Techniques To Identify Microtremor Wave Contributions and Impact To Seismic Site Characterization

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 28

J Seismol (2024) 28:345–372

https://doi.org/10.1007/s10950-024-10189-5

RESEARCH

Techniques to identify microtremor wave contributions


and impact to seismic site characterization
Hema Sharma · Sheri Molnar · Aamna Sirohey

Received: 21 June 2023 / Accepted: 11 January 2024 / Published online: 7 February 2024
© The Author(s), under exclusive licence to Springer Nature B.V. 2024

Abstract We investigate whether varying wavefield wave contributions. The impact to seismic site char-
contributions are the likely cause to variation in micro- acterization is that the depth and VS of the resonator
tremor horizontal-to-vertical spectral ratio (MHVSR) (half-space) layer are overestimated consistently by an
amplification shape between six sites in the relatively average of 28% compared to the often-default-assumed
homogeneous geologic setting of Windsor, Ontario. We Rayleigh ellipticity forward amplification model. Our
quantify the MHVSR shape in terms of peak broadness study demonstrates the importance of correctly iden-
and its fitness using mathematical functions to identify tifying wave type contributions of the microtremor
potential wave type contributions. We develop a tech- wavefield for the proper estimation of VS depth profiles,
nique that uses particle motion plots of cross-correlated especially to obtain correct thickness of the sediment
microtremor recordings to establish the dominant wave layer and resonator VS and thereby the average VS of the
types contributing to the microtremor wavefield within upper 30 m (VS30).
three important frequency bandwidths (below, span-
ning, and above the fundamental peak frequency). We Keywords MHVSRs · Microtremor wavefield ·
investigate the variability in the inverted shear-wave Cross-correlations · Particle motion plots · Surface
velocity (VS) depth profile by performing 21 MHVSR wave modes · Bedrock properties
inversions with varying Rayleigh, Love, and body

1 Introduction
Highlights
• Multiple forward model approximations of an MHVSR
shape are evaluated based on varying microtremor
Earthquake site effects (amplification, liquefaction, etc.)
wavefield contents. are often the main cause of earthquake damage, distant
• Novel use of particle motion plots of cross-correlations from the epicentre. Amongst them, seismic wave ampli-
to identify waves contributing to the microtremor fication results from the general reduction in the soil’s
wavefield.
• We find that fundamental mode Rayleigh ellipticity notably
shear-wave velocity (VS) and density (their product is
overestimates the resonator depth and its VS and VS30. known as seismic impedance) towards the Earth’s sur-
face. For a single soil layer over an elastic half-space,
H. Sharma (*) · S. Molnar (*) · A. Sirohey the site resonance frequency (f0) from vertical propaga-
Dept. Earth Sciences, The University of Western Ontario,
tion of horizontal shear (SH) waves is,
London, ON, Canada
e-mail: hsharm3@uwo.ca f0 = VSav ∕4h (1)
S. Molnar
e-mail: smolnar8@uwo.ca

Vol.: (0123456789)
13
346 J Seismol (2024) 28:345–372

where VSav is the single soil layer’s average VS, and wave contributions could be ignored as noise. Later,
h is soil thickness. Site amplification is determined Lachet and Bard (1994) illustrated that the f0HV can
empirically using soil base-to-surface or rock-to-soil also be explained by the ellipticity curve of funda-
spectral ratios from earthquake recordings (Borcherdt mental mode Rayleigh wave and is equivalent to f0.
1970). Empirical site amplification is also evaluated Castellaro (2016) showed that f0HV coincides with
using the horizontal-to-vertical spectral ratio from the lowest frequency minimum in the vertical com-
either earthquake or microtremor (ambient vibration) ponent Fourier spectrum because of vanishing ver-
recordings from a single seismometer. A microtremor tical energy when Rayleigh wave motion changes
horizontal-to-vertical spectral ratio (MHVSR) is the from prograde to retrograde. When Rayleigh waves
ratio of the average of horizontal components’ Fourier dominate, the vertical component Fourier spec-
spectra and the vertical component’s Fourier spec- trum is minimum at f0HV and maximum at ⁓2f0HV
trum of a tri-axial seismometer’s microtremor record- (Castellaro 2016). Hence, there are two important
ings (Nogoshi and Igarashi 1971; Nakamura 1989). identifiers of an empirical MHVSR dominated by
The reader is referred to the most recent review of Rayleigh ellipticity: the vertical component Fourier
the microtremor method (Molnar et al. 2022) which spectrum amplitude is always a minimum at result-
documents evolution in understanding of the MHVSR. ing in amplitude divergence between the horizon-
An MHVSR amplification spectrum provides reliable tal and vertical component spectra (termed as ‘eye
site frequency(ies) related to soil thickness and VS in shape’ pattern by Castellaro and Mulargia 2009),
the subsurface (known as seismic site conditions). An and MHVSR de-amplification occurs at ⁓2f0HV,
MHVSR amplification spectrum can also provide reli- i.e. a trough is present in the MHVSR at ~ 2. Cas-
able amplification estimates at site resonance (peak) tellaro and Mulargia (2009) use the first identifier
frequency(ies) when there is a strong seismic imped- to confirm f0HV is of stratigraphic origin. For a
ance contrast. MHVSRs can therefore be inverted to linear elastic and homogeneous half-space, sur-
determine subsurface seismic site conditions, but it face waves exist as only Rayleigh waves (Aki and
is not often accomplished due to significant model Richards 1980). Earth’s layering gives rise to addi-
non-uniqueness and continued distrust of MHVSR tional surface wave modes, including Love waves
amplification as a reliable measure of site amplifica- which arise due to constructive interference of up-
tion. Joint inversion of a MHVSR spectrum with a going and down-going SH waves (Aki and Richards
site’s surface wave dispersion curve is often used to 1980). Konno and Ohmachi (1998) showed that
reduce model non-uniqueness. The remaining hin- for high impedance contrasts, Love waves increase
drance to sole MHVSR inversion is an appropriate for- the MHVSR peak amplitude. The Airy phase of
ward prediction model. Without the knowledge of the Love waves can also give rise to resonance (Ben-
microtremor wavefield composition, the appropriate Hador and Buchen 1999) and thereby increase the
forward prediction model is unknown. The evolution horizontal amplitude. Hence, an important indicator
in the understanding of which seismic wave type(s) of the presence of Love wave contributions to the
dominates the microtremor recording and thereby microtremor wavefield is an increased horizontal
the MHVSR in approximately chronological order is amplitude; the higher the amplitude of the horizon-
as follows: SH waves, Rayleigh waves, both Rayleigh tal components within the ‘eye-shape’ frequency
and Love waves or diffracted waves, i.e. the total bandwidth, the more Love wave contributions there
wavefield comprising body (primary, and horizontal are to the wavefield.
and vertical shear waves, P, SV and SH waves respec- Bonnefoy-Claudet et al. (2008) showed that
tively) and surface waves. the contribution of surface and body waves in the
In the formative Nakamura (1989) publication, microtremor wavefield varies according to the
the site’s microtremor fundamental-mode resonance S-wave impedance contrast between the sediment
frequency (f0HV) is shown to be equivalent to f0 and and bedrock layer, as well as the source geometry
thereby it was assumed that body waves dominate (the direction and distance of the source from the
the microtremor wavefield, while Rayleigh surface measurement station). They concluded that the

