A New Autoregressive Moving Average Modeling of H - V Spectral Ratios To Estimate The Ground Resonance Frequency

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Engineering Geology 280 (2021) 105957

Contents lists available at ScienceDirect

Engineering Geology
journal homepage: www.elsevier.com/locate/enggeo

A new autoregressive moving average modeling of H/V spectral ratios to


estimate the ground resonance frequency
Arantza Ugalde a, *, Juan José Egozcue b, César R. Ranero a, c
a
Institute of Marine Sciences - CSIC, Pg. Marítim de la Barceloneta 37-49, 08003 Barcelona, Spain
b
Technical University of Catalonia, Dept. Civil and Environmental Engineering, Jordi Girona, 1-3, 08034 Barcelona, Spain
c
Institució Catalana de Recerca i Estudis Avançats, ICREA, Passeig de Lluís Companys, 23, 08010 Barcelona, Spain

A R T I C L E I N F O A B S T R A C T

Keywords: We propose a new method to estimate the horizontal-to-vertical (H/V) spectral ratio using microtremor mea­
Microtremor surements. The technique is based on modeling the H/V transfer function by means of an AutoRegressive Moving
H/V spectral ratio Average (ARMA) filter. As compared with the conventional, Fourier-based spectra processing routines, this
ARMA model
method is efficient in extracting the fundamental resonant frequency with a higher spectral resolution. We
demonstrate its usefulness by an application to several sites in northeastern Iberian Peninsula. For the data sets
examined, our approach shows improvement over the Fourier transform-based method under adverse experi­
mental conditions. We conclude that the new technique provides robust estimates of the fundamental resonance
frequency and can be effective to characterize empirically the unconsolidated sediment thickness and bedrock
geometry.

1. Introduction amplitude spectrum of the horizontal and vertical components of


microtremor measurements at a single seismic station. The main peak of
The horizontal-to-vertical (H/V) spectral ratio analysis using the the H/V curve has proven, empirically, to provide the fundamental
background microtremor (ambient-vibration) wavefield (Nakamura, horizontal-component shear-wave (SH) resonance frequency (f0) of soft
1989, 2019) is a widely used technique to estimate the fundamental sedimentary layers (e.g., Bard, 2008). As for the amplification level, it
resonance frequency of the ground. This parameter is used in seismic may depend on the local geological conditions and is used as a lower-
microzonation studies to map areas that exhibit local site effects, or bound estimate of earthquake site response for engineering purposes
amplification of seismic ground motions due to the surface geology. (e.g., Lermo and Chávez-García, 1994; Molnar and Cassidy, 2006). The
Microtremor wavefield results from a combination of natural (sea spectral ratio is sensitive to strong acoustic impedance contrasts be­
waves, wind, rainfall) and anthropogenic (traffic, industrial activities) tween the bedrock and the overlying sediments and can be used to
sources that show temporal and spatial changes. Long period (greater determine the geotechnical parameters (depth to bedrock and shear-
than 2 s) ambient-vibration signals, also called microseisms, are asso­ wave velocities) of the subsurface layers. Considering a simple model
ciated with ocean waves at deep ocean and coastal regions, and prop­ of one layer over a half-space, an impedance contrast is theoretically
agate as Rg and Lg phases over continental paths. Short-period expected at
microtremors originate mainly from human activities and consist of
h = β/4f0 , (1)
body (P and S) and surface (Rayleigh and Love) waves (e.g., Okada,
2003). The ground motion amplitude spectrum at a site will depend on where β is the average S-wave velocity of the soil layer (Haskell, 1960;
the vibration sources and on the thickness and physical properties of the Ibs-von Seht and Wohlenberg, 1999). The H/V spectral ratio curve has
uppermost soil layers, which act like a filter that may increase the been modeled theoretically based on the diffuse field approach that
duration and amplitude of the recorded signal in certain frequencies (e. takes into account the full ambient-vibration wavefield (García-Jerez
g., Molnar and Cassidy, 2006). et al., 2016; Sánchez-Sesma, 2017), and by Rayleigh wave ellipticity.
The H/V method involves computing the ratio between the One reason of the lack of agreement regarding the correct physical

* Corresponding author.
E-mail addresses: a.ugalde@icm.csic.es (A. Ugalde), juan.jose.egozcue@upc.edu (J.J. Egozcue), cranero@cmima.csic.es (C.R. Ranero).

https://doi.org/10.1016/j.enggeo.2020.105957
Received 30 April 2020; Received in revised form 15 October 2020; Accepted 5 December 2020
Available online 9 December 2020
0013-7952/© 2020 Elsevier B.V. All rights reserved.
A. Ugalde et al. Engineering Geology 280 (2021) 105957

