by waveform inversion and
the decreasing velocities in the middle crust. The ab- MZK. The Moho is found at a depth of 37 km, which is
solute value of VS in the depth range from about 2 km similar to station PSI and causes a phase at the rather
to about 10 km are very high compared to normal con- high delay time of ≈ 5 s in the receiver functions. Evi-
tinental upper crust making a mainly felsic composition dences for inclined structures related to subduction are
unlikely (Christensen and Mooney, 1995). In the lower also found at BSI. A discontinuity displayed by a phase
crust below 25 km depth VS increases again to normal around 10 s and dipping at 20◦ ± 8◦ to the north-east
values of 4.2 kms−1 , a velocity increase which is visible indicates the subducting oceanic slab at 90 − 110 km
in the receiver functions by a phase at 3 s. Wu et al depth and confirms earlier results (Gudmundsson and
(2005) find low crustal velocities to the Northwest of Sambridge, 1998; Widiyantoro and van der Hilst, 1996).
PSI more or less halfway to station BSI but average
values for the region around PSI. The Moho is formed
by a gradual velocity increase between 35 and 45 km
depth, causing a phase at ≈ 4.5 s. This is in accordance
with the deep Moho at ≈ 38 km found by Sakaguchi
et al (2006) in the area. A low velocity zone in the man- 6.4 Comparison to other island arcs
tle around 55 km depth may also be caused by partial
melting in the upper mantle. At station GSI VS is low
The structure beneath the stations in Malaysia shows
within the upper 5 km, indicating a thick sedimentary
a clear separation between upper and lower crust and
layer. This seems reasonable since GSI is situated on
overall crustal thicknesses between 30 to 35 km com-
the island Nias which is part of the accretionary wedge
parable to continental average structure. The revealed
formed by marine sediments scraped-off the subducting
models beneath the stations in Indonesia show over-
oceanic plate. The gradual velocity increase between
all more heterogeneity, a thicker crust and high veloci-
10 and 20 km probably images the top of subducting
ties in the upper and lower crust. These characteristics
oceanic plate and is represented in the receiver func-
are typical for island arc models (Rudnick and Foun-
tions by a broad phase around 2 s. Azimuthal depen-
tain, 1995; Christensen and Mooney, 1995). However,
dence of delay times of converted phases and polarity
the crustal VS velocities are high in comparison to other
reversal on the T-components provide evidence for dip
island arcs and indicate an enhanced mafic contribution
of the subduction to the north-east. This is in accor-
to these layers (Rudnick and Fountain, 1995). However,
dance with the direction of subduction at the Sunda
without knowledge of VP and density structure a de-
Arc and recent observations (Kopp et al, 2001; Franke
tailed association with specific rock types is not possi-
et al, 2008; Singh et al, 2008). Within the slab another
ble. The crustal structures found here are quite compa-
north-east dipping structure is indicated. For the cor-
rable to the one observed by Azano et al (1985) for the
responding observed converted phases the dip is also
Japanese Archipelago close to Izu Peninsular. They de-
expressed in azimuthal variation of times and their po-
termined P-velocities but S-velocities are comparable if
larity reversal on the T-components (Fig. 11). In both
VP /VS -ratios of 1.73 and 1.78 are assumed for the crust
cases an exact quantification of the dip is difficult due
and mantle, respectively. They observed VP ≈ 6.0 km/s
to interference of different phases in receiver functions.