Vol:. (1234567890)
13
J Seismol (2024) 28:345–372 347

microtremor wavefield is mainly composed of Ray- The f0HV or the entire MHVSR spectrum is
leigh and Love waves for high impedance contrast often inverted for subsurface site conditions even
sites (> 4); Rayleigh waves exist, but the wavefield though the appropriate forward model to predict
is dominated by Love waves for moderate imped- the MHVSR (physical basis of the microtremor
ance contrast sites (> 3 and ≤ 4); and is mainly wavefield) for the site is often unknown. Several
composed of Love and body waves for low imped- researchers have tried to reduce the Love and body
ance contrast sites (≤ 3), with minor contribution wave content in the microtremor recording to fit
from Rayleigh waves. Along with the impedance the MHVSR as a Rayleigh ellipticity function. Fäh
contrast, the source geometry also plays a role in et al. (2001) modified empirical MHVSRs by multi-
determining the microtremor wavefield compo- plying them with a scalar quantity of √2 to remove
sition. Body waves attenuate faster than surface the influence of Love and SH waves prior to inver-
waves (Aki and Richards 1980), so that away from sion using a Rayleigh ellipticity forward model.
the source, the microtremor wavefield is dominated Rayleigh wave ellipticity was also computed by
by surface waves. Conversely, when microtrem- Hobiger et al. (2009) using the random decrement
ors are recorded closer to their sources, there is technique (RayDec), which reduces the Love and
some contribution of body waves along with sur- body waves from the single-station recordings. The
face waves. Albarello and Lunedei (2011) showed open-source Geopsy software (www.​geopsy.​org,
that when sources are nearby, for a high imped- version 3.4.2, Wathelet et al. 2020), developed by
ance contrast, body waves dominate the wavefield SESAME (Bard et al. 2004), inverts the unmodified
below the f0HV; for frequencies above 2f0HV, surface empirical MHVSR spectra as Rayleigh wave ellip-
waves dominate the wavefield; and at intermediate ticity functions using horizontal and vertical eigen-
frequencies, spanning the f0HV and 2f0HV, the wave- functions from Aki and Richards (1980). Arai and
field can be influenced by both surface and body Tokimatsu (2004) developed an inversion scheme
waves. The diffraction or splitting of wave types that accounts for Love waves and higher modes of
due to lateral variability in subsurface ground con- Rayleigh and Love waves in the forward modelling
ditions is known to broaden f0HV and in extreme of MHVSRs. The theoretical amplification function
cases result in plateaued MHVSRs with no clear in the Matlab-based ModelHVSR routine developed
f0HV (Uebayashi 2003). by Herak (2008) is based on vertically incident P
The most recent proposal to explain the micro- and SV waves. In most of the cases, these approxi-
tremor wavefield is that of a diffuse wavefield mations work. The theoretical amplification func-
assumption (DFA; Sánchez-Sesma et al. 2008) tion of the HVInv algorithm (http://​www.​ual.​es/​
which assumes that a diffusion-like phenomenon Grupo​sInv/​hv-​inv/) developed by García-Jerez et al.
takes place when microtremors are generated by (2016) is based on diffractions of the total wavefield
various sources, due to multiple scattering of seis- (Sánchez-Sesma et al. 2008).
mic waves. If a three-dimensional (3D), diffuse, Several array-based studies were also accom-
equipartitioned, harmonic displacement vector field plished to understand particle motions of the recorded
is established within an elastic medium, the Green’s microtremor wavefield. Hobiger et al. (2011) illus-
function can be approximated by the cross-corre- trated an array-based method, MUSIQUE, that can
lation of recordings at two points in the medium. estimate the Rayleigh wave particle motions and
When these two points coincide, the amplification can determine if they are prograde or retrograde and
function is given in terms of the imaginary part hence is useful in separating the wave modes. Degree-
of the Green’s functions (Piña-Flores et al. 2020). of-polarization has been calculated to estimate Ray-
Overall, the respective contribution of wave types to leigh ellipticity from the microtremors (Schimmel
the microtremor wavefield depends on the source(s) and Gallart 2004; Poggi and Fäh 2010; Berbellini
distribution and site conditions and ultimately et al. 2019). Poggi and Fäh (2010) used high-reso-
impacts the MHVSR amplification, i.e. the shape or lution frequency-wavenumber (HRFK) technique to
morphology of the MHVSR curve. filter out Love and SH wave content to acquire the

Vol.: (0123456789)
13
348 J Seismol (2024) 28:345–372

Fig. 1  Location map of six array sites in Windsor. Top inset test holes; CH, cross-hole surveys; VSP, vertical seismic pro-
shows an example of the three-sensor array geometry. The co- filing) of the Ministry of Transportation of Ontario are also
located invasive VS profiling locations (CPT, cone penetration shown

Rayleigh ellipticity. Tapia et al. (2020) identify body site amplification (forward) modelling routine
and surface waves using particle motion plots from identified from the microtremor recordings and/
landslide recordings. In a different study, cross-cor- or the MHVSR curve leads to improvement in the
relations of microtremor recordings and their particle inverted VS profiles and hence seismic site charac-
motion plots were used by Yeh et al. (2013) to esti- terization. To accomplish this, we invert MHVSR
mate a VS model at a fault zone in Taiwan. curves by assuming various wavefield combina-
In this study, we perform a detailed analysis tions including Rayleigh, Love and body waves,
of the MHVSR curves at six seismic array sites in as well as higher surface wave modes. The impact
Windsor, Ontario, Canada (Fig. 1) to test novel to seismic site characterization amongst the 21
techniques in identifying wave type contributions amplification models is evaluated in terms of
of microtremor recordings. Our goal is to deter- four chosen metrics: depth and VS of the resona-
mine the physical basis of the MHVSR and thereby tor (half-space), the velocity ratio at the top of the
determine the most appropriate forward amplifi- resonator and the time-averaged VS of the upper
cation model for inversion either from the micro- 30 m, VS30.
tremor recordings (array-based) or directly from
the MHVSR curve (single-station). We examine
the functional form of MHVSR using two math- 2 Variation in theoretical amplification function
ematical functions (Gaussian and Lorentzian). with changing wavefield
We further examine the particle motion plots of
cross-correlated microtremor recordings to ana- To demonstrate variation in the shape of theoretical
lyse the body and surface wave contributions to site amplification dependent on the assumed wave-
the recorded microtremor wavefield. We also test field content (forward model), we calculate theo-
whether the effort to improve inversion fitness of retical site amplification functions for a simple one-
the MHVSR curve of the appropriate theoretical layer soil model (VS0 = 200 m/s) of 10 m thickness

Vol:. (1234567890)
13
J Seismol (2024) 28:345–372 349

Fig. 2  A For a single soil


layer model, B theoretical
amplification functions are
computed as Rayleigh ellip-
ticity, SH transfer function,
or the DFA considering
different wave combina-
tions: 100% Rayleigh waves
(R), 50% Rayleigh and 50%
Love waves (R + L) and
Rayleigh, Love and body
waves (R + L + BW)

over an elastic half-space (VS1 = 600 m/s) (Fig. 2) ellipticity. The combination of body and surface
based on Rayleigh ellipticity (gpell algorithm of the waves to the wavefield broadens and shifts the peak
Geopsy software package, Wathelet et al. 2020), the frequency lower, towards the expected SH-wave f0.
transverse shear-wave transfer function (gpsh algo- Amplification below the peak frequency is increased
rithm within Geopsy) and varying seismic wave when Love waves or body waves are present. Fig-
contributions based on the DFA (HVInv tool of ure 2 reinforces that different combinations of
García-Jerez et al. 2016). Figure 2 shows that the contributing seismic waves affect the shape of the
calculated Rayleigh ellipticity functions are consist- MHVSR curve.
ent regardless of the available tool (gpell or HVInv).
The Rayleigh ellipticity spectrum exhibits f0HV at
6.6 Hz with a trough at 8.3 Hz; the curve is asym- 3 Geologic setting
metrical with a gentler slope below f0HV. When an
equal amount of Love wave contribution is added The city of Windsor, Ontario, is in the southernmost
(50% Rayleigh and 50% Love waves, termed R + L), part of Canada, on the bank of the Detroit River chan-
f0HV does not change but its amplification increases nel connecting Lake Huron with Lake Erie via Lake
significantly (3.9 to 9.3). Amplification increases St. Clair (Fig. 1). The Quaternary deposits range
because the added Love wave content increases the from a few meters to 50 m in thickness and strongly
horizontal component amplitudes. The theoretical reflect the underlying bedrock surface (Hudec 1998).
SH transfer function has a peak frequency of 5 Hz, The general stratigraphy from base to top (oldest to
where f0 = VSav/4 h = 200/4(10) = 5 Hz, and peak youngest) applicable to the six array sites consists of
amplification of 3, where A0 = VS1/VS0 = 600/200 = 3. Middle Devonian Detroit River Group limestone and
A combination of body (P, SV, and SH) and sur- dolostone bedrock (Easton and Carter 1995; Hudec
face (equal contribution of R and L) waves (termed 1998), overlain by a few meters of Catfish Creek grav-
R + L + BW) results in the same 5 Hz peak fre- elly till (the basal till/limestone interface), and the
quency as the SH transfer function’s f0, but the peak thickest unit of clayey-to-silty Tavistock Till, which
is broader, and amplification is higher at and below is then overlain by glaciolacustrine silty clay. A unit
the peak frequency. Figure 2 demonstrates that Ray- of fine-to-coarse grained deltaic channel sand overlies
leigh ellipticity functions for a moderate imped- the unweathered silty clay unit west of Huron Church
ance contrast site (A0 = 3) exhibits a narrow and Road (Hudec 1998) and is likely present at array sites
high amplitude peak shape compared to a lower and 5 and 6 (Fig. 1). Overall, the geologic stratigraphy in
broader SH transfer function amplification response. Windsor corresponds to sites that are relatively later-
The addition of Love waves (R + L) increases only ally continuous, thereby ideal for one-dimensional
the peak amplification compared to the Rayleigh site amplification studies.