model of the H/V spectral ratio is that the composition of the noise recognized in the recordings as an increase of all the three-component
wavefield at the H/V peak frequencies is not known (see Lunedei and power spectra at certain frequencies (e.g., Castellaro and Mulargia,
Malischewsky, 2015, for a review). All the theories proposed are able to 2009). However, other artifacts can be difficult to detect.
reproduce the main peak frequency of the H/V spectral ratio; however, Assuming the H/V spectral ratio function in Eq. (2) can be modeled
the soil profile inverted from experimental data depends on the as an ARMA process, the resonance frequencies of the soil still can be
modeling of the whole H/V spectral ratio curve. Within this theoretical recovered under these unfavorable conditions. ARMA models are suit­
framework, the H/V spectral ratios have been inverted to estimate the able representations of processes involving rational spectra and have
subsurface shear wave velocity structure. The estimated velocity models been applied successfully for a long time in many scientific fields,
can be improved by combining information of other geophysical and including geophysics and engineering (e.g., Kozin, 1988; Li and Kareem,
geotechnical data during the inversion (e.g, Piña-Flores et al., 2017; 1990; Glaser, 1996; Akintug and Rasmussen, 2005; Ólafsson and
Rosa-Cintas et al., 2017; Perton et al., 2018). Sigbjörnsson, 2011).
Due to its lower cost, environmental friendliness and relative ease of
computing with respect to other geotechnical and geophysical methods, 3. The autoregressive moving average model
the H/V technique has been used extensively for geologic and engi­
neering studies (e.g., Leyton et al., 2013; Molnar et al., 2018; Famiani A stationary process affected by additive nonstationary random noise
et al., 2020; Molnar et al., 2020; Rohmer et al., 2020). Motivated by the can describe the horizontal (h) and vertical (v) components of ambient-
increasing use of the H/V method, standard criteria for data acquisition, vibration recordings at a point x. The model can be written in the time
processing and interpretation were provided in the frame of the Euro­ domain as:
pean project SESAME (Site EffectS assessment using AMbient Excita­
h[n] = uh [n] + wh [n]
tions) (SESAME, 2004; Bard, 2008). Although standardized recording (3)
v[n] = u3 [n] + wv [n]
and processing routines are used for estimating the H/V ratio, it has been
observed that measurement conditions can produce misleading results where uh[n] = u1[n] + ju2[n] is a complex signal that combines the two
(e.g., Castellaro and Mulargia, 2009). Moreover, the spectral processing orthogonal horizontal components, u3[n] is the vertical component
methodologies used may also influence the final estimates (e.g., Dia­ signal, 0 ≤ n ≤ N − 1, N is the length of the data record, j is the imaginary
gourtas et al., 2001; Anthony et al., 2020). In this study, we devised a unit, and x is implicit to simplify the notation. The signals are assumed
computational approach that overcomes adverse measurement condi­ to be uncorrelated with the additive noise in the horizontal wh and
tions and spectral processing limitations. The method is based on vertical wv components.
AutoRegressive Moving Average (ARMA) models that are high- The horizontal-to-vertical ratio of Eq. (3) can be expressed in poly­
resolution spectral analysis methods used to model rational transfer nomial form as:
function processes. It is not a physical model of the process but a
mathemathical approach to enhance the spectral estimation of the H/V Uh (z) = S(z)U3 (z) (4)
ratio. This parametric modeling of data is an alternative to the con­
ventional use of non-parametric spectral techniques based on the fast where z is the shift operator. Mathematically, S(z) can be considered a
Fourier Transform (FFT). It improves the inherent performance limita­ filter transfer function between the input u3[n] and output uh[n]. Linear
tions of the FFT approach, such as frequency resolution and data win­ filters with rational transfer functions are related to general ARMA
dowing (Kay and Marple, 1981). We proposed the original idea behind models through the parameterization:
this approach in Ugalde et al. (1998), and it is further developed in this B(z)
∑q − k
k=0 b[k]z
work. Here we describe the modeling used to estimate the fundamental S(z) = = ∑ (5)
A(z) 1 + pk=1 a[k]z− k
resonance frequency of soils through H/V spectral ratios and illustrate
the new method using data from the permanent Catalan Seismic where B(z) and A(z) are two polynomials having, in this case, complex (b
Network (north-eastern Iberian peninsula; ICGC, 2000). [k]) and real (a[k]) coefficients, (p, q) represent the order of the model, z
= ejω∆t and ∆t is the sampling interval. In view of these considerations,
2. The problem we devised that the horizontal-to-vertical ratio can be actually estimated
using ARMA spectral techniques.
The spectral ratio of the horizontal (H) and vertical (V) components By parameterizing the horizontal and vertical components of the
of microtremor recordings at a point x can be expressed as: signal with the same number of coefficients, we assume that our H/V
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ process can be modeled with a general ARMA (p, p) model. Among the
H
(x, ω) =
E1 (x, ω) + E2 (x, ω)
(2) different approaches proposed to estimate the model coefficients, we use
V E3 (x, ω) the modified covariance method. It consists of minimizing the forward
and backward prediction errors in the least squares sense and is thought
where Ei(x, ω) represent the spectral energies of the horizontal (i=1, 2) to enhance the spectral peaks. Considering a forward linear prediction,
and vertical (i=3) components and ω is the angular frequency. Ac­ for which the estimate at time index n is based on p samples indexed
cording to the standard procedures (e.g., SESAME, 2004), Eq. (2) is earlier in time, the general ARMA system equation for the output uh[n]
evaluated by computing, and then smoothing, the Fourier power spectra and input u3[n] stationary signals is given by:
of the vertical and horizontal recordings over several overlapping time ∑p ∑p
windows. The H/V ratio is calculated for every single time window and k=0
a[k](h[n − k] − wh [n − k] ) = k=0
b[k](v[n − k] − wv [n − k] ) (6)
is finally averaged, so that the source deconvolution is performed in
Multiplying each side of Eq. (6) by v[n − m], taking expectations, and
each time window. Instead, Eq. (2) can be evaluated through first
after some algebra, we get the forward prediction error as:
averaging the horizontal and vertical spectra for all windows and then
∑p ∑p
computing the ratio of the averaged spectra. Using this technique, the ef [m] = − b[k]rvv [m − k] + a[k]rhv [m − k] (7)
effect of the source deconvolution is balanced over many time windows,
k=0 k=0

which increases stability (e.g., Perton et al., 2018; Spica et al., 2018). Here m = − (N − 1) + p, …, N − 1 and ril[s] are the autocovariance (i
Finite duration local nonstationarity or isolated outliers (point de­ = l) or crosscovariance (i ∕
= l) functions at time lag s, estimated as:
fects) in the data segments may affect the power spectra and, thus, the
statistical behavior of the spectral ratios estimates. Some transient
nonstationary signals of natural or anthropogenic origin can be

