See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/328792257

Shallow Ambient‐Noise 3D Tomography in the Concepción Basin, Chile:

Implications for Low‐Frequency Ground Motions

Article in Bulletin of the Seismological Society of America · October 2018

DOI: 10.1785/0120180061

CITATIONS READS

7 628

5 authors, including:

Gonzalo Andrés Montalva Diego Inzunza

University of Concepción University of Concepción
68 PUBLICATIONS 644 CITATIONS 1 PUBLICATION 7 CITATIONS

SEE PROFILE SEE PROFILE

Felipe Leyton German A Prieto

University of Chile National University of Colombia
71 PUBLICATIONS 1,170 CITATIONS 98 PUBLICATIONS 2,798 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

CYCLO The seismic cycle along subduction zones View project

Array processing for earthquake source analysis View project

All content following this page was uploaded by Gonzalo Andrés Montalva on 11 December 2018.

The user has requested enhancement of the downloaded file.

 Bulletin of the Seismological Society of America, Vol. XX, No. XX, pp. –, – 2018, doi: 10.1785/0120180061

Shallow Ambient-Noise 3D Tomography in the Concepción Basin,

Chile: Implications for Low-Frequency Ground Motions
by Diego A. Inzunza, Gonzalo A. Montalva, Felipe Leyton, German Prieto, and Sergio Ruiz

Abstract Forecasting surface seismic intensities is the main objective of much of

the research in seismic engineering. The seismic demand for tall buildings, bridges,
wind farms, and other existing slender structures is predominantly at low frequencies,
but these structures are usually outside the range of the design spectra prescribed by
codes. This article presents a 3D tomography model of the Concepción basin, Chile,
where the shear-wave velocity structure is estimated to 500-m depth. The maximum
basin thickness is approximately 160 m, but the average is about 100 m. The under-
lying bedrock is composed of two different units. The interface between the two
bedrock units, at a depth of about 460 m, causes a low-frequency resonance at
0.5 Hz. The two bedrock units have shear-wave velocities close to 1000 and
3700 m=s, respectively. The resonance in the neighborhood of 0.5 Hz is controlled
by the thickness and V S value of the first bedrock unit from roughly 100 to 460 m.
This frequency coincides with the surface intensities recorded at the basin during the
2010 M w 8.8 Maule earthquake. Shear-wave velocity (V S ) tomography models from
dispersion curves show heterogeneities in half-space stiffness that predict the observed
differences in low-frequency (long-period) ground motions.

Introduction
The Concepción basin is located in the center of the corresponds to the first-order Bessel function, which can be
second most important metropolitan area in Chile. This used to obtain the Rayleigh-wave dispersion curve of a site.
alluvial basin was formed by the Biobío River and consists Chávez-García et al. (2005) generalized the SPAC method to
of a relatively homogeneous layer of sandy sediments different spatial configurations and combined frequency with
(Montalva et al., 2016). time-domain correlograms to enrich the interpretation of the
The Concepción basin has historically been subjected to data. A large body of work on correlograms of microtremors
strong shaking by subduction zone earthquakes in 1570 allowed the development of seismic interferometry (e.g.,
(Ms 8.3), 1657 (Ms 8.0), and 1751 (M w 8.5), which motivated Weaver and Lobkis, 2002; Snieder, 2004; Roux et al., 2005;
the decision to move the city to its current location. The new Sabra, Roux, and Kuperman, 2005). Seismic interferometry
location has experienced events of equal magnitude: 1835 focuses on extracting the coherent component of micro-
(Ms 8.5), 1960 (Ms 8.5), and 2010 (M w 8.8). The last event seisms among receivers from the average of the correlograms
caused great damage to downtown Concepción and revealed of their signals. Early studies using this technique conducted
the limited predictive capability of the Chilean national seismic large-scale ambient-noise tomography (e.g., Moschetti et al.,
regulations (Nch433 Of. 2009) for long-period ground mo- 2007; Yang et al., 2007). In this study, it is used to character-
tions. Site effects strongly effected seismic demand (Assimaki ize the much smaller volume of the Concepción basin (e.g.,
et al., 2012; Montalva et al., 2016) and often govern the dam- Snelson et al., 2007; Delorey and Vidale, 2011; Shirzad and
age distribution during large earthquakes, hence the need to Shomali, 2014a).
characterize them appropriately. For this study, a 51-day Previous qualitative uses of microtremors in Concepción
ambient-noise survey was conducted in Concepción to assess include the work of Leyton et al. (2012) and Montalva et al.
the shear-wave velocity (V S ) structure of the basin and to study (2016), in which horizontal-to-vertical spectral ratios were
its relation to site effects (coordinate stations in Table 1). successfully linked to damage after the 2010 event. Midor-
Aki (1957) proposed the spatial autocorrelation (SPAC) ikawa et al. (2014), using SPAC characterized a V S profile
technique that uses ambient-noise measurements to deter- to a depth of 200 m, computed the amplification function, but
mine V S structure. In his work, Aki shows that the ratio be- it did not fully capture the observed response spectra at the
tween the azimuthal average of the cross-spectral functions Concepción strong-motion station (CONC) during the 2010
and the cross-spectra functions of a central reference sensor Maule earthquake.

BSSA Early Edition / 1

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/doi/10.1785/0120180061/4587491/bssa-2018061.1.pdf

by Universidad de Concepcion, gmontalva
 2 D. A. Inzunza, G. A. Montalva, F. Leyton, G. Prieto, and S. Ruiz

Figure 1. (a) Metropolitan Concepción: the black box contains the stations used. The triangle represents CONC station (triggered),
stars represent stations used for tomography, and squares represent unused stations. (b) Geologic map and station locations within the
Concepción basin.