(accordingly VS ≈ 3.5 km/s) for the upper crust and
It follows a low-velocity zone near 30 km depth and a
VP ≈ 6.8 km/s (accordingly VS ≈ 3.9 km/s) for the
velocity increase at 35 − 40 km depth, causing a phase
lower crust. Both values compare well to the S-velocities
at ≈ 4 s delay time. This low-velocity zone might be in-
of 3.3−3.6 km/s (upper crust) and 4.0−4.2 km/s (lower
dicative of serpentinisation due to water intruding the
crust) found for the Malaysian stations and BSI in this
oceanic plate through faults related to bending of the
study. Additionally, Azano et al (1985) also observed
slab (Ranero et al, 2005; Hacker et al, 2003). The sub-
the Conrad discontinuity at a depth of 15 to 20 km
sequent positive conversion indicates a velocity increase
beneath Honshu and the Moho at the relatively great
and the lower limit of the serpentinised zone.
depth of 35 − 40 km. Therefore, the characteristics of
At station BSI we find VS = 2.8 kms−1 at shal- the crustal structure of the Japanese island arc and the
low depth corresponding to sedimentary rocks (com- Sunda arc show similarities.
 21

7 Conclusions on Sumatra and Nias show more complex crustal struc-

tures. A division in upper and lower crust as for the
Receiver functions are often stacked to increase the sig- Malaysian stations is not identified. In contrast differ-
nal to noise ratio with respect to single traces. How- ent low velocity zones are observed. Close to the trench
ever, the summation may obscure and broaden wave- those may be due to serpentinisation of the subduct-
forms mainly of multiply converted phases when a large ing slab. At distances of 120 to 150 km from the trench
range of epicentral distances are included. As a result low S-velocities may be related to partial melting of
converted phases generated at sharp discontinuities and crustal material and to volcanic activity. Additionally,
at gradients may have similar behaviour. the Moho is found at greater depths for stations close to
The combination of the different methods to analyse the trench (35−45 km) than for the Malaysian stations.
receiver functions, i.e. the modelling of VS,app (T ), wave- Receiver functions of station GSI, located closest to
form inversion and MZK, results in VS -depth models the trench, show evidences for north-east dipping struc-
and resolved non-uniqueness of the individual methods. tures formed by subduction. Here, the top of the sub-
We propose that VS is initially retrieved by modelling ducting plate is found between 10 to 20 km depth. Near
of VS,app (T ) and used as input to waveform inversion. station BSI the subducting slab was observed. Here, a
The resulting VS -depth model is delivered to our mod- dip of 20◦ ± 8◦ to the north-east at depths between 90
ification of the common grid search approach by Zhu and 110 km was retrieved from delay times of the Ps-
and Kanamori (2000) to compute VS /VP . converted phase. We conclude that receiver functions
To estimate the depth of discontinuities and can be used to observe the slab and even to retrieve the
VP /VS the common grid search approach by Zhu and local inclination of the slab surface. Again, this provides
Kanamori (2000) was modified (MZK). The grid search a mean to infer information about the slab in regions
approach by Zhu and Kanamori (2000) was extended to where seismic stations are too isolated for the applica-
multiple layers. Semblance analysis was applied to re- tion of other geophysical methods such as tomography.
trieve model parameters. Constraining VS instead of VP
increases resolution. In case of velocity gradients sem- Acknowledgements The considered seismic stations are op-
blance is high over a broad depth range and insensitive erated by GEOFON (GFZ German Research Centre for Geo-
to VP /VS . This may help to discriminate between sharp sciences), Malaysian National Seismic Network, Pacific21, and
seismic discontinuities and gradients. the Japan-Indonesia Seismic Network. Data were obtained from
GEOFON and IRIS. We thank W. H. Geissler, V. Schlindwein
We conclude that modelling VS,app is a useful tech- and an anonymous reviewer for their comments that helped to
nique to reveal local absolute S-velocities. This can be improve the manuscript.