2
A. Ugalde et al. Engineering Geology 280 (2021) 105957

1 ∑N− s (1984). The estimated coherences from all time windows were
rhh [s] = h[n + s]h* [n]
N n=0 geometrically averaged. Coherence values can help to interpret the H/V
spectral ratio curve.
1 ∑N− s
rvv [s] = v[n + s]v[n] (8) Since we propose a method based on different data processing, we
N n=0
compared our results with the popular, well-established, open-source
1 ∑N− s Geopsy software package (e.g., Wathelet et al., 2020), based on SES­
rhv [s] = h[n + s]v[n]
N n=0 AME’s standard approach. We calculated the power spectra using the
FFT for Hanning-tapered, 50% overlapping time windows of 80 s and
where * denotes the complex conjugate and s = 0, ± 1, …, ± (N − 1). The smoothed the power spectra by means of the Konno and Ohmachi
relations rhh[− s] = rhh*[s], rvv[− s] = rvv[s] and rhv[− s] = rvh*[s] yield (1998) weighting function using a smooth parameter of 2π/b=0.4,
results for negative lags. where b is the width of the smoothing window.
Similarly, we consider the general ARMA system equation for
backward linear prediction, in the sense that the estimate at time index n 4.1. CFON site
is based on p samples indexed later in time. Multiplying by h*[n + m],
taking expectations, and after simplification, we obtain the backward Fig. 1 shows the H/V spectral ratio obtained for station CFON. At this
prediction error as site, a clear peak emerges at fARMA = ± 3.75 ± 0.10 Hz that stabilizes for
0
∑p ∑p
eb [m] = *
b[k]rhv [m − k] − *
a[k]rhh [m − k] (9) model order p = 60. The spectral ratios for positive and negative fre­
k=0 k=0
quencies show a difference in amplitude of about 13% at f0. CFON is
Model parameters a[k] (real) and b[k] = b1[k] + jb2[k] (complex) located in a mountain area at an elevation of 973 m, which suggests the
can be estimated by minimizing the forward and backward prediction influence of topographic or subsurface structure effects on the amplitude
errors: at the resonant frequency (e.g., Kawase et al., 2019; Matsushima et al.,
∑N− 1 ⃒ ⃒2 2014). In this case, the main frequency peaks show low coherence
efb = m=− (N− 1)+p
μ⃒ef [m] ⃒ + η|eb [m] |2 (10) values. Previous geophysical characterization studies at CFON (avail­
able at https://www.icgc.cat/; last accessed April 2020) found an upper
where μ and η are the error weights. To find the filter coefficients that layer of recent, low compacted soil of quaternary colluvial materials
minimize Eq. (10), we set the partial derivatives of efb with respect to a with S-wave velocities β = 281 m/s extending to ~10 m depth. An un­
[k], b1[k] and b2[k] equal to zero for k = 0, …, p, and a[0]=1. Thus, we derlying second layer of completely to moderately altered substrate with
obtain a system of 3p+2 linear equations for 3p+2 unknowns that we β = 536 m/s extends to a depth H~40 m where overlays bedrock. Using
solve using the Gauss-Jordan elimination algorithm (see Appendix A). these values, the fundamental frequency of the profile can be estimated
Once the model parameters are estimated, the transfer function (H/V) by means of the simplified equation (method 7 of Dobry et al., 1976):
(ω) can be computed using Eq. (5). ⎛ ⎞1/2
The critical problem of ARMA modeling is the determination of the ∑n
(H− zm (i) )2
⎜ 4 h(i) ⎟
best model order p, which would be equivalent to set the amount of
2
1 ⎜ v(i)

f = ⎜∑n i=1 ⎟ (11)
smoothing. Too small values of p may cause data underfitting, in which 2
2π ⎝ i=1 (x(i) + x(i + 1) ) h(i) ⎠
case the obtained spectral ratio is too smooth to allow high spectral
resolution. On the contrary, a too high order may cause data overfitting,
thus leading to an unstable model. Our compromise criterion between where n is the number of layers, H − zm is the midpoint depth of layer i, h
underfit and overfit is to compute the transfer function in Eq. (5) for is the thickness of each layer, and x(i + 1) = x(i) + h(i)(H − zm(i))/β(i)2 is
increasing model order p until the frequency of the H/V peak stabilizes. the estimated fundamental mode of vibration at the upper boundary of
layer i. Eq. (11) gives a value of f=3.73 Hz, which is in accordance with
4. The H/V spectral ratio the resonance frequency of 3.75 Hz estimated with the ARMA-based
approach.
Because we were interested in testing carefully the new methodol­ Fig. 1(b) shows the H/V spectral ratio obtained for station CFON
ogy, we focused on high quality seismic data recorded in good seis­ using Geopsy. The computed resonance frequency is fFFT 0 =3.87 ± 0.13
mometer vaults. For this purpose we considered several permanent, Hz. We note that, taking into account the error intervals, results of the
broadband seismic stations of the Catalan Seismic Network (ICGC, ARMA approach and the FFT-based method are in agreement for this
2000). We used 45 min of data recorded during quiet-weather, night- site.
time hours on June 30, 2019. In the model-based ARMA analysis the
length of the data sequence does not affect the frequency resolution; 4.2. CORI site
therefore in this approach we used 512-sample (5.12 s) length, 50%
overlapping windows. For each window, we computed the auto and A comparison between ARMA and the FFT-based method perfor­
crosscovariance functions of the vertical and horizontal components mance is also presented in Fig. 2 for CORI seismic station. No resonance
according to Eq. (8), estimated the model coefficients by solving the frequency can be identified in this case with either method, which agrees
system of Eq. (A.3), and computed the transfer function using Eq. (5). with the installation of CORI on class A ground type site (bedrock
Finally, we geometrically averaged the H/V spectral ratios from all time outcrop) according to Eurocode 8 (EC8, 2004).
windows. We repeated this procedure for increasing values of p. As a
result of the ARMA complex-valued approach, negative and positive 4.3. ICJA site
frequencies were calculated. According to Eq. (5), negative frequencies
represent a rotation in the complex plane. Therefore, the asymmetry of To further assess the ARMA model performance, we analyzed data
the curves indicates the possible directional dependence of the H/V recorded at station ICJA, which is deployed in an urban area at the
spectral ratio, because the real and imaginary parts of b[k] parameterize Institute of Earth Sciences Jaume Almera in the city of Barcelona. Fig. 3
the two horizontal components of motion. (a) displays the spectral ratio curve obtained for model order p = 100,
The complex coherence function between the horizontal and vertical which presents the maximum amplification at fARMA0 = ± 2.66 Hz. The
components was also computed following the procedure described in coherence is low for the main peak. A very similar value of fFFT
0 =2.69 Hz
Bendat and Piersol (1986), where the autospectral and cross-spectral is obtained using the FFT-based method (Fig. 3b). Considering a mean
density functions were estimated according to Lagunas and Gasull shear-wave velocity of about 422 m/s from previous studies in this zone

3
A. Ugalde et al. Engineering Geology 280 (2021) 105957

Fig. 1. (a) 50% percentile of the H/V spectral ratio for station CFON (Catalan Seismic Network) using the ARMA model for order p = 60. The grey area encloses the
25% and 75% percentile regions. The dashed line represents the coherence between the horizontal and vertical components. The frequency of the main peak for
negative (f− ) and positive (f+) frequencies is plotted. (b) H/V spectral ratio using Geopsy. The dashed lines indicate the standard deviation.