Vol.: (0123456789)
13
350 J Seismol (2024) 28:345–372

4 Empirical MHVSR data MHVSR is calculated by dividing the mean horizon-


tal spectrum by the vertical spectrum for each time
Bilson Darko et al. (2020) accomplished non-invasive window then averaged for all time windows for each
passive seismic testing to obtain MHVSRs and funda- site. Figure 3 shows the average MHVSRs with their
mental-mode Rayleigh wave dispersion curves at six standard deviation for the six Windsor seismic array
seismic array sites in Windsor, Ontario (Fig. 1) for sites of Bilson Darko et al. (2020). f0HV is determined
blind comparison of inverted VS profiles with invasive based on two criteria: (1) amplification is > 2, and
(standard and cone) penetration testing. Microtremor (2) f0HV corresponds to divergence between the mean
recordings of 15 to 20-min duration were collected by horizontal spectrum and the vertical spectrum.
Bilson Darko et al. (2020) using three tri-axial Trom- For the sites with a reliable (low standard devia-
ino® velocimeters with 128 Hz sampling in a triangu- tion) MHVSR spectrum, the f0HV is consistently 2 Hz.
lar array geometry (inset of Fig. 1). To calculate the Sharma and Molnar (2019) demonstrated consist-
MHVSR, the full duration microtremor recording is ency in this f0HV of ~ 2 Hz at an additional 45 sites
segmented into time windows of frequency-depend- across Windsor. The MHVSR is not reliable when
ent lengths (50 cycles per window length) and cosine the standard deviation is high, often below f0HV due
tapered. Time windows with high amplitude local to poor sensor-to-ground coupling, climatic effects or
transients are removed using an anti-triggering filter transients from nearby sources (Molnar et al. 2022).
based on short-to-long term averaging, overlapped Bilson Darko et al. (2020) obtained a broadened and
by 5%, and Fourier transformed into amplitude fre- reduced MHVSR peak amplification at sites 2 and 6
quency spectra and smoothed using a Konno and (grey curves in Fig. 3) in comparison to a single clear
Ohmachi (1998) filter with the smoothing constant set MHVSR peak amplification at the other four sites.
to 40. The similarity of the two individual horizontal This was relatively puzzling as a rapid variation in
component Fourier amplitude spectra is checked to the geology is not expected. Instead, they noted this
confirm uniform subsurface site conditions prior to change in MHVSR shape could be attributed to bad
calculating their quadratic mean. The time-averaged weather conditions and/or poor sensor coupling.

Fig. 3  Average MHVSR curves (solid line) with one standard Darko et al. (2020). Red, blue and green lines represent peak
deviation (dashed lines) for sites 1 to 6 (A to F). Grey curves half-width, peak height and peak base respectively
in B and F are original MHVSR spectra determined by Bilson

Vol:. (1234567890)
13
J Seismol (2024) 28:345–372 351

Bilson Darko et al. (2020) therefore discarded these be determined from only two of the four seismom-
unreliable MHVSRs in their joint dispersion curve and eter recordings at both sites (black curves in Fig. 3
f0HV inversions (i.e. performed dispersion curve only B and F). Thus, we interpret that sensor coupling is
inversions) for VS profiles at sites 2 and 6. the dominant factor impacting MHVSR reliability for
The largest amplitude source of microtremors in these sites in Windsor. The reliable MHVSR spectrum
our frequency range (0.5–15 Hz) at the six sites is for site 2 is similar to site 5’s MHVSR with a wide
road traffic, particularly along a four-lane expressway, f0HV base and the presence of a lower amplification
as there are no other major noise sources nearby. Sites left shoulder peak. For site 6, the reliable MHVSR
1, 3 and 4 are located the furthest from the express- spectrum maintains a broadened f0HV. The MHVSR
way, at distances of 220 to 300 m. The MHVSR spec- shape at site 6 is consistent over the recording dura-
tra for these sites correspond to a tall and narrow f0HV tion and therefore anomalous compared to the other
shape, most consistent with a theoretical Rayleigh sites. Having ruled out climatic and temporal effects,
ellipticity function, conveying that the microtremor the variability in the MHVSR shape amongst the six
wavefield is likely dominated by Rayleigh waves. In sites must therefore be primarily the result of site
contrast, the MHVSR shape for the other three sites (2, conditions or wavefield effects.
5 and 6), located at distances between 70 and 100 m To convey MHVSR shape quantitatively, we report
of the expressway, is variable and unexpected. The MHVSR peak(s) according to their peak height and
change in MHVSR shape at site 5 includes a minor half-width from which to calculate peak broadness
narrow peak in the left flank of the peak (we infor- (Fig. 3; Table 1). The peak height is calculated by
mally term this as a ‘left shoulder’) which does not subtracting the MHVSR amplification at the peak’s
impact site characterization; a clear f0HV is obtained. baseline from the maximum peak amplification, and
In contrast, the MHVSR shape at sites 2 and 6 (grey half-width is calculated as the frequency bandwidth
curves in Fig. 3 B and F) impacts their reliability or of the MHVSR peak at half-height. The peak broad-
use for site characterization; a clear f0HV cannot be ness is calculated by dividing the half-width by the
determined for site 2, and the broadness of f0HV at site peak height and multiplying by a factor of 100. The
6 increases its uncertainty. broadest peak at site 6 is ~ 5.5 times broader than the
To preclude MHVSR reliability issues due to narrowest peaks of sites 3 and 4 (Table 1).
sensor-to-ground coupling, climatic effects, or near- There is a spatial trend in the MHVSR shape
source transients, we revisited and carefully rede- amongst the six sites: moderate f0HV amplification
ployed four tri-axial Tromino® seismometers in a with a wide base at sites 1 and 2 in the southeast;
linear array with about 20 m sensor spacing at sites highest and narrow f0HV amplification at sites 3 and 4;
2 and 6. The MHVSR processing steps are the same and greater variation in the f0HV shape (left shoulder
as described above. The extra care and attention peak, broadness) at sites 5 and 6 in the northwest. If
applied to seismometer deployment during our sec- wavefield contributions are not considered, our inter-
ondary testing improves MHVSR reliability for the pretation would be that there is likely lateral (2D) var-
given recording, yet reliable MHVSR curves could iability (Uebayashi 2003) towards the northwest (sites
5 and 6). However, the observed variations in MHVSR
shape may be controlled by variations in wavefield
contributions amongst the six sites given the uniform
Table 1  Quantitative measures of MHVSR shape
subsurface ground conditions in southwestern Wind-
Site Peak height Peak half-width Peak broadness sor. Sites 1, 3 and 4 are likely dominated by Rayleigh
1 7.96 0.63 7.9 waves as they are farther from the source, and sites 2,
2 6.72 0.89 13.2 5 and 6 may have contributions from other wave types
3 9.92 0.5 5.0 (likely body wave) owing to their closer proximity to
4 10.18 0.59 5.8 anthropogenic microtremor sources. We proceed to
5 6.8 0.68 10.0 develop novel techniques to further identify potential
6 5.03 1.43 28.4 wave contributions from the microtremor recordings
and/or MHVSR spectra.

Vol.: (0123456789)
13
352 J Seismol (2024) 28:345–372

5 Microtremor wavefield content of materials or minerals (e.g. Yuan and Mayanovic


from mathematical quantification 2017).
of the MHVSR Figure 4 shows the fitting of Gaussian and Lor-
entzian distributions to each theoretical amplification
We propose that MHVSR shape is indicative of the function (Fig. 2) of our single soil layer over an elas-
underlying microtremor wavefield composition (e.g. tic half-space example from Section 2. We determine
Figure 2) and seek novel techniques to identify and/ and report the peak broadness (B, as calculated and
or separate these wavefield contributions from the shown in Table 1) and the percentage contribution of
MHVSR amplification spectrum. A potential tech- the Gaussian and Lorentzian distributions (Fig. 4). A
nique is the fitting of different mathematical functions linear model is used to fit the linear offset in amplifi-
with unique distributional shapes to identify unique cation across all spectral frequencies. The SH transfer
MHVSR shapes due to seismic wave type(s) embed- function with a symmetrical and broad peak is fit by
ded within the empirical MHVSR spectrum. We uti- a single peaked function, either a 100% Gaussian or a
lize the curve_fit function from the Python scipy.opti- 100% Lorentzian. For all other theoretical amplifica-
mize library that is based on a nonlinear least squares tion functions, a combination of both Lorentzian and
regression to fit a combination of mathematical func- Gaussian functions is required to fit the non-symmet-
tions (linear, Gaussian and Lorentzian) with the tar- rical and less broad peaked amplification function.
get amplification and MHVSR curves. This idea stems When the linear fit has a non-zero slope, particularly
from prominent use of theoretical distributions (e.g. to adjust for amplification offset differences above
Gaussian, Lorentzian, Gaussian–Lorentzian, Voigt) f0HV, Love and/or body waves are present. Below
to fit experimentally measured Raman spectroscopy f0HV, the amplification offset is fit by the Lorentzian
peaks and thereby identify the surface chemistry distribution rather than the linear function. We note

Fig. 4  Mathematical model fit (combination of Gaussian, Lor- example: A R, B SH, C R + L and D R + L + BW. Legend
entzian and linear functions) of different theoretical amplifica- shows the peak broadness (B) and % contribution of Gaussian
tion functions for a single soil layer over an elastic half-space (G) and Lorentzian (L) functions