This work uses the results of a temporary seismic net- The results are then compared with the recorded ground
work of 13 broadband and accelerometer stations deployed motions at CONC.
in the Concepción metropolitan area. It focuses on six of The travel-time tomography models consider only down-
these stations located in and around the Concepción basin. town Concepción (Fig. 1). As shown by Cabas (2016), the
Time-domain correlograms and cross-spectra functions are assumption of homogeneity of the half space at the base of
calculated, and the resulting dispersion data were combined. a soil profile is not always realistic and can have significant
This allows the characterization of the dispersion curve over implications. This assumption is tested for the case of Concep-
a broader frequency range. The information is used to com- ción using a 3D model of V S to a depth of 500 m.
pute 1D site amplification functions that include not only
sediments overlying a half space, but the velocity structure Location
is also defined to a depth of 500 m, which in this case in-
cludes two bedrock layers affecting the Concepción basin. The Concepción basin (Fig. 1a) is located along the coast
of south-central Chile. The sediments in the basin are pri-
marily sands and silty sands deposited by the Biobio River.
The geological map of Concepción (Vivallos et al., 2010,
Table 1 Fig. 1b) shows Pleistocene to Holocene alluvial deposits under
Geographical Coordinate Stations downtown Concepción. Galli (1967) showed the basin is
Station Latitude (°) Longitude (°) underlain by a weathered granite, chiefly tonalities. V S mea-
Accelerometer 01 −36.851323 −73.092869 surements of tonalites from the “El Teniente” mine (located
Accelerometer 02 −36.699158 −73.120811 roughly 120 km south of Santiago, Chile) by Rojas (2014)
Accelerometer 03 −36.823575 −73.055062 show V S values in the order of 3200 m=s at depths of 350 m.
Accelerometer 04 −36.829245 −73.034579 The Concepción basin is typically less than 100 m thick.
Accelerometer 05 −36.777340 −73.092168
Accelerometer 06 −36.833860 −73.050231
Poblete and Dobry (1968) reported that bedrock formed of
Accelerometer 07 −36.743539 −72.997855 granite rock appears at a depth of about 100 m in a borehole
Broadband 01 −36.819570 −73.073752 near the eastern margin of the basin (Fig. 1b, borehole A).
Broadband 02 −36.823780 −73.047889 The inversion of gravity data by Montalva et al. (2016) using
Broadband 03 −36.866753 −73.072227 the data collected by Vivallos et al. (2010) confirmed this
Broadband 04 −36.812755 −73.042261
Borehole A −36.822183 −73.044958
depth and estimated that the basin depth varies from zero
Borehole B −36.818595 −73.035214 (outcropping rock) to about 160 m. This inversion was
confirmed by a recent 80-m-deep borehole drilled for the
BSSA Early Edition

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/doi/10.1785/0120180061/4587491/bssa-2018061.1.pdf

by Universidad de Concepcion, gmontalva
 Shallow Ambient-Noise 3D Tomography in the Concepción Basin, Chile 3

“Site Effects Observatory on Alluvial Basins” project at

another location within the basin (Fig. 1b, borehole B).

Methodology
The thickness of the Concepción basin varies from 0 to
160 m and is typically less than 100 m. Dispersion curves of
the bedrock and sediments beneath downtown Concepción
were estimated based on the microtremor records of the field
campaign. This procedure used correlograms in the temporal
domain and cross spectra in the spectral domain. Time-
domain correlograms between pairs of microtremors records
were computed for the six stations within the basin. The
correlograms of microtremors between two stations are
described by Sabra, Gerstoft, et al. (2005) as:
ZT
Cij
r1 ; r2 ; τ  νi
r1 ; tνj
r2 ; t  τdt;
EQ-TARGET;temp:intralink-;df1;55;557
1
0

in which t is the time; νi
r1 ; t and νj
r2 ; t are the recorded

Figure 2. Histogram of group velocity between all stations.
microtremors in the positions r1 and r2 , respectively; T is the
length of the time window; Cij
r1 ; r2 ; τ is the correlogram;
frequency; and Snc is a known value of the phase slowness
and τ is the time lag. Sabra, Gerstoft, et al. (2005) demon-
with its corresponding associated frequency ωn :
strated that the temporal derivatives of the correlograms are
approximations to the velocity Green’s functions, hence Z 
ω
Sp
ω  ω −1 Su
ωdω  ωn sc :
n
EQ-TARGET;temp:intralink-;df3;313;460
3
dCij ωn
EQ-TARGET;temp:intralink-;df2;55;445 ≈ −Gij
r1 ; r2 ; t  Gji
r2 ; r1 ; −t;
2
dt The cross spectra were calculated in the spectral domain to
calculate the Rayleigh-wave phase velocity. The Chávez-
in which Gij
r1 ; r2 ; t is the velocity Green’s functions be-
tween two locations. Sabra, Gerstoft, et al. (2005) showed that García et al. (2005) methodology was used to estimate
when the Green’s function is not interpreted, it is not necessary the first-order Bessel’s function J0 as:
to take the derivative; hence, the correlograms are used di-  
ω×r
rectly. The method used for processing the data is similar ρ
r; ω  J 0 ;
4EQ-TARGET;temp:intralink-;df4;313;368

c
ω
to the one described by Bensen et al. (2007), which compiles
different methodologies. Only the vertical component of the in which ρ is the azimuth average of the cross-spectra
data was used, with the mean, trend, and instrument response coefficient.
removed. The sampling frequency was reduced to 10 Hz to Ekström et al. (2009) proposed that when computing
decrease the computation time; this simplification does not the mean of the cross-spectra functions between a pair of
cause problems because the range of frequencies studied is stations, the functions will not always have the behavior
0.4–1.5 Hz. Windows of 600-s length were selected, a 5% of the first-order Bessel’s functions, but the zero crossing will
cosine taper was applied to each trace, and a four-pole Butter- retain these characteristics; hence at these points, the phase
worth bandpass filter with a 15% width from the central fre- velocity can be computed through the argument of the
quency was applied. To achieve more homogeneous signals Bessel’s functions using the following equation:
and reduce the effects of transients, the temporal (1 bit; Larose ω0 × r
et al., 2004) and spectral normalization of the traces (Shapiro EQ-TARGET;temp:intralink-;df5;313;221 C
ω0   ;
5
Zn
et al., 2006) were carried out.
Correlograms with group velocities between 0.3 and 3.5 in which C
ω0  is the phase velocity, ω0 is the angular
(km=s) were selected because they are the range where most frequency in which the Bessel’s function is zero, r is the
velocities are found in the histograms as shown in Figure 2. interstation distance, and Zn is the nth zero crossing in
The signal-to-noise ratio (SNR) was calculated using the Shir- the Bessel’s function.
zad and Shomali (2014b) method. Correlograms having SNR The time-domain phase velocities have been comple-
greater than 2 were selected. Finally, an average of 912 cor- mented with those calculated in the spectral domain to
relograms was used for each path, which were then stacked. improve the characterization of the dispersion curve. Figure 3
The group velocities were integrated as shown in equa- shows the dispersion curve between stations BB01 and BB04
tion (3) to obtain the phase velocity, in which Sp and Su are using both time and frequency domains. An inversion of the
phase and group slowness, respectively; ω is the angular dispersion curves was carried out to construct a tomography
BSSA Early Edition