Fig. 2. (a) H/V spectral ratio for station CORI using the ARMA model for order p = 60. The grey area encloses the 25% and 75% percentile regions. (b) H/V spectral
ratio for station CORI using Geopsy. The dashed lines indicate the standard deviation.

Fig. 3. (a) H/V spectral ratio for station ICJA using the ARMA model for order p = 100. The grey area encloses the 25% and 75% percentile regions. (b) H/V spectral
ratio for station ICJA using Geopsy. The dashed lines in indicate the standard deviation.

4
A. Ugalde et al. Engineering Geology 280 (2021) 105957

(Cadet et al., 2011) using Eq. (1) we infer a theoretical sediment these observations, we computed the power spectral density of ambient
thickness of about 40 m at this site. This result is in agreement with the noise at station CARA (Fig. 5a). Two persistent, narrow spectral peaks
lithology found at the adjacent Almera-1 borehole (Jurado and Salvany, are clearly recognized at 10 Hz and 12.5 Hz, which are compatible with
2016) that sampled Quaternary sediment to a depth of 41 m overlying continuous quasi-harmonic signals of industrial origin (e.g., Anto­
Paleozoic basement rocks. novskaya et al., 2019). CARA seismic station is surrounded by several
hydroelectric power facilities. In particular, Vielha hydropower station
is located at about 1.8 km from CARA site. This facility is equipped with
4.4. CARA site water turbines rotating at 750 rpm that explain the observed spectral
peak at 12.5 Hz. The spectral peak at 10 Hz may be generated by six
We found that the ARMA approach and the FFT-based method yield hydroelectric facilities with turbine rotation velocities of 600 rpm
significantly different results for particular datasets. Fig. 4(a, b) com­ located at distances less than 7.5 km from CARA. These observations
pares the resonance frequency obtained using data from CARA borehole indicate that the main H/V spectral ratio peak observed is induced by
seismic station, which is deployed at 27 m depth on class A ground. the 12.5 Hz frequency generated at the hydroelectric power station. To
Using data recorded during night-time hours, the ARMA method detects filter out the signals excited by the industrial facilities we applied a
a main peak frequency fARMA0 = ± 12.51 ± 0.04 Hz that stabilizes for notch filter to the data and recomputed the H/V spectral ratio. The
model order p = 70; however, no significant peak can be identified using resulting power spectral density is plotted in Fig. 5 (b). Fig. 6 shows that
Geopsy for the same data. no spectral peak of stratigraphic origin can be inferred from the ARMA
Nevertheless, using data recorded during day-time hours, the FFT- model at CARA site using the notch-filtered data. But a frequency peak
based method yields a main peak frequency fFFT 0 =11.92 ± 1.02 Hz still emerges at fFFT
0 =10.44 Hz using the method based on the FFT.
(Fig. 4d). For the same day-time data, the ARMA results stabilize earlier Checking H/V in all directions of the horizontal plane, we note a strong
(model order p = 30) and the frequency pattern is unaltered (Fig. 4c); azimuthal variation that presents a maximum around 18◦ (+180◦ )
however, the H/V spectral ratio curve presents higher (more than 4 (Fig. 7). These results will be discussed later.
times) amplification levels than for nigh-time data. An increase of the
coherence is not observed. Energy directionality of the main peak is also
noted as an increase of about 17% in the peak amplitude between 4.5. ARBS site
negative and positive frequencies (Fig. 4c).
To test the possible influence of high-frequency noise sources on We also applied the ARMA method to data recorded at ARBS, which

Fig. 4. H/V spectral ratio for station CARA using the ARMA model (a, c) and Geopsy (b, d). The grey area in the left column encloses the 25% and 75% percentile
regions. The dashed lines in the right column indicate the standard deviation. Results for night-time (a, b) and day-time hours (c, d) are plotted. See the text
for discussion.

5
A. Ugalde et al. Engineering Geology 280 (2021) 105957

Fig. 5. Vertical (Z), East-West (E) and North-South (N) power spectral density (PSD) at station CARA for day-time hours in dB related to (m/s2)2/Hz before (a) and
after (b) notch-filtering with a quality factor Q (centre frequency/bandwidth) of 150.

Fig. 6. (a) H/V spectral ratio for station CARA using the ARMA model for order p = 60. The grey area in (a) encloses the 25% and 75% percentile regions. (b) H/V
spectral ratio for station CARA using Geopsy. The dashed lines indicate the standard deviation. Day-time, notch-filtered data at center frequencies 10.0 Hz and 12.5
Hz are used.

fFFT
0 =2.25 ± 0.34 Hz emerges using the FFT-based method for the same
data set (Fig. 8b). The frequency of the main peak changes during the
day in the range 2–3 Hz, having the minimum values at night and the
maximum ones during the 8–15 h time interval. These observations
support a relationship of the main peak with anthropogenic noise
sources, given that ARBS is located at ~8 km distance from Andorra la
Vella (capital of the Principality of Andorra). Moreover, by analyzing the
H/V variation as a function of azimuth we note that the spectral ratio
presents the maximum values around 69◦ (Fig. 9). The directivity
observed may be related to the direction of the noise sources. However,
ARBS is deployed at the crest of a ridge, and the obtained preferential
direction N69E corresponds to the direction of the largest slope, that
presents maximum height drops of >60%. Therefore, if the H/V spectral
ratio peak observed with the FFT-based method in this site was inter­
preted as the resonance frequency of the ground, the directionality
might be related to the effects of surface topography that is generally
observed to cause larger amplification on the horizontal components
Fig. 7. Azimuth variation of the H/V spectral ratio for station CARA preferentially along the direction perpendicular to the ridge axis (e.g.,
using Geopsy. Chávez-García et al., 1996; Massa et al., 2014). These observations will
be discussed in the next section.
is a 30-m-depth borehole seismic station deployed on rock at an eleva­
tion of 2145 m in the Pyrenean range. Fig. 8(a) shows coherence values
around 0.5 and a relatively flat spectral ratio curve, which may indicate
a lack of high acoustic impedance at depth. However, a peak at