Vol:. (1234567890)
13
J Seismol (2024) 28:345–372 353

Fig. 5  Mathematical model fit (combination of Gaussian, Lorentzian and linear functions) of the empirical MHVSR spectrum for
sites 1 to 6 (A to F). Legend similar as in Fig. 4

an alternative approach could be to fit specific (fixed) sloped linear offset to achieve fitness of the empirical
mathematical distributions to directly compare the MHVSR curve at f0HV and above.
percentage contribution of each distribution’s unique Figures 4 and 5 illustrate that when SH waves
shape to the theoretical amplification spectrum dominate, the MHVSR amplification curve is
amongst the considered examples. symmetrical and best fit by a single peaked
Figure 5 shows mathematical functions fit to the distribution (Gaussian or Lorentzian) with a linear
empirical MHVSR spectra at the six Windsor seis- offset of (near) zero slope. When surface waves
mic array sites. The broader and more symmetrical dominate, the MHVSR curve is asymmetric and
MHVSR peak at site 6 is fit equally by either a sin- more than one peaked function form is required
gle Gaussian or Lorentzian distribution (dotted box to fit the MHVSR at f0HV and below, with the need
in Fig. 5) and is suggestive that the microtremor of a sloped linear offset above f0HV to account for
wavefield is dominated by upward propagating SH increasing amplification due to Love and/or body
waves. All other sites are best fit by a minimum of wave contributions.
two mathematical distributions. Sites 3 and 4 are best
fit by a tall Lorentzian distribution, noting a second
short Lorentzian distribution is required to fit the 6 Microtremor wavefield content
empirical peak’s base. Sites 1, 2 and 5 are best fit from cross‑correlated microtremor recordings
by a combination of a narrow Gaussian distribution
(central peak) and broader and lower amplitude Lor- MHVSR analysis for our six Windsor seismic array
entzian distribution (peak’s base). Overall, the Lor- sites in the previous two sections show that in
entzian distribution is best suited to fit the MHVSR addition to surface waves, sites 2, 5 and 6 may also
curve at f0HV and below, with the need of a secondary have contributions from body waves, especially site
Gaussian or Lorentzian distribution combined with a 6. We seek to better understand the microtremor

Vol.: (0123456789)
13
354 J Seismol (2024) 28:345–372

Fig. 6  Cross-plots of low frequency (0.5–1 Hz) cross-correla- panels show the site’s MHVSR curve (black) compared to the
tions grouped by similar response: A, B sites 1 and 4, C, D five other site’s MHVSR (grey), and blue shading highlights the
sites 2 and 6 and E, F sites 3 and 5 respectively. Uppermost 0.5–1-Hz frequency band

wavefield contributions at each Windsor site using component microtremor recording at the pseudo
the cross-correlations of the simultaneous passive- source with the vertical, radial (R, parallel azimuth
seismic array recordings at two receivers. The between the seismometer pair), and transverse (T,
typical assumption of a microtremor wavefield is perpendicular azimuth) component microtremor
that uniform energy is coming at a receiver from all recordings at the receiver. This provides us with the
azimuths, and microtremors are uncorrelated in time three orthogonal cross-correlations, vertical (ZZ),
and space (Colombi et al. 2014). We performed a transverse (ZT) and radial (ZR), for each site. These
cursory verification that the microtremor wavefield orthogonal cross-correlations are calculated at three
was azimuthally distributed within our frequency select frequency bandwidths: 0.5–1 Hz, below f0HV;
range of interest (0.5–15 Hz) at each site using 1–5 Hz, spanning f0HV and 2f0HV; and 5–15 Hz, above
frequency-wavenumber analysis. To understand 2f0HV, from microtremor recordings that are bandpass
the energy partitioning at each of the six sites, we filtered for each case. For each frequency bandwidth,
consider one of the two seismometers as a pseudo particle motion plots of the three orthogonal cross-
source and the other as a receiver. Cross-correlations correlations are generated to understand how the
are then obtained from the microtremor recordings vertical component energy at the pseudo source has
at the pseudosource and the receiver. The two partitioned amongst the three components at the
seismometers were spaced 15 m apart for sites 1, 3, receiver.
4 and 5 (Bilson Darko et al. 2020) and 20 m apart Figure 6 displays particle motion plots of the low
for sites 2 and 6 during our secondary testing. Cross- frequency (0.5–1 Hz) orthogonal cross-correlations.
correlations are computed using a 3-s time delay For sites 1 and 4 (Fig. 6A and B), the horizontal plane
and averaged over the full duration microtremor (ZT-ZR) plots show the lowest horizontal amplitude
recording. We cross correlate the vertical (Z) motions of the six sites, indicating minimal SV/SH/

Vol:. (1234567890)
13
J Seismol (2024) 28:345–372 355

Fig. 7  Cross-plots of mid frequency (1–5 Hz) cross-correla- site’s MHVSR curve (black) compared to the five other site’s
tions grouped by similar response: A–C sites 1, 3 and 4, and MHVSR (grey), and green shading highlights the 1–5-Hz fre-
D–F sites 2, 5 and 6 respectively. Uppermost panels show the quency band

Love wave contributions. Most of the pseudosource ⁓2f0HV trough) orthogonal cross-correlations. Sites
energy remains in the vertical direction, more so for 1 and 3 (Fig. 7A and B) display vertically polarized
site 4 than 1, with little energy transferring to radial motions indicative of P wave motion or vertically
(see ZZ vs ZR plot) and transverse (see ZZ vs ZT dominant Rayleigh wave elliptical motion, the latter
plot) direction, indicating a dominance of Rayleigh/P- of which is more likely. For an ellipticity ratio,
waves. The crossplots for sites 2 and 6 not only are the amplitude is maximum at f0HV and minimum
most similar to each other (Fig. 6C and D), but also at 2f0HV; plots would therefore be dominated by
have similarities to sites 1 and 4 with the highest vertical energy present at the ellipticity minimum’s
amplitude motions in the vertical direction. However, frequency. For site 4 (Fig. 7C), more energy
more of the energy at sites 2 and 6 partitions into the has partitioned into the horizontal components
horizontal plane (relatively equal in both horizon- particularly in the radial direction, indicating the
tal directions) leading to motions which indicate as wavefield is dominated by Rayleigh waves. At this
a mixture of Rayleigh, SV and SH/Love waves. The mid-frequency band, we had expected the cross-
cross-plots for sites 3 and 5 (Fig. 6E and F) have plots at sites 3 and 4 to be most similar to each other
the highest amplitudes in the radial direction sug- with their consistent MHVSR shape. For sites 2, 5
gesting contribution from polarized SV waves. The and 6 (Fig. 7D to F), more energy is partitioned into
MHVSR amplitude at these two sites is the highest at the transverse direction indicating contributions
the low frequencies compared to all other sites, also from Love waves and/or SH waves in addition to
consistent with body wave contributions below f0HV Rayleigh waves. The broadened f0HV shape for site
(Fig. 2B). 6 is consistent with increased SH wave content or
Figure 7 shows particle motion plots of the mid- more similarity to a theoretical transfer function
frequency (1–5 Hz, spanning the f0HV peak and (Fig. 2B).

Vol.: (0123456789)
13
356 J Seismol (2024) 28:345–372

Fig. 8  Cross-plots of high frequency (5–15 Hz) cross-cor- site’s MHVSR (black) compared to the five other site’s MHVSR
relations grouped by similar response: A, B sites 1 and 6 and (grey), and red shading highlights the 5–15-Hz frequency band
C–F sites 2 to 5 respectively. Uppermost panels show the

Figure 8 shows particle motion plots of the high (Sager et al. 2018), we have confidence in our conclu-
frequency (5–15 Hz, above f0HV) orthogonal cross- sions as they are consistent with earlier microtremor
correlations. Rayleigh waves (elliptical disc in ZZ-ZR wavefield contribution findings (e.g. Albarello and
plane) are present in all cases. Sites 1 and 6 (Fig. 8A Lunedei 2011).
and B) exhibit elliptical motions or a mixture or Ray-
leigh and SH waves, whereas for sites 2 to 5 (Fig. 8C
to F), the highest amplitudes occur in the ZZ-ZR 7 Impact to seismic site characterization
plane indicating Rayleigh wave dominance.
Overall, the particle motion plots of the cross-cor- Previous sections examined the contributions of
relations at the six Windsor sites (Figs. 6, 7, and 8) different wave types in the recorded microtremor
indicate a mixture of body and surface waves below wavefield at the six Windsor sites. In this section,
f0HV, other than for sites 3 and 5. At mid frequen- we investigate the variability in inverted VS models
cies (spanning f0HV and 2f0HV), Rayleigh or P-waves from inversion of the average MHVSR spectrum
dominate sites 1, 3 and 4, and sites 2, 5 and 6 are at the six Windsor sites due to the chosen forward
dominated both by body and surface waves. Since amplification model. This variance will be evaluated
sites 1, 3 and 4 are farther from the dominant micro- in terms of impact to seismic site characterization
tremor source (road expressway), Rayleigh waves for four selected metrics: (a) resonator (half-space)
would likely dominate as P waves will have attenu- depth, DR; (b) resonator VS, VSR; (c) the velocity
ated more at these higher frequencies. Above 2f0HV, ratio at the top of the resonator, CV = VSR/VSR-1;
there is dominance of surface waves at all sites. It and (d) the time-averaged VS of the top 30 m, VS30.
should be noted that although retrieval of pure body In addition, we evaluate how our understanding
waves using cross-correlations can be problematic of the microtremor wavefield at the Windsor sites