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/doi/10.1785/0120180061/4587491/bssa-2018061.1.pdf

by Universidad de Concepcion, gmontalva
 4 D. A. Inzunza, G. A. Montalva, F. Leyton, G. Prieto, and S. Ruiz

Figure 3. Dispersion curve between stations BB01 and BB04. (a) Stacked correlograms of 51 days, envelopes (thick line), and black
diamonds mark the envelope’s maximum amplitude time lag. (b) Group velocity from correlograms and phase velocity using Ekström et al.
(2009). (c) Cross-spectra function with zero crossings.

model for V S. The problem starts from the integral of the

travel times (t) for a given slowness S along a ray path; this EQ-TARGET;temp:intralink-;df11;313;416 min f
Sk − αΔSk :
11
is expressed in the equation (6), in which dl is the differential
To build a tomography model, a system of equations must be
line element along the ray path
Z solved to find the unknown slowness Si constant for each cell i.
Considering the ray path is velocity dependent (i.e., the inver-
t
l  S
ldl;
6
sion process is a nonlinear one), to keep tomographic studies
EQ-TARGET;temp:intralink-;df6;55;380

ray path
for shallow seismic surveys within linear bounds, a bias of a
which can be expressed in its discrete matrix form as indi- few percent in the estimated models due to the simplified linear
cated in equation (7). This is the fundamental relation to find, approach can be accepted (Kugler et al., 2007). The analysis
in which L is the run-length matrix of each ray path for each assumes linear ray theory by assuming that the velocity model
cell that is crossed, S is the slowness matrix to find, and t is varies smoothly over the distance of a wavelength. The process
travel time for each ray (i.e., the travel-time length between is still a nonlinear iterative method because the travel times are
two points divided by the phase velocity estimated from the recalculated for each ray path until the misfit is minimized. For
dispersion curve): an overdetermined system (restrictions greater than unknowns)
L × S  t;
EQ-TARGET;temp:intralink-;df7;55;258
7 as in this case, the solution of slowness (S) that satisfies the
system of normal equations, shown in equation (8), must be
found. The solution to the system of normal equations using
EQ-TARGET;temp:intralink-;df8;55;228 LT LS  LT t;
8 a misfit function coincides with the solution that minimizes the
quadratic function shown in equation (9). The gradient is there-
fore given by equation (10). The conjugate gradient method
shown in equation (11) was implemented to solve the problem,
misfit function penalty function
EQ-TARGET;temp:intralink-;df9;55;189

f
S  z}|{  z}|{ in which α is the search line and Δs is the gradient. The penalty
1 η2 T function shown in equation (9) was introduced because the

LS − tT
LS − t S S
2 2 number of Hessian (H) conditions is large
H  LT L. Sev-
misfit function penalty function eral different models of slowness can predict the same data. To
 z}|{  z}|{ ;
9 fix this problem, restrictions were added to keep the solution
1 T T 1 η 2
close to the origin, and the term η was given the value of 0.1;
S L LS − tT LS  tT L ST S
2 2 2 this value was determined using a trade-off curve.
The initial tomography model was built from an inter-
EQ-TARGET;temp:intralink-;df10;55;71 Δf
S  LT
LS − t  η2 S;
10 polation of V S profiles using the Picozzi et al. (2009) method
BSSA Early Edition

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/doi/10.1785/0120180061/4587491/bssa-2018061.1.pdf

by Universidad de Concepcion, gmontalva
 Shallow Ambient-Noise 3D Tomography in the Concepción Basin, Chile 5

Figure 4. Values of phase velocity for each cell from travel times at each frequency.

as implemented in other studies (Hannemann et al., 2014; factor elements defined by the adaptative biweight estima-
Cheng et al., 2015; Szanyi et al., 2016). The idea is to tion method. L2 incorporates the constraints, as shown in
incorporate the constraints on the system indicated in equa- equation (11). The upper part shows the m × n L1 matrix,
tion (7) and solve by minimizing the difference between ob- which is the longitude matrix indicated in equation (7), in
served and calculated travel times as shown in equation (12), which m is the number of rays and n is the number of cells.
in which Δt
k is the normalized difference between observed In the lower part of matrix L2 (see equation 13), the n × n
and calculated travel times and t0 − tc =to , ΔS
k is the nor- submatrix K indicates the number of rays that cross the cell
malized difference between the slowness for the kth iteration times the cell’s ellipticity (Kissling, 1988). This term is cal-
S
k − S
k−1 =S
k . The process starts from the slowness of culated by dividing each cell into four quadrants and
the homogeneous first model S
k1 , which are obtained assigning to each quadrant the distance of the rays that cross
from a linear fit between the observed travel times and the it by constructing a ray density matrix, as is also the case of
interstation distances. The mean of these slowness values
the matrix L in equation (7). Then singular values for each
was used in the model and initial travel times t
k1 were then
cell are computed (λ1 , λ2 ), and finally the ellipticity is calcu-
computed using equation (7). The new solution S
k1 was
lated as λ1 =λ2 . δ2 is the damping factor of the model, which
obtained solving equation (12)
was determined to be 0.5 using trial and error, which as in-
EQ-TARGET;temp:intralink-;df12;55;168 WΔt
k  L2 ΔS
k ;
12 dicated by Long and Kocaoglu (2001) aims to stabilize the
inversion process. The n × n submatrix M is used to smooth
the results in the horizontal plane.
  Solving equation (10) for several frequency ranges
WL1
L2  ;
13 allows the computation of dispersion curves for each cell
Kδ2 M
EQ-TARGET;temp:intralink-;df13;55;130

as shown in Figure 4 (Brenguier et al., 2007). The V S profiles

in which W is an m × m diagonal matrix (Tukey, 1974) used are obtained by inverting the dispersion curves using Dinver
to stabilize the iteration process and is made up of weighting (Wathelet, 2008), which implements the neighborhood
BSSA Early Edition