6
A. Ugalde et al. Engineering Geology 280 (2021) 105957

Fig. 8. (a) H/V spectral ratio for station ARBS using ARMA for model order p = 50. The grey area encloses the 25% and 75% percentile regions. (b) H/V spectral ratio
for station ARBS using Geopsy. The dashed lines indicate the standard deviation.

of inverting for the optimum filter coefficients that model the process.
The strongest contribution of this new method is its robustness for
estimating the natural frequency of the ground with high spectral res­
olution. Other intrinsic advantages of the method are its applicability to
short data lengths and the avoidance of window-biasing effects. The
technique is less of an art than the FFT-based method and it can be
automated. However, it requires an extensive investigation of the model
that represents the data through the determination of the appropriate
order of the model p. The computational complexity of the method de­
pends on the model order. When p is high, then the computational
complexity will be large compared to the FFT-based methods. Several
criteria and empirical methods have been proposed over the years to
determine the model order, but they do not always work in practical
cases (including our data). Here we devised an empirical approach
adapted to our problem, which consists in running the code for
increasing values of p until the H/V spectral ratio stabilizes. The algo­
rithm performance is mostly affected by the empirical procedure chosen
to determine the model order; however, high performance computing
Fig. 9. Azimuth variation of the H/V spectral ratio for station ARBS
using Geopsy. may speed up considerably this process.

5. Discussion and conclusions 5.2. ARMA approach performance

5.1. The spectral processing We tested the new technique with seismic data acquired by several
permanent stations of the Catalan seismic network and compared the
The H/V spectral ratio of microtremor recordings is conventionally obtained results with those computed using the widely recognized
computed by means of non-parametric spectral techniques based on the Geopsy software based on classical spectral analysis methods. The pre­
Fourier Transform. Several parameters actually used for the analysis dominant frequencies obtained with the ARMA approach and Geopsy
such as the size of the data windows, percentage of overlap and the level show good correspondence for measurements at CFON, CORI and ICJA
of smoothing, may bias the spectral estimates and lead to misleading sites. CFON station shows one distinctive resonant frequency at 3.75 Hz
results. The smoothing process reduces the variance of the spectral es­ that is related to a lithological interface at a depth of 30 m. CORI pre­
timates, at the cost of compromising the frequency resolution. More­ sents a relatively flat spectral ratio curve as theoretically expected at
over, the spectral estimates contain the contribution of the properties bedrock sites. ICJA site shows a peak of stratigraphic origin at 2.66 Hz
from several neighboring frequencies over the interval of smoothing. that is related to a high impedance contrast at a depth of 40 m.
These amplitudes may be discordant, thus distorting features of interest. Conversely, results for CARA and ARBS present some discrepancies
Therefore, considerable care must be exercised in choosing the right caused by the processing methodologies due to some singular features of
parameters depending on the nature of the input data (e.g., Diagourtas the underlying seismic data. At CARA site, two predominant frequencies
et al., 2001; Anthony et al., 2020). For this reason, previous experience at 10 Hz and 12.5 Hz are clearly identified using the ARMA approach.
is often required when applying these techniques for computing and They are caused by the activity at the existing hydroelectric facilities
interpreting the H/V spectral ratio. close to the site that generates continuous quasi-harmonic mechanical
Here we proposed a new technique that is expected to overcome the oscillations (Antonovskaya et al., 2019). The main peak at 12.5 Hz is
above-mentioned difficulties. It is a parametric technique based on caused by the closest facility, having water turbines rotating at 750 rpm
modeling the H/V time series by an ARMA filter in which the system is and located at 1.8 km distance from the seismic station. The rest of the
represented by a polynomial transfer function expressed in terms of the closest hydropower facilities in the area (six), which are located at
model parameters. The method for estimating the spectral ratio consists straight-line distances of less than 7.5 km from the site, have turbine
rotation velocities of 600 rpm and are responsible for the spectral peak

