Auxiliary Material for Paper 2009JB007021
C. E. Acton and K. Priestley
Bullard Laboratories, University of Cambridge, Cambridge, United Kingdom.
V. K. Gaur
Indian Institute of Astrophysics, Bangalore, India.
S.S. Rai
National Geophysical Research Institute, Hyderabad, India.
Complete citation: Acton, C. E., K. Priestley, V. K. Gaur, and S. S. Rai, (2010),
Group velocity tomography of the Indo-Eurasian collision zone, J. Geophys. Res., 115,
B12335, doi:10.1029/2009JB007021.
Introduction
The auxiliary file contains seven figures giving additional details of the surface wave
analyses.
Figure S1 gives an example of data processing and group velocity dispersion
measurement. Figure S2 shows the effect of varying the a priori smoothness parameter in
the tomographic inversion for the group velocity dispersion maps. The checkerboard
resolution tests for Rayleigh waves are shown in the paper. Here we include additional
resolution tests. Figure S3 shows the Love wave checkerboard tests and Figure S4 gives
the Rayleigh wave synthetic plate tests which illustrates how well isolated features are
resolved in different parts of the region. Figure S5 shows the results of a correlation tests
between group velocity dispersion data derived from CRUST2.0/AK135 and this study.
Figure S6 shows the results of a correlation tests between Ritzwoller and Levshin [1998]
Rayleigh wave group velocities and the Rayleigh wave group velocities determined in
this study. Figure S7 shows a depth resolution tests for inversion of group velocity
dispersion for shear velocity. A detailed description for each figure is given in the
accompanying caption.
2009jb007021-fs01.pdf
Figure S1. Group velocity dispersion curves were calculated for individual sourcereceiver paths using the multiple filter taper analysis (MFT) of Herrmann and Ammon
[2004]. In this procedure the data are filtered by a set of narrowband Gaussian filters
centered at discrete frequencies and a group velocity-period map is constructed from the
arrival times of the maximum amplitudes at the center frequency of each filter. (a)
Vertical component seismograms for a magnitude 5.6 event recorded at Hyderabad from
a distance of 2177 km. The top trace shows the raw vertical component prior to removal
of the instrument response, followed by the filtered and decimated vertical component,
the isolated fundamental mode after phase match filtering and the residual after removal
of the fundamental mode. The fundamental mode Rayleigh wave is picked out in red. (b)
The corresponding group-velocity diagram as displayed by the MFT graphical interface,
colored and contoured according to the amplitude of the seismogram at different
frequencies. The fundamental mode can be picked out along a ridge in the contours.
Dispersion is measured on the displacement trace.
2009jb007021-fs02.pdf
Figure S2. Effect of varying the a priori smoothness parameter (apsl) which is a limit on
the standard deviation of the slowness across a 10° reference distance. (a)–(c) show 10 s
Rayleigh wave group velocity maps for a range of values of apsl with a corresponding
map of the estimated error from the tomography, (d)–(f), plotted below each one. The
scale for group velocities shown in (a)–(c) is given by (g), and (h) plots the relationship
between apsl and the sum of the squares of the residual fits between observed and
synthetic travel times. The red diamond marks the chosen value of apsl for 10 s period
which can be seen to lie on the corner of the curve although the precise value is chosen
subjectively with regard to the physical reasonableness of the results, the correspondence
to known features and the presence of artifacts such as streaking and speckling.
2009jb007021-fs03.pdf
Figure S3. Checkerboard tests for 20 s Love wave group velocity maps with cells of
three different sizes: (a) 5°, (b) 7.5°, (c) 10°. The corresponding recovery maps produced
by inverting synthetic travel times through the checkerboard model for the same paths as
those used in the tomographic inversion of real data are shown below. The recovered
velocities suffer from the poorer path coverage of the Love wave dataset compared to the
Rayleigh wave dataset and a 5° checkerboard pattern is not well resolved except for the
regions of highest path density in the northeastern Himalaya, southeastern Tibet and Iran.
However 7.5° cells are reasonably well recovered across continental regions with some
smearing across the Indian Shield and 10° cells are very well recovered throughout the
region except for the Arabian Sea.
2009jb007021-fs04.pdf
Figure S4. Plate tests for the 20 s Rayleigh wave group velocity map with a 3° low
velocity square cell positioned over (a) the northeastern Himalaya, (b) the north central
Indian Shield and (c) the east Indian coast next to the Bay of Bengal. These tests have
been performed to address the limitations of standard checkerboard tests in testing for
directional bias in the recovery of velocities or the production of velocity anomalies in
the tomographic inversion as neighboring cells can obscure the effects. Each cell has a
velocity of 2.5 km s-1 within a 4.0 km -1 homogeneous background. The method used is in
every other way identical to the checkerboard tests included in the main body of the
paper. The corresponding recovery maps produced by inverting synthetic travel times
through the velocity model for the same paths as those used in the tomographic inversion
of real data are shown directly below, (d)–(f). In all cases a 3° cell can clearly be resolved
with limited smearing. Amplitude recovery is best in d) where 99% of the velocity is
recovered in the center of the cell and slightly poorer in e) and f) where recovered
amplitudes are 10% to 15% too high in the center of each cell.