Vol:. (1234567890)
13
J Seismol (2024) 28:345–372 357

impacts site characterization, i.e. comparison of the MHVSR inversions is that of the resonator (half-
default-assumed fundamental mode Rayleigh wave space) and its depth (Fig. 15 of the Appendix).
ellipticity (R0) inverted VS model with that of the We initiate Monte Carlo sampling with the HVInv
best-fit inverted VS models of f0HV (peak best, Pb, program which switches to optimized simplex down-
model) and the entire MHVSR curve (curve best, hill sampling as the inversion converges. HVInv cal-
Cb, model) amongst 20 other inverted VS models. culates the misfit as,
Our aim therefore is to evaluate fitness of f0HV,
∑N (MHVSRi − H∕V i )2
the shape around f0HV (peak flanks), as well as misfit = (2)
the entire MHVSR curve shape with varying wave i=1 2
𝜎MHVSR
i
contributions in the forward model.
The current state-of-practice of inverting MHVSR where N is the number of frequency samples, H/V
data for subsurface velocity (depth profile) mod- is the theoretical amplification function calculated
els primarily uses f0HV (either as a single frequency using the DFA, and σ is one standard deviation of the
value, or as several amplification values with fre- MHVSR. The total number of models generated for
quency that define the peak), which may also involve each inversion varies between 10,000 and 15,000. We
higher peak frequencies (when applicable) and less calculate post-inversion fitness using the natural loga-
commonly the entire MHVSR spectrum because the rithm of the modelled amplification function, H/V(f),
appropriate forward amplification model to predict and the natural logarithm of the MHVSR spectral data
the MHVSR (physical basis of the microtremor wave- as (Ikeda et al. 2013),
field) for the site is often unknown. We test the fitness ∑N � �
of the empirical MHVSR data (entire curve) at the six 1 i=1
ln MHVSRi ∙ ln(H∕V i )
fitness =
Windsor sites for 21 different amplification forward N2

∑N � � 2 ∑N � � 2
( i=1 (ln MHVSRi )) ( i=1 (ln H∕V i ))
models with wavefield combinations that include
(3)
equal contributions of either Rayleigh waves (R mod-
els), Rayleigh and Love waves (R + L models) or Ray- where N is the number of frequency samples. The
leigh, Love and body waves (R + L + BW models) as maximum fitness of the 21 amplification models is
well as the fundamental (0) and first two higher sur- evaluated in terms of Eq. 3 for both the full MHVSR
face wave modes (1, 2). curve (0.5 to 15 Hz, except 1 to 15 Hz for site 5) and
To perform the suite of 21 MHVSR inversions limited to the mid-frequency bandwidth that includes
for each of the six Windsor sites with varying wave f0HV (1 to 5 Hz) to determine the Cb and Pb models,
contributions in the amplification forward model, respectively, for each site.
we use the HVInv algorithm of García-Jerez et al.
(2016). The model parameterization consists of four 7.1 Wavefield contributions that best fit MHVSR
uniform velocity layers including the elastic half- curves with a tall and narrow ­f0HV
space corresponding to surficial soil, soil and glacial
till layers over the half-space. For all four layers, Figure 9 shows the fitness of the suite of 21 minimum
Poisson’s ratio varies from 0.2 to 0.45 and density misfit amplification models with varying wavefield
is held constant at 2000 kg/m3. We constrain the a contributions (A to C) in comparison to Windsor site
priori VS and soil thickness (h) parameter bounds 1’s MHVSR data and in terms of its residual (H/V
for the upper three layers based on inversion of the model - MHVSR data) with frequency. The corre-
site’s dispersion curve (see Fig. 14 in the Appendix). sponding VS depth profile of the 21 minimum misfit
Inversion of the site’s dispersion data provided models for Windsor site 1 is also shown in Fig. 9. We
by Bilson Darko et al. (2020) was accomplished maintain the use of low-frequency (below f0HV), mid-
using the Dinver algorithm of the Geopsy software frequency (includes f0HV) and high-frequency (above
(Wathelet et al. 2020) with relatively wide model f0HV) bandwidths of the previous section for discus-
parameter bounds. For the 21 MHVSR inversions, sion (dashed vertical lines in Fig. 9). The differ-
wide parameter bounds are used for the top depth ence in peak amplifications of models and MHVSRs
(DR) and VS (VSR) of the resonator. Hence, the appears to be large for some models, owing to high
greatest freedom of model parameters in the 21 and narrow amplification at f0HV that some models

Vol.: (0123456789)
13
358 J Seismol (2024) 28:345–372

Fig. 9  Comparison of MHVSR data (circles) and 21 amplifica- for Windsor site 1. The corresponding VS depth profile of the
tion models (coloured lines) and their residual with frequency 21 models is also shown. Pb and Cb refer to peak best and
considering A Rayleigh, B Rayleigh and Love and C Rayleigh, curve best models respectively determined by Eq. 3
Love and body wave contributions in the amplification model

have difficulty in fitting. It is readily apparent that The addition of fundamental mode Love waves
the default assumed R0 amplification model fits the (R0L0) leads to the highest fitness of the f0HV peak
narrow and high amplification of the f0HV peak. The (Pb model) by improving fitness of the peak’s flanks
inclusion of higher Rayleigh wave modes (R1, R2; while reducing fitness of the peak’s maximum. At low
Fig. 9A) produce higher amplification within the nar- frequencies, Rayleigh and Love wave contribution
rowest portion of the f0HV peak maximum in better amplification models best fit the empirical amplifica-
agreement with the MHVSR data but with reduced tion (Fig. 9B), which is overestimated if body waves
fitness within the mid-frequency bandwidth, i.e. the are included (Fig. 9C). At high frequencies, amplifi-
peak’s maximum is fit more than the entire peak. cation fitness appears to be improved mostly by the
There is an observable trend of decreasing fitness addition of Love waves (Fig. 9B and C compared to
of the f0HV peak at high amplification with the addi- Fig. 9A). Amongst all the 21 models, R1L2 is the Cb
tion of Love wave (some models; Fig. 9B) to Love model and provides the highest fitness of the entire
and body wave contributions (all models; Fig. 9C). MHVSR curve.

Vol:. (1234567890)
13
J Seismol (2024) 28:345–372 359

Fig. 10  MHVSR inversion results for Windsor site 5; same format as Fig. 9

The MHVSR fitness results of site 1 are generally noting higher mode contributions are not evaluated
repeated for sites 2, 3 and 4 with high and narrow f0HV by our microtremor wavefield analyses in Sections 6
amplification (see Figs. 16, 17, and 18 in the Appen- and 7.
dix), noting Rayleigh wave contributions best fit f0HV
for all sites (Pb model), and with (sites 2 and 3) and 7.2 Wavefield contributions that best fit MHVSR
without (site 4) the need of Love wave contributions curves with an anomalously shaped ­f0HV peak
to best fit the entire MHVSR curve (Cb model). Over-
all, the fitness of the 21 amplification models is con- Figure 10 shows the suite of 21 MHVSR inversion
sistent with our initial assumption that sites 1, 3 and results with varying wavefield contributions (A to C)
4 are most Rayleigh ellipticity-like (should be mod- for site 5. The anomalously shaped f0HV peak is best
elled as R0) as well as our further understanding of fit with both Rayleigh and Love wave contributions
the microtremor wavefield composition at site 1 from (Pb model in Fig. 10B), and the entire MHVSR curve
earlier sections: f0HV and mid-frequency bandwidth is best fit when all wave types contribute (Cb model
is dominated by Rayleigh waves, but Love wave con- in Fig. 10C). The f0HV left shoulder is present in some
tributions are likely present over the entire MHVSR of the theoretical amplification functions when Love
spectrum. We note the possibility that the high ampli- waves and body waves are added (Fig. 10B and C).
fication of f0HV may be the result of higher Rayleigh The suite of corresponding VS profiles is not signifi-
mode contributions (see also site 2 results in Fig. 16), cantly different suggesting this morphological change

Vol.: (0123456789)
13
360 J Seismol (2024) 28:345–372

Fig. 11  MHVSR inversion results for Windsor site 6; formatting as in Fig. 9

to site 5’s f0HV is driven primarily by the addition of curve is best fit when all wave types contribute,
Love or Love and body waves rather than merging noting the fit is very similar amongst these models,
of two narrow resonance modes driven by seismic i.e. the fitness of Pb and Cb models is similar due to
impedance layering (discussed further for site 6). greater symmetrical shape of site 6’s MHVSR. The
R0 model does not fit the f0HV amplification as well
7.3 Wavefield contributions that best fit MHVSR as models that include other wave contributions. For
curves with a symmetrical and broader ­f0HV some models, the central f0HV peak is modelled as
two narrow peaks, e.g. Rayleigh wave models with
The MHVSR inversion results with varying wavefield higher modes included (Fig. 11A) and some models
contributions (A to C) for site 6 are shown in Fig. 11. that include all wave types (Fig. 11C). It is therefore
Residuals for site 6 are the lowest compared to possible to explain the broadened f0HV shape for
other sites, due to the symmetry in the f0HV peak’s site 6 as result of additional Love and body wave
flanks. Similar to site 5, the anomalous broad but contributions and/or contributions by higher Rayleigh
symmetrical f0HV peak is best fit with both Rayleigh modes or the presence of two narrow resonance
and Love wave contributions and the entire MHVSR modes driven by seismic impedance layering (i.e.