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/doi/10.1785/0120180061/4587491/bssa-2018061.1.pdf

by Universidad de Concepcion, gmontalva
 6 D. A. Inzunza, G. A. Montalva, F. Leyton, G. Prieto, and S. Ruiz

algorithm as proposed by Sambridge (1999). These profiles ity structure for depths up to 500 m. The time-domain
are then interpolated, and a V S tomography model is obtained. processing of the ambient noise shows information in a wider
The second model computes the 3D V S tomography range of frequencies than the spectral domain, as also
model in one step as it was proposed by Pilz et al. (2012). observed by Chávez-García and Rodríguez (2007).
This method has been implemented by Hannemann et al. The analysis focused on the Concepción downtown
(2012) and Amoroso et al. (2017). A large part of this second where the Maule earthquake was recorded at CONC
method uses stages (or equations) of the previously presented (Fig. 1b), a triggered permanent station. It used stations
tomography model to minimize the difference between ob- BB01, BB02, BB04, AC03, AC04, and AC06, with intersta-
served and calculated travel times as shown in equation (12). tion distances from 0.64 to 3.65 km. The dispersion curves
The difference of this more direct method is that the range of for the area are shown in Figure 5a, which are complemented
the calculated matrix increases and carries a higher computa- at higher frequencies using the phase velocities calculated by
tional cost. The vector Δt
k is r × f long, in which r is the Montalva et al. (2016) and using the SPAC technique, with
number of rays and f is the number of frequencies used. both sets of data merging into a continuous dispersion curve,
ΔS
k is the m × n difference of slowness, in which m is as one would expect from compatible data sets.
the number of cells per layer and n is the number of layers. Two bedrock layers (bedrock units I and II) can be iden-
Matrix L2 (in equations 12 and 13) is composed of the tified below the basin sediments. A first approximation
rf × mn L1 matrix in the upper part and by the mnf × mnf for the V S profile in the basin was made using the fitted
K matrix in the lower part. The K submatrix is the product of dispersion curve through ordinary least-squares shown in
two functions, the first (equation 14) is the same as in the Figure 5a and then inverting it using the neighborhood algo-
previous method, and the second depends on the vertical rithm (Wathelet, 2008; Fig. 5b). The inverted profile has
coordinates (equation 15). This second function is calculated three significant layers (see Table 2), showing two abrupt
using the analytic solution for the vertical displacement V S changes at 93 and 464 m.
(Hill, 2010) and is shown in the following equation: Figure 6a shows the transfer function (TF) calculated
following the 1D elastic wave propagation theory (Thomson,
EQ-TARGET;temp:intralink-;df14;55;457 u3
z  memkz − sepkz ;
14 1950) and using only the V S values. The Q-value used for
in which z < 0, k is the wavenumber, m  0:8475, each layer followed the relation Q  V S =10 (Midorikawa
p  0:3933, and s  1:4679. The final value of the second et al., 2014). Unlike the three-layer model, the two-layer
function (w) is determined by normalizing average value of model in Figure 6a fails to produce a peak close to 0.5 Hz.
u3 within each layer as Hence, it is postulated that the 0.5-Hz peak can be attributed
R d2 to the second—and deeper—impedance contrast.
u3 dz Using the 5% damped pseudospectral acceleration re-
w  Rd1∞ u dz ;
15 EQ-TARGET;temp:intralink-;df15;55;384

0 3 corded at CONC for the Maule event as a reference (Fig. 6b),

the three-layer TF explains the peak close to 0.5 Hz. How-
in which d1 and d2 are the lower and upper bounds of
ever, the function does not explain the peak near 0.67 Hz that
each layer.
the transverse spectrum shows. This may be because of sev-
The mnf × mn M matrix, described by Ammon and
eral reasons such as that the inverted dispersion curve is an
Vidale (1993), was not used in this study because one of
average of the measured points (Fig. 5a) or that 3D effects are
the objectives was to identify strong impedance contrasts.
relevant in the response.
Equation (12) allows the computation of each cell’s phase
slowness (here, cell is referred to as the entire height of the
model on each horizontal division). Considering a Poisson Tomography
module between 0.2 and 0.4, the relation between the slow-
The 1D analysis performed is able to explain the low-
ness (S) and V S is expressed by equation (16), in which
frequency response observed at 0.5 Hz in the CONC station,
C  0:92. The final V S tomography model in the second
and it should be noted that this is the only station that showed
approach was calculated by solving for the V S value in
a strong spectral acceleration (SA) value at this low
the following equation:
frequency for the 2010 Maule earthquake (Bastías and Mon-
1 talva, 2016). However, this still does not explain the differ-
S :
16
ence in the horizontal response spectra at CONC station. The
EQ-TARGET;temp:intralink-;df16;55;188

CV S
tomography of the basin explained earlier is used to shed
The values of V S in the tomography maps are interpolated light into this difference.
using kriging (Oliver and Webster, 2015). The experiment used the six stations in downtown
Concepción with 15 ray paths (Fig. 7a). A 4 × 4 grid with
Results and Discussions 15 effective cells was used; each cell was 872 m in latitude
and 587 m in longitude with layer interfaces at 0, 60, 100,
The dispersion curve, characterized at frequencies from 150, 200, 300, 400, and 500 m. The first 3D model imple-
0.4 to 1.5 Hz, allowed the inversion of the shear-wave veloc- mented from interpolation of individual V S profiles is shown
BSSA Early Edition

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/doi/10.1785/0120180061/4587491/bssa-2018061.1.pdf

by Universidad de Concepcion, gmontalva
 Shallow Ambient-Noise 3D Tomography in the Concepción Basin, Chile 7

Figure 5. (a) Dispersion curves data for downtown stations BB01, BB02, BB04, AC03, AC04, and AC06; (b) V S profile; and (c) physical
model interpreted from the profile.