7
A. Ugalde et al. Engineering Geology 280 (2021) 105957

around 10 Hz. The order of the ARMA model defines the complexity of resolution than the classical methods. Because of the complex-valued
the signal (i.e., the number of harmonic components that can be approach, the asymmetry of the curves between negative and positive
modeled). For this reason, the high signal-to-noise ratio signals at CARA frequencies indicates the possible directional dependence of noise
for day-time hours can be modeled with a lower number of parameters sources. The empirical interpretation of the H/V spectral peaks obtained
than signals recorded during night-time hours, which require a more with the ARMA approach in terms of the depth of the impedance
complex model to enhance the spectral peaks. Whereas the exact fre­ contrast agrees well with the available geophysical and geotechnical
quencies of the disturbances emerge with the ARMA technique no information. The proposed method may be useful to estimate with
matter at what time the measurements are taken, the FFT-based method improved precision the depth of the engineering bedrock, which can be
fails to detect the peaks using data from night-time hours. However, two used to optimize the distribution of geotechnical campaigns in engi­
broad H/V peaks emerge around these frequencies when the spectral neering projects.
ratio is computed using day-time data (when the demand of electricity is Data availability
high and so is the signal-to-noise ratio). Although the spectral peaks Seismic data (waveforms and instrument response) have been ob­
detected are not related to stratigraphic features, this is an example of tained through the Federation of Digital Seismograph Networks (FDSN)
the superior performance of the ARMA approach to extract predominant web service available at https://www.icgc.cat (last accessed April
frequencies from stationary data. After processing the data with a notch 2020). ARBS station is deployed at La Rabassa, Andorra, in collaboration
filter, no resonance frequency peak can be identified in the resulting between the Institut Cartogràfic i Geològic de Catalunya (ICGC) and the
spectral ratio using the ARMA technique, but the FFT-based method still Institut d’Estudis Andorrans (IEA). ICJA station belongs to the Institute
detects a broad peak around 10.44 Hz, which presents a strong direc­ of Earth Sciences Jaume Almera (http://labsis.ictja.csic.es/edtarray, last
tivity with a maximum around 18◦ (+180◦ ) (Fig. 7). If we interpret this accessed April 2020).
frequency as the resonance frequency of the ground, the directivity
might be connected to the topography of the site, as CARA is deployed at
a mountain ridge, and the direction 198◦ N points to the direction of the Declaration of Competing Interest
maximum slope, having an average height drop of ~46%. This would
indicate that surface topographic effects might affect the results in this The authors declare that they have no known competing financial
site, causing large amplifications on the horizontal components prefer­ interests or personal relationships that could have appeared to influence
entially along the direction perpendicular to the ridge axis, as it has been the work reported in this paper.
observed before (e.g., Massa et al., 2014). However, if the frequency
peak observed at CARA is related to anthropogenic noise, the directivity Acknowledgements
observed could be related to the direction of the noise sources. The fact
that the ARMA approach does not identify the spectral peak favors this We are very grateful to the Editor and two anonymous reviewers for
second interpretation. The same behavior is observed at ARBS. In this taking the time and effort necessary to review the manuscript. We
case, the ARMA approach yields a near-flat spectral ratio curve, whereas sincerely appreciate all valuable comments and suggestions, which
the FFT-based method detects a peak at 2.25 Hz. The predominant fre­ helped us to improve the quality of the article. This research was
quency at ARBS changes during the day, which suggests an anthropo­ partially supported by Euskontrol, S. A., through the ISURI-1 project.
genic origin of the noise sources. Also, it presents the maximum H/V The activity of J. J. Egozcue was supported by the project METhods for
amplification at an azimuth of 69◦ , which again may correspond to the COmpositional analysis of DAta (CODAMET), Ministerio de Ciencia,
direction of the largest average slope or to the direction of the noise Innovación y Universidades (Ref: RTI2018-095518-B-C22, 2019-2021).
sources. More detailed analyses would be required at CARA and CARBS This is a contribution of the Barcelona Center for Subsurface Imaging
to reach a conclusion. that is a Grup de Recerca de la Generalitat de Catalunya. Geopsy package
To conclude, the proposed ARMA method is a powerful algorithm to can be downloaded from http://www.geopsy.org/ (last accessed April
estimate the resonance frequency of soils using microtremor records. We 2020). Results of the ARMA method were plotted using the Generic
showed that it yields more robust results with a higher spectral Mapping Tools (GMT) software (Wessel et al., 2013).

Appendix A. Appendix

To minimize the forward and backward prediction errors in Eq. (10), we set
∂efb ∂efb ∂efb
= 0 (l ∕
= 0), a[0] = 1; = 0; =0 (A.1)
∂a[l] ∂b1 [l] ∂b2 [l]
This leads to the set of equations:
∂efb ∑∑ ( * *
)
= a[k] rhv [m − l]rhv [m − k] + rhh [m − l]rhh [m − k]
∂a[l] k m

( *
)
− b1 [k] rhv [m − l]rvv [m − k] + rhh [m − l]rhv [m − k]
( *
)
− jb2 [k] rhv [m − l]rvv [m − k] + rhh [m − l]rhv [m − k] = 0

∂efb ∑∑ ( )
= *
− a[k] rvv [m − l]rhv [m − k] + rhv [m − l]rhh [m − k] (A.2)
∂b1 [l] k m

( *
)
+ b1 [k] rvv [m − l]rvv [m − k] + rhv [m − l]rhv [m − k]
( *
)
+ jb2 [k] rhv [m − l]rhv [m − k] = 0

8
A. Ugalde et al. Engineering Geology 280 (2021) 105957

∂efb ∑∑ ( )
*
= − ja[k] rvv [m − l]rhv [m − k] + rhv [m − l]rhh [m − k]
∂b2 [l] k m

( *
)
+ jb1 [k] rhv [m − l]rhv [m − k]
( *
)
+ b2 [k] rvv [m − l]rvv [m − k] + rhv [m − l]rhv [m − k] = 0,

where m = − (N − 1) + p, …, N − 1; k = 0, …, p; l = 0, …p; and we have considered μ = η = 1/2.


Eq. (A.2) can be expressed in matrix form as:
R⋅f = R0 , (A.3)

where f = (a[1]⋯a[p]b1[0]⋯b1[p]b2[0]⋯b2[p])T and


⎛∑ ( * *
)⎞
− rhv [m − 1]rhv [m] + rhh [m − 1]rhh [m]
⎜ m ⎟
⎜ ⎟
⎜ ⋮ ⎟
⎜∑ ( ) ⎟
⎜ − rhv [m − p]rhv [m] + rhh [m − p]rhh [m] ⎟
* *
⎜ ⎟
⎜ m ⎟
⎜ ∑( ⎟
⎜ *
) ⎟
⎜ r [m]r [m] + r [m]r [m] ⎟
⎜ vv hv hv hh ⎟
⎜ m ⎟
⎜ ⎟

R0 = ⎜ ⋮ ⎟
∑ ⎟
⎜ ( *
) ⎟
⎜ r vv [m − p]rhv [m] + r hv [m − p]r [m] ⎟
⎜ hh ⎟
⎜ m ⎟
⎜ ∑ ( ) ⎟
⎜ * ⎟
⎜ j r vv [m]r hv [m] + r hv [m]rhh [m] ⎟
⎜ m