especially useful at isolated stations where no other in-
formation are available. Furthermore, we found that the
MZK significantly improves the widely used algorithm
of Zhu and Kanamori (2000). Therefore, future receiver References
function studies may benefit from the proposed modi-
fications. Ammon CJ (1991) The isolation of receiver function
The crustal and upper mantle seismic structure of effects from teleseismic P waveforms. Bull Seismol
the Sunda plate is only partly constrained from previ- Soc Am 81:2504–2510
ous studies. We study the shear wave structure under- Ammon CJ, Randall GE, Zandt G (1990) On the
neath six permanent seismic broadband stations along nonuniqueness of receiver function inversions. J Geo-
the Sunda Arc on Nias and Sumatra and on Malaysia phys Res 95:15,303–15,318
by means of receiver functions. The analysis of re- Azano S, Wada K, Yoshii T, Hayakawa M, Misawa Y,
ceiver functions reveals consistent local crustal and up- Moriya T, Kanazawa T, Murakami H, Suzuki F, Kub-
per mantle VS models for the different stations. High ota R, Suyehiro K (1985) Crustal structure in the
shear wave velocities near the surface for stations on northern part of the Phillippine Sea plate as derived
Malaysia are indicative of crystalline rocks. At near- from seismic observations of Hatoyama–off Izu Penin-
trench stations on Sumatra low near-surface VS corre- sula explosions. J Phys Earth 33:173–189
spond to thick sedimentary layers. Baroux E, Avouac JP, Bellier O, Sébrier M (1998) Slip-
In general stations on Malaysia reveal crustal struc- partitioning and fore-arc deformation at the Sunda
tures comparable to other island arcs (compare Miller Trench Indonesia. Terra Nova 10:139–144
and Christensen, 1994), i. e. an upper crust divided from Birch F (1961) The velocity of compressional waves
a lower crust by the Conrad discontinuity and the Moho in rocks to 10 kilobars, part 2. J Geophys Res
in relative great depth between 30 and 38 km. Stations 66(7):2199–2224
22

Cassidy JF (1992) Numerical experiments in broad- Seismic Network (GRSN). Geophys J Int 121:191–
band receiver function analysis. Bull Seismol Soc Am 202
82:1453–1474 Kopp H, Flueh ER, Klaeschen D, Bialas J, Reichert C
Chang SJ, Baag CE, Langston CA (2004) Joint inver- (2001) Crustal structure of the central Sunda margin
sion of teleseismic receiver functions and surface wave at the onset of oblique subduction. Geophys J Int
dispersion using the Genetic Algorithm. Bull Seismol 147:449–474
Soc Am 94:691–704 Kosarev G, Makeyeva LI, Vinnik LP (1987) Inversion of
Christensen NI, Mooney WD (1995) Seismic velocity teleseismic P-wave particle motion for crustal struc-
structure and composition of the continental crust: ture in Fennoscandia. Phys Earth Planet Inter 47:11–
A global view. J Geoph Res 100(B7):9761–9788 24
Crow MJ, Barber AJ (2005) Map: Simplified geological Krueger F, Scherbaum F, Rosa J, Kind R, Zetsche F,
map of Sumatra. vol 31, Geological Society, London, Hoehne J (2002) Crustal and upper mantle struc-
Memoirs ture in the Amazon region (Brazil) determined with
Franke D, Schnabel M, Ladage S, Tappin DR, Neben broadband mobile stations. J Geophys Res 107:2265,
S, Djajadihardja YS, Müller C, Kopp H, Gaedicke C DOI 10.1029/2001JB000598
(2008) The great Sumatra-Andaman earthquakes – Küpper FJ (1958) Theoretische Untersuchungen über