Figure 6. (a) Transfer functions (TFs) for two and three layer profiles. (b) Horizontal response spectra at the CONC station for the 2010
Maule earthquake.

in Figure 7, where phase slowness maps (Fig. 7a) are made

Table 2
V S Profile
for frequencies 0.55, 0.8, 1.0, and 1.5 Hz. For the 1.0- and
1.5-Hz frequencies, some cells were not crossed by ray
Layer (m)
paths; in these cases, the information on the dispersion curve
Start End V S (m=s)
was complemented with microtremors measurements at local
0 49 292 scale previously processed by Montalva et al. (2016).
49 93 386
The dispersion curves for every cell were inverted,
93 464 993
464 – 3695 obtaining a shear-wave velocity profile for each cell (15 in
total). These profiles are interpolated resulting in the model

BSSA Early Edition

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/doi/10.1785/0120180061/4587491/bssa-2018061.1.pdf

by Universidad de Concepcion, gmontalva
 8 D. A. Inzunza, G. A. Montalva, F. Leyton, G. Prieto, and S. Ruiz

Figure 7. Tomography model from interpolated profiles. (a) Phase velocity map for 0.8 Hz, (b) 3D model down to a depth of 500 m,
(c) cross section at 1–1′, and (d) plan view at a depth of 400 m; the gray lines show major roads. The color scale in (d) was changed to
highlight velocity changes in the depth slice. Black arrows show the directions of different stiffness associated with different seismic demand.

presented in Figure 7b. A cross section through the model is CONC has a strong amplification close to 0.5 Hz, as shown
presented in Figure 7c. In the first 150 m, the shape of the in Figure 9. This suggests that the 0.5-Hz peak (in the CONC
basin is shown with varying sediment thickness; below this, spectra) is controlled by a structure deeper than the basin
the bedrock unit I layer is observed with shear-wave veloc- sediments.
ities around 1000 m=s. Finally, between 300 and 400 m Three TFs are shown in Figure 9. The first results from
bedrock, unit II layer appears, which is the layer that causes the curve fitted to the point cloud in Figure 5a and the inver-
the amplification at low frequencies (i.e., 0.5 Hz). sion of that dispersion curve. The second and third results
The second tomographic model, which obtains the 3D show that TFs shown in Figure 9 were computed from 1D
V S model developed in one step from the Rayleigh-wave profiles at CONC location taken from each of the tomography
travel times, is shown in Figure 8. The depth limits for the models. Figure 9 also shows that the different TFs change
layers were 0, 60, 100, 150, 200, 300, 400, and 500 m. somewhat significantly in high frequencies; however, the
Frequencies 2.2, 2.0, 1.8, 1.5, 1.2, 1.0, and 0.9 Hz are used, low-frequency peak that was the objective of the model is
obtaining the final model after iteration 98. Cross sections present in all of them. This is expected because although
through the model, presented in Figure 8b,c, show the changes in bedrock V S can have a strong impact on surface
150 m of sediment. Under the sediment layer, the first bed- intensities, in this case, the impedance contrast is large in all
rock unit I layer has velocities close to 1000 m=s, along with three cases. Therefore, the TF changes mainly in amplitude.
the second bedrock unit between 400 and 500 m with veloc- The functions provide a reasonable explanation for the longi-
ities around 2500 m=s. This is the difference between the tudinal spectrum of the 2010 Maule earthquake ground
two models. The computational time of the second tomog- motion recorded at CONC, which is located in the deeper zone
raphy model is 10 times the first model because of the range of the basin. Although the intent of this article is not to com-
of matrices to be resolved. The V S profiles interpolation pare tomographic methods, the V S interpolation method best
model produces V S variations at large depths that are com- captures the surface response at 0.5 Hz as well as the variation
patible with the different surface response in the longitudinal between longitudinal and transverse components. There is in-
and transverse directions (Fig. 7d). This spatial variability is sufficient evidence to say which method is better, but in this
less evident in the one-step model. case, the V S interpolation fits the observations better.
Although the second bedrock unit does not appear at the Because the depth to the bedrock unit I layer or the base
same depth in both models, the 1D TF for both models at of the sediment has the shape of a Mexican hat (see Montalva

BSSA Early Edition

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/doi/10.1785/0120180061/4587491/bssa-2018061.1.pdf

by Universidad de Concepcion, gmontalva
 Shallow Ambient-Noise 3D Tomography in the Concepción Basin, Chile 9

Figure 8. One-step tomography model. (a) 3D model up to a depth of 500 m, (b) cross-section view 1–1′, (c) cross-section view A–A′,
and (d) V S map at a depth of 500 m; gray lines show major roads. The color scale in (d) was changed to highlight velocity changes in the depth
slice.

(RotD00, RotD50, and RotD100) are the

0th, 50th, and 100th percentiles of the
SA azimuthal analysis for each frequency
(or period). The longitudinal spectrum
closely matches the maximum azimuthal
value, which suggests that the hetero-
geneity of the half space could partly
explain the difference between the longi-
tudinal and transverse spectra. For this
hypothesis to hold true, the impedance
contrast would need to be greater in the
longitudinal direction. This is the case
and is shown by arrows in Figure 7d.
Figure 9. (a) TFs at CONC station. Continuous line represents the global dispersion Average shear-wave velocities are
curve adjusted to the point cloud as an initial estimate. Dashed line represents the V S calculated for the area of two ellipses
profile taken from the one-step tomography model. Dotted line represents the V S profile centered at CONC (−36:83°, −73:05°),
from the interpolated profiles tomography. (b) Horizontal response spectra at the CONC
station for the 2010 Maule earthquake.
with major axes in the longitudinal and
transverse directions, respectively. The V S
values are taken from the 3D tomography
et al., 2016), basin margin effects (e.g., Pilz et al., 2011) con- at 400 m, where the bedrock unit II layer begins. The ellipses
tribute to the surface response. This makes surface intensity dimensions are 95.2 m for the semiminor axis and 726.1 m
and damage variable despite a unique site class according to for the semimajor axis, as shown in Figure 11a. The azimu-
the Chilean and U.S. seismic codes. thal average values of V S (i.e., rotating the ellipses) are
An azimuthal analysis of the horizontal spectra at the shown in Figure 11b. Higher V S values were observed close
CONC station is shown in Figure 10, where the spectra are to the east–west direction axes and lower than the north–
computed for every rotation angle, and the three shown south axes. The same analysis was performed for the