⎜ ⎟
⎜ ⋮ ⎟
⎜ ⎟
⎝ ∑( ) ⎠
*
j rvv [m − p]rhv [m] + rhv [m − p]rhh [m]
m

References Glaser, S., 1996. Insight into liquefaction by system identification. Géotechnique 46 (4),
641–655. https://doi.org/10.1680/geot.1996.46.4.641.
Haskell, N.A., 1960. Crustal reflection of plane SH waves. J. Geophys. Res. 65 (12),
Akintug, B., Rasmussen, P.F., 2005. A Markov switching model for annual hydrologic
4147–4150. https://doi.org/10.1029/JZ065i012p04147.
time series. Water Resour. Res. 41 (9), W09424 https://doi.org/10.1029/
Ibs-von Seht, M., Wohlenberg, J., 1999. Microtremor measurements used to map
2004WR003605.
thickness of soft sediments. Bull. Seismol. Soc. Am. 89 (1), 250–259.
Anthony, R.E., Ringler, A.T., Wilson, D.C., Bahavar, M., Koper, K.D., 2020. How
Institut Cartogràfic i Geològic de Catalunya (ICGC), 2000. Catalan Seismic Network.
processing methodologies can distort and bias power spectral density estimates of
International Federation of Digital Seismograph Networks. Other/Seismic Network.
seismic background noise. Seismol. Res. Lett. 91 (3), 1694–1706. https://doi.org/
https://doi.org/10.7914/SN/CA.
10.1785/0220190212.
Jurado, M.J., Salvany, J.M., 2016. Scientific drilling in the campus: Almera-1 borehole,
Antonovskaya, G., Kapustian, N., Basakina, I., Afonin, N., Moshkunov, K., 2019.
unraveling urban subsurface geology in Barcelona (Spain). Geotemas 16, 617–620.
Hydropower Dam State and its foundation soil survey using industrial seismic
Kawase, H., Nagashima, F., Nakano, K., Mori, Y., 2019. Direct evaluation of S-wave
oscillations. Geosciences 9 (4), 187. https://doi.org/10.3390/geosciences9040187.
amplification factors from microtremor H/V ratios: double empirical corrections to
Bard, P.-Y. (Ed.), 2008. The H/V technique: results of the SESAME Project. Bull. Earthq.
“Nakamura” method. Soil Dyn. Earthq. Eng. 126, 105067. https://doi.org/10.1016/
Eng. 6 (1), 1–147 (special issue).
j.soildyn.2018.01.049.
Bendat, J.S., Piersol, A.G., 1986. Random Data: Analysis and Measurement Procedures.
Kay, S.M., Marple, S.L., 1981. Spectrum analysis – a modern perspective. Proc. IEEE 69
Wiley, New York (407 pp).
(11), 1380–1419. https://doi.org/10.1109/PROC.1981.12184.
Cadet, H., Macau, A., Benjumea, B., Bellmunt, F., Figueras, S., 2011. From ambient noise
Konno, K., Ohmachi, T., 1998. Ground-motion characteristics estimated from spectral
recordings to site effect assessment: the case study of Barcelona microzonation. Soil
ratio between horizontal and vertical components of microtremor. Bull. Seismol.
Dyn. Earthq. Eng. 31 (3), 271–281. https://doi.org/10.1016/j.soildyn.2010.07.005.
Soc. Am. 88, 228–241.
Castellaro, S., Mulargia, F., 2009. The effect of velocity inversions on H/V. Pure Appl.
Kozin, F., 1988. Autoregressive moving average models of earthquake records. Probab.
Geophys. 166, 567–592. https://doi.org/10.1007/s00024-009-0474-5.
Eng. Mech. 3 (2), 58–63. https://doi.org/10.1016/0266-8920(88)90016-1.
Chávez-García, E.J., Sáchez, L.R., Hatzfeld, D., 1996. Topographic site effects and HVSR.
Lagunas, M., Gasull, A., 1984. An improved maximum likelihood method for power
A comparison between observations and theory. Bull. Seismol. Soc. Am. 86,
spectral density estimation. IEEE Trans. Acoust. Speech Signal Process. 32 (1),
1559–1573.
170–173. https://doi.org/10.1109/TASSP.1984.1164292.
Diagourtas, D., Tzanis, A., Makropoulos, K., 2001. Comparative study of microtremor
Lermo, J., Chávez-García, F.J., 1994. Are microtremors useful in site response
analysis methods. Pure Appl. Geophys. 158, 2463–2479. https://doi.org/10.1007/
evaluation? Bull. Seismol. Soc. Am. 84, 1350–1364.
PL00001180.
Leyton, F., Ruiz, S., Sepúlveda, S.A., Contreras, J.P., Rebolledo, S., Astroza, M., 2013.
Dobry, R., Oweis, I., Urzua, A., 1976. Simplified procedures for estimating the
Microtremors’ HVSR and its correlation with surface geology and damage observed
fundamental period of a soil profile. Bull. Seismol. Soc. Am. 66 (4), 1293–1321.
after the 2010 Maule earthquake (Mw 8.8) at Talca and Curicó, Central Chile. Eng.
EC8, 2004. Eurocode 8 : Design of structures for earthquake resistance. Part 1: General
Geol. 161, 26–33. https://doi.org/10.1016/j.enggeo.2013.04.009.
rules, seismic actions and rules for buildings. In: European Norm, European
Li, Y., Kareem, A., 1990. ARMA systems in wind engineering. Probab. Eng. Mech. 5 (2),
Committee for Standardisation, European Committee for Standardisation Central
49–59. https://doi.org/10.1016/S0266-8920(08)80001-X.
Secretariat, Rue de Stassart 36, B-1050 Brussels, Belgium.
Lunedei, E., Malischewsky, P., 2015. A review and some new issues on the theory of the
Famiani, D., Brunori, C.A., Pizzimenti, L., Cara, F., Caciagli, M., Melelli, L., Mirabella, F.,
H/V technique for ambient vibrations. In: Ansal, A. (Ed.), Perspectives on European
Barchi, M.R., 2020. Geophysical reconstruction of buried geological features and site
Earthquake Engineering and Seismology, Geotechnical, Geological and Earthquake
effects estimation of the Middle Valle Umbra basin (Central Italy). Eng. Geol. 269,
Engineering, vol. 39, pp. 371–394. https://doi.org/10.1007/978-3-319-16964-4_15.
105543. https://doi.org/10.1016/j.enggeo.2020.105543.
Massa, M., Barani, S., Lovati, S., 2014. Overview of topographic effects based on
García-Jerez, A., Piña-Flores, J., Sánchez-Sesma, F.J., Luzón, F., Perton, M., 2016.
experimental observations: meaning, causes and possible interpretations. Geophys.
A computer code for forward calculation and inversion of the H/V spectral ratio
J. Int. 197 (3), 1537–1550. https://doi.org/10.1093/gji/ggt341.
under the diffuse field assumption. Comput. Geosci. 97, 67–78. https://doi.org/
Matsushima, S., Hirokawa, T., De Martin, F., Kawase, H., Sánchez-Sesma, F.J., 2014. The
10.1016/j.cageo.2016.06.016.
effect of lateral heterogeneity on horizontal-to-vertical spectral ratio of microtremors