Imaging the boundary between the ruptures of the die Mehrfachaufstellung von Geophonen. PhD thesis,
great 2004 and 2005 earthquakes. Earth Planet Sci Johannes-Gutenberg-Universität Mainz
Lett 269(1–2):118–130 Masturyono, McCaffrey R, Wark DA, Roecker SW,
Gill J (1981) Andesites and Plate Tectonics. Springer, Ibrahim G, Sukhyar (2001) Distribution of magma
New York beneath the Toba caldera complex, north Sumatra,
Gorbatov A, Domı́nguez J, Suárez G, Kostoglodov V, Indonesia, constrained by three-dimensional P wave
Zhao D, Gordeev E (1999) Tomographic imaging velocities, seismicity snd gravity data. Geochem Geo-
of the P-wave velocity structure beneath the Kam- phys Geosyst 2
chatka peninsula. Geophys J Int 137(2):269–279 Menke W (1989) Geophysical data analysis: discrete
Gudmundsson O, Sambridge M (1998) A regionalized inverse theory. Academic Press, New York
upper mantle (RUM) seismic model. J Geophys Res Metcalfe I (2000) The Bentong-Raub suture zone. Jour-
103:7121–7136 nal of Asian Earth Sciences 18(6):691–712, DOI
Hacker BR, Peacock SM, Abers GA, Holloway SD 10.1016/S1367-9120(00)00043-2
(2003) Subduction factory 2. Are intermediate- Miller DJ, Christensen NL (1994) Seismic signature
depth earthquakes in subducting slabs linked to and geochemistry of an island arc: A multidiscipli-
metamorphic dehydration reactions? J Geophys Res nary study of the Kohistan accreted terrane, north-
108(B1):2003, DOI 10.1029/2001JB001129 ern Pakistan. J Geophys Res 99(B6):11,623–11,642,
Hamilton W (1979) Tectonics of the Indonesian Region. DOI 10.1029/94JB00059
US Geol Surv Prof Pap 1048:345 Neidell NS, Taner MT (1971) Semblance and other
Hoang-Trong P, Behe L (1972) Extended distances and coherency measures for multichannel data. Geophys
angles of incidence of P-waves. Bull Seismol Soc Am 36(3):482–497
62(4):885–902 Peacock SM (1993a) Large-scale hydration of the
Jacobsen BH, Svenningsen L (2008) Enhanced unique- lithosphere above subducting slabs. Chemi-
ness and linearity of receiver function inversion. cal Geology 108(1-4):49–59, DOI 10.1016/0009-
Bull Seismol Soc Am 98(4):1756–1767, DOI 2541(93)90317-C
10.1785/0120070180 Peacock SM (1993b) The importance of blueschist
Julià J, Vila J, Macià R (1998) The receiver structure → eclogite dehydration reactions in subduct-
beneath the Ebro basin, Iberian peninsula. Bull Seis- ing oceanic crust. Geological Society of Amer-
mol Soc Am 88:1536–1547 ica Bulletin 105(5):684–694, DOI 10.1130/0016-
Kawakatsu H, Watada S (2007) Seismic Evidence for 7606(1993)105<0684:TIOBED>2.3.CO;2
Deep-Water Transportation in the Mantle. Science Pubellier M, Rangin C, Le Pichon X, Dotsea working
316(5830):1468–1471, DOI 10.1126/science.1140855 group (2005) Dotsea: A synthesis of deep marine data
Kennett B (1991) IASPEI 1991 Seismological Tables. in Southeast Asia. Mém Soc géol France 176
Research School of Earth Sciences, Australian Na- Ranero CR, Villasenor A, Morgan JP, Weinrebe
tional University, Canberra W (2005) Relationship between bend-faulting at
Kind R, Kosarev GL, Petersen NV (1995) Receiver trenches and intermediate-depth seismicity. Geochem
functions at the stations of the German Regional Geophys Geosystems 6(12)
 23

Ritzwoller MH, Shapiro NM, Barmin MP, Levshin AL Tibet from a seismological passive source experiment.