BSSA Early Edition

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/doi/10.1785/0120180061/4587491/bssa-2018061.1.pdf

by Universidad de Concepcion, gmontalva
 10 D. A. Inzunza, G. A. Montalva, F. Leyton, G. Prieto, and S. Ruiz

accurate and may have significant impacts

on surface ground motions.

Conclusions
The shear-wave velocity structure for
downtown Concepción was estimated to a
depth of 500 m using ambient-noise
dispersion curves from 0.4 to 1.5 Hz.
Three layers were identified: (1) a thin ve-
neer of alluvial deposits fills the basin of
about 100 m, (2) this overlays a bedrock
unit, and (3) another bedrock unit is iden-
tified at a depth of 464 m. The impedance
Figure 10. Response spectra for as-recorded and rotated horizontal components of contrast between the two bedrock units
the CONC station records for the 2010 Maule earthquake. causes a low-frequency resonance at about
0.5 Hz.
The two tomography models pre-
shallower depths of the model. With the averaged V S values sented here show differences in the V S values calculated
from surface to the depth of the second impedance contrast, for these bedrock layers, showing that there is epistemic un-
the TF was calculated for the longitudinal and transverse di- certainty in these methods (i.e., same data can produce differ-
rections recorded at CONC station, as shown in Figure 11c. ent V S estimates). However, despite these differences, the V S
Although the TF of the actual record is not available, one can structure shows these two bedrock units in both models. The
observe the relation between the computed TF and the spec- V S of bedrock unit II is consistent with unweathered tonalites
trum for the recorded ground motion. as expected from the geologic map pattern in Figure 1b. The
Ruiz and Saragoni (2009) discussed the spectral peaks V S of bedrock unit II may represent weathered rock units,
of several strong ground-motion records, including Mexico likely tonalites.
City (1985) and Valparaíso (1985). They postulate that when The high-seismic demand near 0.5 Hz observed at sta-
there is more than one peak in the spectrum, the first one is tion CONC for the Maule earthquake cannot be explained by
associated with the seismic source, and the second one is a homogeneous half space. The V S tomography allowed the
associated with the vibration of the soil deposits. In this estimation of deeper V S profiles even while using a 1D-type
work, it is proposed that the second peak is attributable to analysis, and it explains the 0.5-Hz peak of the recorded
site effects (i.e., soil and bedrock layers). In seismic hazard ground-motion spectra. Therefore, a deep (500 m in this
analyses, site effects are typically considered via V S profiles case) velocity structure would be necessary to explain these
that end in a homogeneous half space (Cabas, 2016), but as it low-frequency amplifications. The proposed TFs explain the
has been shown in this case, this simplification might not be observed spectra at CONC station for both longitudinal

Figure 11. Ellipse rotation. (a) Ellipse position, (b) azimuthal V S averaged values at a depth of 400 m, (c) TFs of averaged V S in
longitudinal and transverse directions, and (d) horizontal response spectra at the CONC station for the 2010 Maule earthquake.

BSSA Early Edition

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/doi/10.1785/0120180061/4587491/bssa-2018061.1.pdf

by Universidad de Concepcion, gmontalva
 Shallow Ambient-Noise 3D Tomography in the Concepción Basin, Chile 11