9
A. Ugalde et al. Engineering Geology 280 (2021) 105957

inferred from observation and synthetics. Bull. Seismol. Soc. Am. 104 (1), 381–393. system using the diffuse field assumption (DFA). Geophys. J. Int. 208 (1), 577–588.
https://doi.org/10.1785/0120120321. https://doi.org/10.1093/gji/ggw416.
Molnar, S., Cassidy, J.F., 2006. A comparison of site response techniques using weak- Rohmer, O., Bertrand, E., Mercerat, E.D., Régnier, J., Pernoud, M., Langlaude, P.,
motion earthquakes and microtremors. Earthquake Spectra 22 (1), 169–188. https:// Alvarez, M., 2020. Combining borehole log-stratigraphies and ambient vibration
doi.org/10.1193/1.2160525. data to build a 3D Model of the lower Var Valley, Nice (France). Eng. Geol. 270,
Molnar, S., Cassidy, J.F., Castellaro, S., Cornou, C., Crow, H., Hunter, J.A., 105588. https://doi.org/10.1016/j.enggeo.2020.105588.
Matsushima, S., Sánchez-Sesma, F.J., Yong, A., 2018. Application of microtremor Rosa-Cintas, S., Clavero, D., Delgado, J., López-Casado, C., Galiana-Merino, J.J.,
horizontal-to-vertical spectral ratio (MHVSR) analysis for site characterization: state Garrido, J., 2017. Characterization of the shear wave velocity in the metropolitan
of the art. Surv. Geophys. 39, 613–631. https://doi.org/10.1007/s10712-018-9464- area of Málaga (S Spain) using the H/V technique. Soil Dynamics and Earthquake
4. Engineering 92, 433–442. https://doi.org/10.1016/j.soildyn.2016.10.016.
Molnar, S., Assaf, J., Sirohey, A., Adhikari, S.R., 2020. Overview of local site effects and Sánchez-Sesma, F.J., 2017. Modeling and inversion of the microtremor H/V spectral
seismic microzonation mapping in Metropolitan Vancouver, British Columbia, ratio: physical basis behind the diffuse field approach. Earth Planets Space 69–92.
Canada. Eng. Geol. 270, 105568. https://doi.org/10.1016/j.enggeo.2020.105568. https://doi.org/10.1186/s40623-017-0667-6.
Nakamura, Y., 1989. A method for dynamic characteristics estimation of subsurface SESAME, 2004. Guidelines for the implementation of the H/V spectral ratio technique on
using microtremor on the ground surface. Q. Rep. Railway Tech. Res. Inst. 30 (1), ambient vibrations. In: Measurements, Processing and Interpretation. WP12-
25–33. Deliverable D23.12. https://doi.org/10.1007/s10518-008-9059-4, p 62. Available
Nakamura, Y., 2019. What is the Nakamura method? Seismol. Res. Lett. 90 (4), as supplementary material at. (last accessed April 2020).
1437–1443. https://doi.org/10.1785/0220180376. Spica, Z.J., Perton, M., Nakata, N., Liu, X., Beroza, G.C., 2018. Site characterization at
Okada, H., 2003. The Microseismic Survey Method. Society of Exploration Geophysicists Groningen gas field area through joint surface-borehole H/V analysis. Geophys. J.
of Japan, translated by Koya Suto. Geophysical Monograph Series No. 12, Society of Int. 212 (1), 412–421. https://doi.org/10.1093/gji/ggx426.
Exploration Geophysicists, Tulsa. Ugalde, A., Egozcue, J.J., Alfaro, A., Pujades, L.G., Canas, J.A., 1998. Estimation of the
Ólafsson, S., Sigbjörnsson, R., 2011. Digital filters for simulation of seismic ground system function of soils using microtremors. Annales Geophysicae, part 4. Nonlinear
motion and structural response. J. Earth Eng. 15 (8), 1212–1237. https://doi.org/ Geophysics & Natural Hazards 16 (4), 1208.
10.1080/13632469.2011.565862. Wathelet, M., Chatelain, J.-L., Cornou, C., Giulio, G.D., Guillier, B., Ohrnberger, M.,
Perton, M., Spica, Z., Caudron, C., 2018. Inversion of the horizontal-to-vertical spectral Savvaidis, A., 2020. Geopsy: a user-friendly open-source tool set for ambient
ratio in presence of strong lateral heterogeneity. Geophys. J. Int. 212 (2), 930–941. vibrationprocessing. Seismol. Res. Lett. 91 (3), 1878–1879. https://doi.org/
https://doi.org/10.1093/gji/ggx458. 10.1785/0220190360.
Piña-Flores, J., Perton, M., García-Jerez, A., Carmona, E., Luzón, F., Molina-Villegas, J. Wessel, P., Smith, W.H.F., Scharroo, R., Luis, J.F., Wobbe, F., 2013. Generic mapping
C., Sanchez-Sesma, F.J., 2017. The inversion of spectral ratio H/V in a layered tools: improved version released. EOS Trans. AGU 94, 409–410. https://doi.org/
10.1002/2013EO450001.

10

You might also like