Vol:. (1234567890)
13
J Seismol (2024) 28:345–372 361

Table 2  Reporting of the four select seismic site characteriza- overestimation of the R0 model; negative differences indicate
tion metric’s values and their comparison in terms of the dif- underestimation of the R0 model (shown with grey shading).
ference and relative percentage change (%) between the default The last row reports the average relative percentage change (%)
assumed R0 model and the Pb model, Cb model or the aver- between the R0 model and the other 20 models for all possible
age (Ave, and one standard deviation (SD)) of all other 20 120 model comparisons of the six Windsor sites; one SD and
models for the six Windsor sites. Positive differences indicate the coefficient of variation (cov) are also reported

CV VS30 Comparison with R0 model (difference, %)


Case DR (m) VSR (m/s)
(VSR/VS3) (m/s) DR VSR CV VS30
Site 1 R0 108 1925 2.4 316
Pb: R0L0 74 1781 2.7 344 34, 31 144, 8 0.3, 13 -28, 9
Cb: R1L2 66 1312 1.9 302 42, 39 613, 32 -0.5, 21 14, 4
Ave_20 (1 SD) 69 (11) 1620 (370) 2.5 (0.5) 315 (20) 39, 36 305, 16 -0.1, 4 1, 0
Site 2 R0 79 2738 4.2 380
Pb: R2 75 2072 3.0 339 4, 5 666, 24 1.2, 29 41, 11
Cb: R2L1 70 1875 2.7 346 9, 11 863, 32 1.5, 36 34, 9
Ave_20 (1 SD) 64 (8) 1781 (290) 2.7 (0.4) 360 (24) 15, 19 957, 35 1.5, 37 20, 5
Site 3 Pb: R0 121 1726 2.3 269
Cb: R1L2 61 1142 1.8 324 60, 50 584, 34 0.5, 22 -55, 20
Ave_20 (1 SD) 82 (26) 1415 (438) 2.2 (0.7) 290 (28) 39, 32 311, 18 0.1, 4 -21, 8
Site 4 Pb, Cb: R0 97 2752 3.6 279
Ave_20 (1 SD) 74 (14) 1537 (553) 2.0 (0.7) 256 (27) 23, 24 1215, 44 1.6, 44 23, 8
Site 5 R0 111 1939 2.6 325
Pb: R1L1 76 1221 1.9 303 35, 32 718, 37 0.7, 27 22, 7
Cb: R0L1BW 82 1635 2.6 297 29, 26 304, 16 0, 0 28, 9
Ave_20 (1 SD) 82 (10) 1513 (413) 2.3 (0.5) 303 (14) 29, 27 426, 29 0.3, 12 22, 7
Site 6 R0 65 1078 2.3 249
Pb: R1L1 48 939 2.1 241 17, 26 139, 13 1, 14 8, 3
Cb: R2L1BW 46 1059 2.4 240 19, 29 19, 2 -0.1, 4 9, 4
Ave_20 (1 SD) 48 (9) 974 (128) 2.2 (0.2) 243 (10) 17, 26 104, 10 0.1, 4 6, 2
Average relave percentage change, SD and cov for the 6 sites 28, 14, 0.5 28, 16, 0.58 24, 18, 0.73 7, 6, 0.86

models that generate two narrow peaks in Fig. 11C 7.4 Choice of theoretical amplification model’s
have similar VS profiles that are different from the impact to seismic site characterization
other model’s VS profiles). Site 6’s dispersion data
(Fig. 14) suggests a stiff near surface layer with The impact to seismic site characterization in terms
apparent mode velocities at the highest frequencies of the four chosen metrics (DR, VSR, CV, and VS30) for
but does not suggest strong contributions by higher the R0, Pb and Cb models for all six Windsor sites is
modes at mid-frequencies or to the f0HV. The change reported in Table 2. In addition, Table 2 reports the
in MHVSR shape at site 6 can be explained by difference and relative percentage change of these
additional wave contributions, consistent with earlier four seismic site characterization metrics between the
interpretations in Sections 6 and 7, but we do not default assumed R0 model and the (1) Pb model, (2)
rule out that the broadness is driven by merging of Cb model and (3) average of the 20 models for each
resonance frequencies. site. Overall, the R0 amplification model’s VS profile

Vol.: (0123456789)
13
362 J Seismol (2024) 28:345–372

Fig. 12  R0, Pb and Cb inversion results for sites 1 to 3 (A to C); formatting as in Fig. 9

determines the deepest and often the highest VS of the default assumed R0 forward model), DR and VSR are
resonator (half-space) consistently amongst the suite likely overestimated by an average of 36% and 16%
of 21 amplification models for the six sites. For site respectively (Table 2). The impact on CV and VS30 for
1, amplification models that best fit f0HV (Pb model) all the six sites will be discussed together later in this
and the entire MHVSR curve (Cb model) include section. This overestimation of DR and VSR by the R0
Love wave contributions (Fig. 12A). In other words, model compared to the Pb and Cb models is generally
the default assumed R0 model for this site may over- consistent for all the sites (Figs. 12 and 13; Table 2).
estimate DR by 31% and 39% and VSR by 8% and 32% The Pb model (site 3; Fig. 12C) and both the Pb and
respectively (Table 2). Had we applied our under- Cb models (site 4; Fig. 13A) are the R0 model, as
standing of wavefield contributions at this site (out- expected from earlier analyses of the MHVSR shape
comes from earlier Sections 6 and 7), we would have and wavefield contributions at these two sites. The
modelled the MHVSR as the R0L0 model (Rayleigh fundamental mode Rayleigh ellipticity (R0) amplifi-
and Love wave contributions with default-assumed cation model assumption is applicable for sites 3 and
fundamental mode contributions); this model 4. Had we blindly inverted site 5’s and 6’s MHVSR as
improves fitness of f0HV compared to the R0 model R0 amplification, we would overpredict DR and VSR
while forfeiting fitness of the f0HV peak maximum, as on average by 27% or 26% and 29% or 10% respec-
desired (R0L0 is the Pb model). For site 1, the impact tively. This is expected since the Pb model includes
to seismic site characterization is that if wavefield Love wave contributions and the Cb model includes
contributions are not accounted for in the forward Love and body wave contributions which are signifi-
amplification model (comparison of the average of cant departures from the R0 model assumption for
all 20 amplification forward model solutions with the these two sites (Fig. 13B and C).

Vol:. (1234567890)
13
J Seismol (2024) 28:345–372 363

Fig. 13  R0, Pb and Cb inversion results for sites 4 to 6 (A to C); formatting as in Fig. 9

We do not observe a consistent impact to ­ CV seismic site characterization when wavefield contri-
amongst the six Windsor sites; the average percent- butions are ignored in the inversion of MHVSR data.
age change in ­CV can be quite high (e.g. 37% and Combining the results for all six sites (bottom row
44% for sites 2 and 4 respectively) or low (e.g. 4% of Table 2), we observe that the largest variance in
for sites 1 and 6), owing to the high variance in VSR the average percentage change compared to the R0
amongst the inversion models. Since this variation model amongst the six Windsor sites occurs in the
in the half-space depth is at depths greater than 30 m VSR (28% ± 16%) and DR (28% ± 14%) metrics, fol-
for all the sites, the impact to VS30 should be 0% (e.g. lowed closely by CV (24% ± 18%). This indicates
site 1; Table 2) but can be as much as 20% (Table 2) that the largest impact to seismic site characteriza-
confirming impact to the VS depth profile above the tion due to the unknown forward model solution is
resonator layer even with constrained a priori model the depth and VS of the resonator (half-space), which
parameters of the upper three layers. We note that also influences CV. Not surprisingly, the VS30 metric
VS30 variance will be less than the other three chosen has the lowest average percentage change (7% ± 6%),
metrics because it is an average measure and that a i.e. resonator VS and depth are often deeper than 30 m
variance in VS30 over ~ 4% (the variance within and and do not play a significant role in the VS30 at these
between different in situ VS profiling methods; Garo- Windsor sites. In terms of the metric’s statistical dis-
falo et al. 2016) is significant. persion (coefficient of variation, cov = SD/Ave), the
In practice, we do not expect 21 forward model largest dispersion corresponds to VS30 (0.86) and then
solutions to be accomplished to find the MHVSR CV (0.73). The cov in the VSR and DR metrics (0.58
curve’s best fit amplification (Cb) model; however, and 0.5) is larger than dispersion between various
our intent is to determine the potential impact to in situ VS(z) profiling methods (0.2; Garofalo et al.