(0.5 Hz) and transverse (0.67 Hz) peaks. The results high- Brenguier, F., N. M. Shapiro, M. Campillo, A. Nercessian, and V. Ferrazzini
light the importance and difficulty of selecting the half space (2007). 3-D surface wave tomography of the Piton de la Fournaise
volcano using seismic noise correlations, Geophys. Res. Lett. 34,
in site-response studies. The choice will modify the surface no. 2, 2–6, doi: 10.1029/2006GL028586.
demand estimates (in this case, for low frequencies). A linear Cabas, A. (2016). Improvements to the assessment of site-specific seismic
analysis incorporating information of impedance contrasts hazards, Ph.D. Thesis, Department of Civil and Environmental
at greater depths is appropriate to improve surface seismic Engineering, Virginia Polytechnic Institute and State University.
Chávez-García, F. J., and M. Rodríguez (2007). The correlation of micro-
estimation.
tremors: Empirical limits and relations between results in frequency
The assumption of a homogeneous half space would be and time domains, Geophys. J. Int. 171, no. 2, 657–664.
erroneous for the Concepción basin. A higher resolution Chávez-García, F. J., M. Rodríguez, and W. R. Stephenson (2005). An
study using more stations is currently being conducted by alternative approach to the SPAC analysis of microtremors: Exploiting
the research group to further validate this finding. stationarity of noise, Bull. Seismol. Soc. Am. 95, no. 1, 277–293.
Cheng, F., J. Xia, Y. Xu, Z. Xu, and Y. Pan (2015). A new passive seismic
method based on seismic interferometry and multichannel analysis of
Data and Resources surface waves, J. Appl. Geophys. 117, 126–135.
Delorey, A. A., and J. E. Vidale (2011). Basin shear-wave velocities beneath
The data used for this investigation are passive records Seattle, Washington, from noise-correlation Rayleigh waves, Bull.
from continuous ambient-noise measurements from two field Seismol. Soc. Am. 101, no. 5, 2162–2175.
campaigns in the metropolitan area of Concepción, Chile. Ekström, G., G. A. Abers, and S. C. Webb (2009). Determination of
These data are available upon request from the authors. surface-wave phase velocities across USArray from noise and Aki’s
spectral formulation, Geophys. Res. Lett. 36, no. 18, 5–9.
The first campaign used two Trillium 120 broadband stations Galli, C. (1967). Geología urbana y suelos de fundación de Concepción y
from Nanometrics, configured to a sampling frequency of Talcahuano, Informe final del proyecto de investigación N°75 de la
200 Hz and monitored for 54 days in the months of Novem- Comisión de Investigación Científica de la Universidad de Concep-
ber and December 2014. The second campaign used 11 sta- ción, Concepción, Chile (in Spanish).
tions that were deployed as illustrated in Figure 1. Seven Hannemann, K., C. Papazachos, M. Ohrnberger, M. Anthymidis, and A.
Lontsi (2012). 3D V S modeling from high-frequency ambient
Meridian Compact PH Nanometrics accelerometers and four noise tomography: The case of northern Mygdonia basin, Euroseistest
CMG-3T Güralp broadband were installed. These instru- area (northern Greece), Second European Conf. on Earthquake
ments have a fundamental period of 120 s and a sampling Engineering and Seismology, Istanbul, Turkey, 25–29 August 2014,
frequency of 100 Hz. The instruments were buried, whenever 1–3.
possible, and located as far away from anthropogenic influ- Hannemann, K., C. Papazachos, M. Ohrnberger, A. Savvaidis, M. Anthy-
midis, and A. M. Lontsi (2014). Three-dimensional shallow structure
ence as possible within a large city. from high-frequency ambient noise tomography: New results for the
Mygdonia basin-Euroseistest area, northern Greece, J. Geophys. Res.
Acknowledgments 119, no. 6, 4979–4999.
Hill, D. P. (2010). Surface-wave potential for triggering tectonic
This work was partly funded by Fondo Nacional de Desarrollo Cien- (nonvolcanic) tremor-corrected, Bull. Seismol. Soc. Am. 102, no. 6,
tífico y Tecnológico (FONDECYT) Award Number 11121404 and the Mil- 2337–2355.
lennium Nucleus “The Seismic Cycle Along Subduction Zones” (CYCLO), Kissling, E. (1988). Geotomography with local earthquake data, Rev.
Millennium Scientific Initiative (MSI) of the Chilean Government Grant Geophys. 26, no. 4, 659–698.
Number NC160025. Data gathered with the Fondequip EQM160015 project Kugler, S., T. Bohlen, T. Forbriger, S. Bussat, and G. Klein (2007).
“Site Effects Observatory on Alluvial Basins” were also instrumental. All Scholte-wave tomography for shallow-water marine sediments,
maps were generated with the Generic Mapping Tools (GMT; Wessel et al., Geophys. J. Int. 168, no. 2, 551–570.
2013). The authors would like to thank the Geopsy team (www.geopsy.org, Larose, E., A. Derode, M. Campillo, and M. Fink (2004). Imaging from one-
last accessed January 2018) for making its software available. bit correlations of wideband diffuse wave fields, J. Appl. Phys. 95,
no. 12, 8393–8399.
Leyton, F., G. Montalva, and P. Ramírez (2012). A preliminary study of
References
seismic microzonation of Concepción based on microtremors, geology
Aki, K. (1957). Space and time spectra of stationary stochastic waves, with and damages patterns, Obras Proy. no. 11, 40–46 (in Spanish).
special reference to microtremors, Bull. Earthq. Res. Ins. 35, 415–457. Long, L. T., and A. H. Kocaoglu (2001). Surface-wave group-velocity
Ammon, C. J., and J. E. Vidale (1993). Tomography without rays, Bull. tomography for shallow structures, J. Environ. Eng. Geophys. 6,
Seismol. Soc. Am. 83, no. 2, 509–528. no. 2, 71–81.
Amoroso, O., G. Festa, P. Bruno, L. D'Auria, G. De Landro, V. Di Fiore, and Midorikawa, S., H. Yamanaka, K. Chimoto, R. Riddell, H. Miura, and K.
A. Zollo (2017). Integrated tomographic methods for seismic imaging Saguchi (2014). Evaluation of site effects on strong-motion records in
and monitoring of volcanic caldera structures and geothermal areas, Concepción during the 2010 Maule, Chile, earthquake, Bull. Seismol.
J. Appl. Geophys. 156, 16–30. Soc. Am. 104, no. 5, 2503–2511.
Assimaki, D., C. Ledezma, G. Montalva, A. Tassara, G. Mylonakis, and R. Montalva, G. A., F. J. Chávez-Garcia, A. Tassara, and D. Jara (2016). Site
Boroschek (2012). Site effects and damage patterns, Earthq. Spectra effects and building damage characterization in Concepción after the
28 (suppl. 1), 55–74. M w 8.8 Maule earthquake, Earthq. Spectra 32, no. 3, 1469–1488.
Bastías, N., and G. Montalva (2016). Chile strong ground motion flatfile, Moschetti, M. P., M. H. Ritzwoller, and N. M. Shapiro (2007). Surface wave
Earthq. Spectra 32, no. 4, 2549–2566. tomography of the western United States from ambient seismic noise:
Bensen, G. D., M. H. Ritzwoller, M. P. Barmin, A. L. Levshin, F. Lin, M. P. Rayleigh wave group velocity maps, Geochem. Geophys. Geosys. 8,
Moschetti, N. M. Shapiro, and Y. Yang (2007). Processing seismic am- no. 8, 1–10.
bient noise data to obtain reliable broad-band surface wave dispersion Oliver, M. A., and R. Webster (2015). Basic Steps in Geostatistics: The
measurements, Geophys. J. Int. 169, no. 3, 1239–1260. Variogram and Kriging, Springer, New York, New York.