(2002) Global surface wave diffraction tomography. J J Geophys Res 102, B 12:27,491–27,500
Geophys Res 107:2335, DOI 10.1029/2002JB001777 Yuan X, Sobolev SV, Kind R, Oncken O, Bock G, Asch
Rudnick R, Fountain D (1995) Nature and composition G, Schurr B, Graeber F, Rudloff A, Hanka W, Wyle-
of the continental crust: A lower crustal perspective. galla K, Tibi R, Haberland C, Rietbrock A, Giese
Reviews of Geophysics 33(3):267–309 P, Wigger P, Röwer P, Zandt G, Beck S, Wallace T,
Rüpke LH, Morgan JP, Hort M, Connolly JA (2004) Pardo M, Comte D (2008) Subduction and collision
Serpentine and the subduction zone water cycle. processes in the Central Andes constrained by con-
Earth Planet Sci Lett 223:17–34 verted seismic phases. Nature 408(6815):958–961
Sakaguchi K, Gilbert H, Zandt G (2006) Converted Zhu L, Kanamori H (2000) Moho depth variation in
wave imaging of the Toba Caldera, Indonesia. Geo- southern California from teleseismic receiver fun-
physical Research Letters 33 tions. J Geophys Res 105:2969–2980
Schön J (1996) Physical properties of rocks, Handbook
of geophysical seismic exploration, vol 18. Elsevier,
Oxford
A MZK: delay times of converted phases in the
Singh SC, Carton H, Tapponnier P, Hananto ND,
case of two layers over a halfspace
Chauhan APS, Hartoyo D, Bayly M, Moeljopranoto
S, Bunting T, Christie P, Lubis H, Martin J (2008) In the case of two layers over a halfspace delay times of P-to-
Seismic evidence for broken oceanic crust in the 2004 S-converted phases are calculated and used in the MZK. The
Sumatra earthquake epicentral region. Nature Geo- following formulas which are an extension to Eqs. (6) apply:
0 s −2 1
science 1:777–781 q −2
tPs = (H1 − H2 )  VS,2 − p2 − − p2 A
V P,2
Subarya C, Chlieh M, Prawirodirdjo L, Avouac JP, VS,2
VS,2
Bock Y, andA J Meltzner KS, Natawidjaja DH, Mc- 0 s −2 1
Caffrey R (2006) Plate-boundary deformation asso- q −2
+ H1  VS,1 − p2 − − p2 A
V P,1
VS,1
ciated with the great Sumatra-Andaman earthquake. VS,1
Nature 440:46–51, DOI 10.1038/nature04522 0 s −2 1
q −2
tPpPs = (H1 − H2 )  VS,2 − p2 + − p2 A
Svenningsen L (2007) Teleseismic receiver functions – V P,2
VS,2
Method development and application . PhD thesis, VS,2
0 s −2 1
University of Aarhus q −2
Svenningsen L, Jacobsen BH (2007) Absolute S- + H1  2
VS,1 − p +
VP,1
VS,1 −p A2
VS,1
velocity estimation from receiver functions. Geo- q −2 q −2
phys J Int 170(3):1089–1094, DOI 10.1111/j.1365- tPpSs = (H1 − H2 ) VS,2 − p2 + 2H1 VS,1 − p2 . (8)
246X.2006.03505.x
Quantities VP,i , VS,i , Hi denote velocities and depth of the lower
Tibi R, Wiens DA, Yuan X (2008) Seismic evidence
boundaries, respectively, for layer i.
for widespread serpentinized forearc mantle along
the Mariana convergence margin. Geophys Res Lett
35(13):L13,303, DOI 10.1029/2008GL034163
Widiyantoro S, van der Hilst R (1996) Structure and
evolution of lithospheric slab beneath the Sunda
Arc, Indonesia. Science 271(5255):1566–1570, DOI
10.1126/science.271.5255.1566
Wiechert E (1907) Über Erdbebenwellen. Teil I:
Theoretisches über die Ausbreitung der Erdbeben-
wellen. Nachrichten von der Königlichen Gesellschaft
der Wissenschaften zu Göttingen, Mathematisch-
Physikalische Klasse
Wu HH, Tsai YB, Lee TY, Lo CH, Hsieh CH, Toan DV
(2005) 3-d shear wave velocity structure of the crust
and upper mantle in South China Sea and its sur-
rounding regions by surface wave dispersion analysis.
Marin Geophys Res 25(1-2):5–27
Yuan X, Ni J, Kind R, Sandvol E, Mechie J (1997)
Lithospheric and upper mantle structure of southern