Vol.: (0123456789)
13
364 J Seismol (2024) 28:345–372

2016) and is also therefore significant. This says that function and reinforces that the microtremor wave-
varying the wave contributions of the forward ampli- field is dominated by shear waves. In contrast, all
fication model (21 tested possible models per site) other sites require two peaked mathematical forms to
will impact seismic site characterization in terms of fit f0HV in combination with a linear offset to fit the
the resonator VS and depth, and although the variance entire MHVSR curve which reinforces that the micro-
in VS30 is small, it is significant or impactful. tremor wavefield is dominated by surface waves (pri-
marily Rayleigh waves) in the mid-frequency f0HV
bandwidth with increasing Love and/or body wave
8 Conclusions contributions at other frequencies. Although the input
data for this type of analysis is relatively simple (the
In this study, changes in MHVSR shape were related empirical MHVSR), the percentage contribution of
to the wave types present in the microtremor record- specific wave types is not determined, and the method
ings at six sites in Windsor, Ontario. The MHVSR’s is therefore relatively lacking for our intended goal.
lowest frequency peak amplification (f0HV) is ~ 2 Hz This technique could grow in popularity if the under-
amongst the six sites, consistent with the relatively lying mathematical function(s) of particular wave
homogeneous geologic setting, yet f0HV amplification type contributions are better defined in future, i.e. as
is higher and narrower for sites 1, 3 and 4 compared practical as its application to identify chemical com-
to sites 2, 5 and 6. The quantification of the f0HV peak position of materials in Raman spectroscopy.
broadness amongst the six Windsor sites conveys that We then proceeded to a two-seismometer tech-
the f0HV peak at site 6 is ~ 5.5 times broader than sites nique that involved examination of particle motion
3 and 4. We then proceeded to develop novel single- plots of the orthogonal microtremor cross-correla-
seismometer and two-seismometer techniques to fur- tions to further differentiate wave contributions in
ther elucidate the potential wave contributions in the the microtremor wavefield at the six Windsor sites.
microtremor recordings at the six Windsor sites with Below f0HV, our analysis confirmed a mixture of body
the practical intent to determine the most applicable and surface waves at most Windsor sites; sites 3 and 5
theoretical amplification (forward) model to invert display the highest amplification which could be gen-
the MHVSR. This is in contrast to existing techniques erated from polarized SV waves. At mid frequencies
to remove non-Rayleigh wave contributions for the (spanning f0HV and 2f0HV), either Rayleigh or P-waves
inversion of MHVSR spectra as theoretical Rayleigh (sites 1, 3, 4) or body and surface waves (sites 2, 5, 6)
ellipticity functions (e.g. Hobiger et al. 2009; Berbel- dominated. Above 2f0HV, there is dominance of sur-
lini et al. 2019). face waves at all sites. The contribution percentage of
We first proposed quantification of the MHVSR individual seismic wave types is not possible in this
itself (single-seismometer method) in terms of the study. The wave contributions and thereby the most
number and shaped nature of different mathematical appropriate theoretical (forward) amplification model
functions (linear, Lorentzian, Gaussian) to identify to invert the f0HV peak or the entire MHVSR spectrum
unique MHVSR shapes due to seismic wave type(s) at each site were identified.
embedded within the empirical MHVSR spectrum. We chose to test post-inversion fitness of the f0HV
When SH waves dominate, the MHVSR amplifica- peak or the entire MHVSR curve at the six Windsor
tion curve is mostly symmetrical and best fit by a sites using 21 theoretical amplification models based
single peaked distribution (Gaussian or Lorentzian) on different combinations of wave contributions: Ray-
with a linear offset of (near) zero slope. When sur- leigh waves, Rayleigh and Love waves or Rayleigh,
face waves dominate, the MHVSR curve is asym- Love and body waves as well as up to two higher sur-
metric and more than one peaked functional form is face wave modes, where equal contributions come
required to fit the MHVSR at f0HV and below, with the from each wave type and their respective modes. As
need of a sloped linear offset above f0HV to account a future scope of work, differing wave contributions
for increasing amplification due to Love and/or body can be assumed. From this suite of possible amplifi-
wave contributions. Hence, site 6’s broad but sym- cation solutions, we could then evaluate the impact to
metrical f0HV is fit by a single peaked mathematical seismic site characterization in terms of the difference

Vol:. (1234567890)
13
J Seismol (2024) 28:345–372 365

and relative percentage change of four chosen metrics to obtain correct thickness of the sediment layer and
(DR, VSR, CV and VS30) between the default-assumed resonator VS and thereby VS30.
fundamental mode Rayleigh wave ellipticity (R0)
model with that of the peak and curve best mod- Abbreviations 3D Three-dimensional; A0 Peak
els (Pb and Cb model respectively). The R0 model amplification; B MHVSR Peak broadness; BW Body
assumption is applicable only for sites 3 and 4; R0 waves; Cb Curve (full MHVSR) best amplification
is the Pb (both sites) and Cb (site 4) model. For all model; CV Ratio of the Vs between the layer above
other sites, some combination of Rayleigh waves with and the resonator (half-space) layer; DR Depth
Love and/or body waves best fits f0HV and the entire (D) of the resonator (R) layer; DFA Diffuse wave-
MHVSR curve; the default-assumed Rayleigh elliptic- field assumption; f0 Theoretical fundamental mode
ity function (R0) forward model would be wrongfully (resonance) site frequency; f0HV Fundamental mode
applied in the inversion of the MHVSR data for these MHVSR peak frequency; h Soil thickness; HRFK
sites. High-resolution frequency-wavenumber; H/V Hor-
In terms of impact to seismic site characterization, izontal-to-vertical ratio; L waves Love waves; L0
the R0 model’s VS profile determines the deepest and Fundamental mode Love wave; L1 First mode Love
often the highest VS of the resonator (half-space) con- wave; L2 Second mode Love wave; MHVSR Micro-
sistently amongst the suite of 21 amplification models tremor horizontal-to-vertical spectral ratio; P waves
for the six sites. In other words, DR (and VSR) is sub- Primary (compressional) waves; Pb Peak (f0HV) best
stantially overestimated by the wrongfully applicable amplification model; R waves Rayleigh waves; R0
R0 model in fitting f0HV or the entire MHVSR curve as Fundamental mode Rayleigh wave; R1 First mode
much as 32% (and 37%) and 39% (and 32%) respec- Rayleigh wave; R2 Second mode Rayleigh wave; S
tively. Had we blindly inverted each site’s MHVSR as waves Secondary (shear) waves; SH waves Trans-
R0 amplification, we would overpredict the four cho- verse shear waves; SV waves Radial shear waves;
sen metrics by an average relative percentage change VS Shear wave velocity; VS30 Time-averaged VS
of 28% (DR), 28% (VSR), 24% (CV) and 7% (VS30). of the top 30 m; VSav Average VS of the soil layer;
Although we do not recommend the practitioner to VSR VS of the resonator (R, half-space) layer; ZR
perform 21 different inversions with varying wave Cross-correlation of the pseudosource’s vertical (Z)
contributions, we do note that comparison of the component recordings and the receiver’s radial (R)
suite of inverted models was informative and gener- component recording; ZT Cross-correlation of the
ally supported previous interpretations of the under- pseudosource’s vertical (Z) component recordings
lying wave contributions, e.g. the inclusion of higher and the receiver’s transverse (R) component record-
Rayleigh wave modes or Love waves improved fit- ing; ZZ Cross-correlation of the pseudosource’s
ness of the maximum amplification, the inclusion of vertical (Z) component recordings and the receiv-
Love and/or body waves improved fitness of the f0HV er’s vertical (Z) component recording
peak while sacrificing fitness of its maximum ampli-
fication and the inclusion of Love and/or body waves Acknowledgements We thank Dr. Gerhard Pratt for his sug-
gestion to perform particle motion analysis of cross-correla-
improved fitness of the entire MHVSR curve. The tions from microtremor recordings. We acknowledge Alex
practitioner may also not have access to surface wave Bilson Darko for supplying microtremor recordings and surface
dispersion data as was used in this study to reduce wave dispersion curve data for the Windsor array sites. The
model non-uniqueness of the MHVSR inversion. manuscript was improved by two anonymous reviewers.
Other near surface information to constrain h or VS
Author contribution Hema Sharma (HS) and Sheri Mol-
(i.e. Equation 1) should be utilized to reduce MHVSR nar (SM) conceptualized the presented work and methodolo-
model non-uniqueness. Our study determined a sig- gies. HS performed all analyses under the supervision of SM.
nificant effect of the chosen forward amplification Aamna Sirohey (AS) performed the analyses of Section 6 and
model in the MHVSR inversion (f0HV or entire spec- helped to draft its writing. HS and SM wrote and edited the
manuscript. SM acquired the funding for this research work.
trum) to the subsequent VS depth profile and demon-
strates the importance of correctly identifying wave Funding Funding for this research was provided by the
type contributions of the microtremor wavefield for Natural Sciences and Engineering Research Council (NSERC)
the proper estimation of VS depth profiles, especially (RGPIN-2016–04634).

Vol.: (0123456789)
13
366 J Seismol (2024) 28:345–372

Appendix

Fig. 14  Input dispersion curves (circles) for array sites 1 (A) misfit model (the VS profile of this minimum misfit model are
to 6 (F) shown in comparison to the theoretical fundamental- shown in Fig. 15) for sites 1 to 6 (A to F)
mode Rayleigh dispersion curve (solid line) of the minimum

Vol:. (1234567890)
13
J Seismol (2024) 28:345–372 367

Fig. 15  A priori VS and h parameter bounds used in the 21 minimum misfit model is coloured grey at and below the cal-
MHVSR inversions (thin grey lines) for array sites 1 (A) to 6 culated resolution depth of the dispersion data, i.e. depth ≈
(F) based the minimum misfit model (black line) from inver- wavelength of the lowest frequency dispersion datum divided
sion of the site’s dispersion curve. The lower portion of the by 3 (Asten and Henstridge 1984)

Vol.: (0123456789)
13
368 J Seismol (2024) 28:345–372

Fig. 16  MHVSR inversion results for Windsor site 2; formatting as in Fig. 9

Vol:. (1234567890)
13
J Seismol (2024) 28:345–372 369

Fig. 17  MHVSR inversion results for Windsor site 3; formatting as in Fig. 9

Vol.: (0123456789)
13
370 J Seismol (2024) 28:345–372

Fig. 18  MHVSR inversion results for Windsor site 4; formatting as in Fig. 9

Data availability Data are available upon request. References


Declarations Aki K, Richards PG (1980) Quantitative seismology—theory
and methods. Freeman Company, San Francisco
Ethics approval and consent to participate Not applicable. Albarello D, Lunedei E (2011) Structure of an ambient vibra-
tion wavefield in the frequency range of engineering inter-
est ([0.5, 20] Hz): insights from numerical modelling.
Consent for publication Not applicable. Near Surface Geophysics 9:543–559. https://​doi.​org/​10.​
3997/​1873-​0604.​20110​17
Competing interests The authors declare no competing interests. Arai H, Tokimatsu K (2004) S-wave velocity profiling by
inversion of microtremor H/V spectrum. Bull Seism Soc
Am 94(1):53–63
Asten MW, Henstridge JD (1984) Array estimator and the use
of microseisms for reconnaissance of sedimentary basins.
Geophysics 49(11):1828–1837
Figs. 14,15,16,17,18.