BSSA Early Edition

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/doi/10.1785/0120180061/4587491/bssa-2018061.1.pdf

by Universidad de Concepcion, gmontalva
 12 D. A. Inzunza, G. A. Montalva, F. Leyton, G. Prieto, and S. Ruiz

Picozzi, M., S. Parolai, D. Bindi, and A. Strollo (2009). Characterization of Geología y Minería. Carta Geológica de Chile, Serie Geología
shallow geology by high-frequency seismic noise tomography, Ambiental Series, 17 (in Spanish).
Geophys. J. Int. 176, 164–174. Wathelet, M. (2008). An improved neighborhood algorithm: Parameter
Pilz, M., S. Parolai, M. Picozzi, and D. Bindi (2012). Three-dimensional conditions and dynamic scaling, Geophys. Res. Lett. 35, no. 9, 1–5.
shear wave velocity imaging by ambient seismic noise tomography, Weaver, R., and O. Lobkis (2002). On the emergence of the Green’s function
Geophys. J. Int. 189, 501–512. in the correlations of a diffuse field: Pulse-echo using thermal phonons,
Pilz, M., S. Parolai, M. Stupazzini, R. Paolucci, and J. Zschau (2011). Ultrasonics 40, nos. 1/8, 435–439.
Modelling basin effects on earthquake ground motion in the Santiago Wessel, P., W. H. F. Smith, R. Scharroo, J. F. Luis, and F. Wobbe (2013).
de Chile basin by a spectral element code, Geophys. J. Int. 187, 929– GMT 5: A major new release of the generic mapping tools, Eos Trans.
945. AGU 94, no. 45, 409–410.
Poblete, M., and R. Dobry (1968). Modelo dinámico del subsuelo de Yang, Y., M. H. Ritzwoller, A. L. Levshin, and N. M. Shapiro (2007).
Concepción, Rev. IDIEM 7, no. 3, 111–136 (in Spanish). Ambient noise Rayleigh wave tomography across Europe, Geophys.
Rojas, I. (2014). Reología y fábrica de la mineralización primaria, J. Int. 168, no. 1, 259–274, doi: 10.1111/j.1365-246X.2006.03203.x.
yacimiento el teniente, Memoria de título Geólogo, Departamento
de Geología, Universidad de Chile, Santiago, Chile (in Spanish).
Roux, P., K. G. Sabra, W. A. Kuperman, and A. Roux (2005). Ambient noise Departamento de Ingeniería Civil
cross correlation in free space: Theoretical approach, J. Acoust. Soc. Universidad de Concepción
Am. 117, 79–84. Víctor Lamas 1290, Casilla 160-C, Correo 3
Ruiz, S., and G. R. Saragoni (2009). Free vibration of soils during large 4070409 Concepción, Chile
earthquakes, Soil Dynam. Earthq. Eng. 29, no. 1, 1–16. diegoinzunza@udec.cl
Sabra, K. G., P. Gerstoft, P. Roux, W. A. Kuperman, and M. C. Fehler (D.A.I.)
(2005). Extracting time-domain Green’s function estimates from
ambient seismic noise, Geophys. Res. Lett. 32, no. 3, 1–5.
Sabra, K. G., P. Roux, and W. A. Kuperman (2005). Emergence rate of Departamento de Ingeniería Civil
the time-domain Green’s function from the ambient noise cross- Universidad de Concepción
correlation function, J. Acoust. Soc. Am. 118, no. 6, 3524–3531. Edmundo Larenas 219, Casilla 160-C, Correo 3
Sambridge, M. (1999). Geophysical inversion with a neighbourhood 4070409 Concepción, Chile
algorithm—I. Searching a parameter space, Geophys. J. Int. 138, gmontalva@udec.cl
479–494. (G.A.M.)
Shapiro, N. M., M. Ritzwoller, and G. Bensen (2006). Source location of the
26 sec micoseism from cross-correlations of ambient seismic noise,
Geophys. Res. Lett. 33, no. 18, 1–5. Centro Sismológico National
Shirzad, T., and Z. H. Shomali (2014a). Shallow crustal structures of the University of Chile
Tehran basin in Iran resolved by ambient noise tomography, Geophys. Blanco Encalada 2002, Casilla #2777
J. Int. 196, no. 2, 1162–1176. 8370415 Santiago, Chile
Shirzad, T., and Z.-H. Shomali (2014b). Extracting seismic body and leyton@csn.uchile.cl
Rayleigh waves from the ambient seismic noise using the rms-stacking (F.L.)
method, Seismol. Res. Lett. 86, no. 1, 173–180.
Snelson, C. M., T. M. Brocher, K. C. Miller, T. L. Pratt, and A. M. Trehu
(2007). Seismic amplification within the Seattle basin, Washington Geosciences Department
state: Insights from SHIPS seismic tomography experiments, Bull. National University of Colombia
Seismol. Soc. Am. 97, 1432–1448. Carrera 30 #45-03
Snieder, R. (2004). Extracting the Green’s function from the correlation of 111321 Bogotá, Colombia
coda waves: A derivation based on stationary phase, Phys. Rev. E 69, gaprietogo@unal.edu.co
no. 4, 046610. (G.P.)
Szanyi, G., Z. Gráczer, E. Győri, Z. Kaláb, and M. Lednická (2016).
Ambient seismic noise tomography of a Loess high bank at Dunas-
zekcső (Hungary), Pure Appl. Geophys. 173, no. 8, 2913–2928. Geophysics Department
Thomson, W. T. (1950). Transmission of elastic waves through a stratified University of Chile
solid medium, J. Appl. Phys. 21, no. 2, 89–93, doi: 10.1063/ Blanco Encalada 2002, Casilla #2777
1.1699629. 8370415 Santiago, Chile
Tukey, J. E. (1974). Introduction to today’s data analysis, Proc. of the Conf. sruiz@dgf.uchile.cl
on Critical Evaluation of Chemical and Physical Structural (S.R.)
Information, Washington, D.C., 24–29 June 1973, 3–14.
Vivallos, J., P. Ramirez, and A. Fonseca (2010). Microzonificación Sísmica Manuscript received 23 February 2018;
de la Ciudad de Concepción, Región del Biobío, Servicio Nacional de Published Online 11 December 2018

BSSA Early Edition

Downloaded fromView
https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/doi/10.1785/0120180061/4587491/bssa-2018061.1.pdf
publication stats
by Universidad de Concepcion, gmontalva