2009jb007021-fs05.pdf
Figure S5. Correlation tests between group velocity dispersion data derived from
CRUST2.0/AK135 and this study. Group velocities were calculated from the
CRUST2.0/AK135 model by extracting the shear velocity structure at points on a 2° grid
within a 20° spherical cap, centered on 22.5° N, 77.5° E, and creating synthetic
dispersion curves. Results for periods 10, 15, 20, 30, 50 and 70 s are plotted with group
velocities from this study on the x axis. A linear regression line is plotted in red, with the
gradient written above each plot. The Pearson's correlation coefficient, r, for the dataset
is also noted. Pearson’s correlation coefficient, r, ranges from +1, signifying a perfect
positive linear relationship, -1, signifying a perfect negative linear relationship. At shorter
periods, there is a great deal of scatter in the data. The majority of this scatter comes from
large errors in our measurements of group velocities across the Indian Ocean at 10 and 15
s due to poor path coverage at these periods. A subset of the data, containing only
continental measurements and defined by a 10° radius spherical cap centered on 30° N,
77.5° E, is plotted for periods 10 and 15 s. This subset is used to calculate the linear
regression line and Pearson's r, but the remainder of the dataset is plotted as pale gray
crosses. Measurements plotted within the blue-dashed lines are all oceanic. It is clear that
if oceanic measurements are excluded there is a better correlation between
CRUST2.0/AK135 and our results at short periods, and that our values of group velocity
are too high in the oceans at 10 and 15 s. Remaining scatter at short periods is likely due
to the underestimation by CRUST2.0/AK135 of sediment thicknesses in areas such as the
Tarim Basin and overestimation of upper crustal velocities in Tibet (Figure 6). At long
periods there is good correlation between group velocities predicted by
CRUST2.0/AK135 and our measurements, CRUST2.0/AK135 gives consistently higher
group velocities, resulting in a decrease in the gradient of the linear regression line as
period increases. This is partly due to the fact that CRUST2.0/AK135 overestimates
average crustal velocities in tectonically active areas such as Tibet and underestimates
crustal thicknesses beneath the plateau. It must also be true that AK135 mantle velocities
are too high for this region.
2009jb007021-fs06.pdf
Figure S6. Correlation tests between Ritzwoller and Levshin [1998] Rayleigh wave
group velocities and this study. Group velocities are sampled from both maps for a
number of periods (20, 25, 30, 40, 50 and 70s) every 2° for a 20° spherical cap centered
on 22.5° N, 77.5° E and plotted with results from this study on the x axis. A linear
regression line is plotted in red, with the gradient written above each plot. The value of
Pearson's r for the data set is also noted.
2009jb007021-fs07.pdf
Figure S7. Resolution tests for a) a cylindrical low velocity zone of radius 500 km
centered on 32.5° E, 85° N and with velocities perturbed by 15% between 0 and 70 km
depth relative to a smoothed reference model based on AK135 and b) a cylindrical low
velocity zone at depth, centered on 34.5° E, 87.5° N, with radius 150 km and velocities
perturbed by 15% between 70 and 100 km depth. The models output from synthetic tests
for input a) and b) are shown respectively in c) and d). All four profiles run from 10° N,
87° E to 40° N, 88° E. Resolution tests were performed to test the recoverability of 3D
heterogeneities in the velocity model. The first test is for a cylindrical low velocity zone
of radius 500 km with velocities perturbed by 15% between 0 and 70 km depth relative to
the reference model, intended to simulate the slow crust of the Tibetan Plateau. Path
averaged velocity structures were calculated for each path through the model which were
then used to create a synthetic dispersion curve for each path. Synthetic dispersion data
were tomographically inverted in the same way as the real data to create group velocity
maps. The shear velocity structure beneath the profile shown in a) was then estimated
from inversion of group velocity dispersion curves extracted every 15 km along the
profile from the group velocity maps. The recovered velocities in b) reasonably
reproduce the input model in a). The base of the low velocity anomaly between 0 and 70
km, representing the slow Tibetan crust, is smeared out by approximately 20 km. The low
upper mantle velocities seen beneath Tibet in this study extend more than 40 km below
the base of the crust and are unlikely to be an artefact of the inversion process. Regions
where the reference model is unperturbed are generally reproduced to an accuracy of 3%
by the inversion. The second test c) is for a cylindrical low velocity zone of radius 150
km with velocities perturbed by 15% between 70 and 100 km. d) shows that a velocity
anomaly of this magnitude and lateral extent (~3°) at this depth and location is resolvable
but does incur a loss in amplitude and smearing.
Herrmann, R. B., and C. J. Ammon (2004), Computer programs in seismology, surface
waves, receiver functions and crustal structure, Saint Louis Univ., Saint Louis, Mo.
Ritzwoller, M. H., and A. L. Levshin (1998), Eurasian surface wave tomography: Group
velocities, J. Geophys. Res., 103(B3), 4839–4878, doi:10.1029/97JB02622.