Vol:. (1234567890)
13
J Seismol (2024) 28:345–372 371

Bard P-Y et al (2004) The SESAME project: an overview and Hudec PP (1998) Geology and geotechnical properties of gla-
main results. Proceedings of the 13th World Conference cial soils in Windsor. In: Karrow PF, White OL (eds)
in Earthquake Engineering, Vancouver, Canada, August Urban geology of Canadian cities, vol 42. Geological
2004, Paper # 2207 Association of Canada - Special Publication, pp 225–236
Ben-Hador R, Buchen P (1999) Love and Rayleigh waves in Ikeda T, Asten M, Matsuoka T (2013) Joint inversion of spa-
non-uniform media. Geophys J Int 137:521–534 tial autocorrelation curves with HVSR for site character-
Berbellini A, Schimmel M, Ferreira AMG, Morelli A (2019) ization in Newcastle, Australia. ASEG Extended Abstr
Constraining S-wave velocity using Rayleigh wave ellip- 1:1–4. https://​doi.​org/​10.​1071/​ASEG2​013ab​315
ticity from polarization analysis of seismic noise. Geo- Konno K, Ohmachi T (1998) Ground-motion characteristics
phys J Int 216:1817–1830 estimated from spectral ratio between horizontal and
Bilson Darko A, Molnar S, Sadrekarimi A (2020) Blind com- vertical components of microtremor. Bull Seism Soc
parison of non-invasive shear wave velocity profiling with Am 88(1):228–241
invasive methods at bridge sites in Windsor, Ontario. Soil Lachet C, Bard P-Y (1994) Numerical and theoretical inves-
Dyn Earthq Eng 129. https://​doi.​org/​10.​1016/j.​soild​yn.​ tigations on the possibilities and limitations of Nakamu-
2019.​105906 ra’s technique. J Phys Earth 42:377–397
Bonnefoy-Claudet S, Kohler A, Cornou C, Wathelet M, Bard Molnar S et al (2022) A review of the microtremor hor-
P-Y (2008) Effects of Love waves on microtremor H/V izontal-to-vertical spectral ratio (MHVSR) method.
Ratio. Bull Seism Soc Am 98(1):288–300 J Seismolog 26:653–685. https://​doi.​org/​10.​1007/​
Borcherdt RD (1970) Effects of local geology on ground s10950-​021-​10062-9
motion near San Francisco Bay. Bull Seism Soc Am Nakamura Y (1989) A method for dynamic characteris-
60:29–61 tics estimation of subsurface using microtremor on the
Castellaro S (2016) The complementarity of H/V and disper- ground surface. Q Rep Railw Tech Res Inst (RTRI)
sion curves. Geophysics 81(6):T323–T338 Japan 30:25–33
Castellaro S, Mulargia F (2009) The effect of velocity inver- Nogoshi M, Igarashi T (1971) On the amplitude characteris-
sions on H/V. Pure Appl Geophys 166:567–592. https://​ tics of ambient noise (part 2). J Atmos Oceanic Technol
doi.​org/​10.​1007/​s00024-​009-​0474-5 24:26–40
Colombi A, Boschi L, Roux P, Campillo M (2014) Green’s Piña-Flores J, Cárdenas-Soto M, García-Jerez A, Seivane H,
function retrieval through cross-correlations in a two- Luzón F, Sánchez-Sesma FJ (2020) Use of peaks and
dimensional complex reverberating medium. J Acoust Soc troughs in the horizontal-to-vertical spectral ratio of ambi-
Am 135(3):1034–1043. https://​doi.​org/​10.​1121/1.​48644​85 ent noise for Rayleigh-wave dispersion curve picking.
Easton RM, Carter TR (1995) Geology of the Precambrian J Appl Geophys 177. https://​doi.​org/​10.​1016/j.​jappg​eo.​
Basement Beneath the Paleozoic of Southwestern 2020.​104024
Ontario. In: Ojakangas RW, Dickas AB, Green JC (eds) Poggi V, Fäh D (2010) Estimating Rayleigh wave particle
Basement Tectonics 10. Proceedings of the International motion from three-component array analysis of ambient
Conferences on Basement Tectonics, vol 10. Springer, vibrations. Geophys J Int 180(1):251–267
Dordrecht, pp 221–264. https://​doi.​org/​10.​1007/​978-​94-​ Sager K, Boehm C, Ermert L, Krischer L, Fichtner A (2018)
017-​0831-9_​26 Sensitivity of seismic noise correlation functions to global
Fäh D, Kind F, Giardini D (2001) A theoretical investigation noise sources. J Geophys Res: Solid Earth 123:6911–
of average H/V ratios. Geophys J Int 145(2):535–549 6921. https://​doi.​org/​10.​1029/​2018J​B0160​42
García-Jerez A, Piña-Flores J, Sánchez-Sesma FJ, Luzón F, Sánchez-Sesma FJ, Perez-Ruiz JA, Luzon F, Campillo M,
Perton M (2016) A computer code for forward compu- Rodriguez-Castellanos A (2008) Diffuse fields in dynamic
tation and inversion of the H/V spectral ratio under the elasticity. Wave Motion 45:641–654
diffuse field assumption. Comput Geosci 97:67–78 Schimmel M, Gallart J (2004) Degree of polarization filter for
Garofalo F et al (2016) InterPACIFIC project: comparison frequency-dependent signal enhancement through noise
of invasive and non-invasive methods for seismic site suppression. Bull Seism Soc Am 94(3):1016–1035
characterization. Part II: inter-comparison between sur- Sharma H, Molnar S (2019) Earthquake site characterization
face-wave and borehole methods. Soil Dyn Earthq Eng of southwestern Ontario for microzonation mapping. 12th
82:241–254 Canadian Conference on Earthquake Engineering, Quebec,
Herak M (2008) ModelHVSR – A Matlab tool to model hori- Canada, June 2019, Paper # 192-jKuB-149
zontal-to-vertical spectral ratio of ambient noise. Comput Tapia M, Guinau M, Roig P, Pérez-Guillén C, Suriñach E, Khaz-
Geosci 34(11):1514–1526 aradze G (2020) Seismic data of a rockslide: Evaluation of
Hobiger M, Bard P-Y, Cornou C, Le Bihan N (2009) Single noise levels, site effects, frequency content and identifica-
station determination of Rayleigh wave ellipticity by using tion of seismic phases. Data Brief 29:105250. https://​doi.​
the random decrement technique (RayDec). Geophys Res org/​10.​1016/j.​dib.​2020.​105250
Lett 36:L14303. https://​doi.​org/​10.​1029/​2009G​L0388​63 Uebayashi H (2003) Extrapolation of irregular subsurface struc-
Hobiger M, Cornou C, Bard P-Y, Le Bihan N (2011) tures using the horizontal-to-vertical spectral ratio of long-
MUSIQUE: A quaternion-based array processing tech- period microtremors. Bull Seism Soc Am 93(2):570–582
nique for surface wave polarization analysis. SSP 2011 Wathelet M, Chatelain J, Cornou C, Giulio GD, Guillier B, Ohrn-
- 2011 IEEE Workshop on Statistical Signal Processing, berger M, Savvaidis A (2020) Geopsy: a user-friendly open-
Nice, France, pp 5–8. https://​doi.​org/​10.​1109/​SSP.​2011.​ source tool set for ambient vibration processing. Seismol
59677​70

Vol.: (0123456789)
13
372 J Seismol (2024) 28:345–372

Res Lett 91:1878–1889. https://​doi.​org/​10.​1785/​02201​ Publisher’s Note Springer Nature remains neutral with regard
90360 to jurisdictional claims in published maps and institutional
Yeh Y, Wen S, Lee K, Chen C (2013) Shear-wave velocity model affiliations.
of the Chukuo fault zone, Southwest Taiwan, from cross
correlation of seismic ambient noise. J Asian Earth Sci Springer Nature or its licensor (e.g. a society or other partner)
75:174–182 holds exclusive rights to this article under a publishing
Yuan X, Mayanovic RA (2017) An empirical study on Raman agreement with the author(s) or other rightsholder(s); author
peak fitting and its application to Raman quantitative self-archiving of the accepted manuscript version of this article
research. Appl Spectrosc 71(10):2325–2338. https://​doi.​org/​ is solely governed by the terms of such publishing agreement
10.​1177/​00037​02817​721527 and applicable law.

Vol:. (1234567890)
13

You might also like