Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
8 An HHT-Based Approach
to Quantify Nonlinear
Soil Amplification and
Damping
Ray Ruichong Zhang
CONTENTS
ABSTRACT
This study proposes to use a method of nonlinear, nonstationary data processing and
analysis, i.e., the Hilbert-Huang transform (HHT), to quantify influences of soil non-
linearity in earthquake recordings. The paper first summarizes symptoms of soil non-
linearity shown in earthquake ground motion recordings. It also reviews the Fou-
rier-based approach to characterizing the nonlinearity in the recordings and
demonstrates the deficiencies. It then offers the justifications of the HHT in addressing
the nonlinearity issues. With the use of the 2001 Nisqually earthquake recordings and
results of the Fourier-based approach as a reference, this study shows that the
159
The first HHT-based component and the Hilbert amplitude spectra can identify
abnormal high-frequency spikes in the recording at sites where strong soil
nonlinearity occurs; this can help to detect the nonlinear sites at a glance.
The HHT-based factor for site amplification is defined as the ratio of marginal
Hilbert amplitude spectra, similar to the Fourier-based one that is the ratio of
Fourier amplitude spectra. The HHT-based factor is effective in quantifying soil
nonlinearity in terms of frequency downshift in the low-frequency range and
amplitude downshift in the intermediate-frequency range.
Hilbert and marginal damping spectra are identified in ways similar to Hilbert
and marginal amplitude spectra. Consequently, the HHT-based factor for site
damping is found as the difference of marginal Hilbert damping spectra, which
can be extracted from the HHT-based factor for site amplification and used as
an alternative index to measure the influences of soil nonlinearity in seismic
ground responses.
8.1 INTRODUCTION
Site amplification is the phenomenon in which the amplitude of seismic waves
increases significantly when they pass through soil layers near the earth’s surface.
It can be illustrated by considering the seismic energy flux along a tube of seismic
rays, which is proportional to the impedance (density × wave speed) and squared
shaking velocity. Since the energy should be constant in the absence of damping,
any reduction in the impedance is compensated by an increase in the shaking velocity,
thus yielding site amplification for seismic waves in soil layers. Site amplification
is a key factor in mapping seismic hazard in urban areas (e.g., [1]) and designing
geotechnical and structural engineering systems on soils (e.g., [2]).
In general, site amplification is not linearly proportional to the intensity of input
seismic motion at bedrock because of soil nonlinearity under large-amplitude earth-
quakes. The extent of soil nonlinearity can be characterized by the change of two
dynamic features of soil layer, i.e., soil resonant frequency and damping, in the
frequency-dependent site amplification. Consensus has been building that the
site-amplification factors in the current codes overemphasize the extent of soil
nonlinearity and thus potentially underestimate the level of site amplification. It has
been demonstrated [3] that the recording-based amplification factors are larger than
those in codes for a certain range of base acceleration intensity. In addition, some
features of site-amplification factors used in codes and guidance for structural design
contradict recent findings from the 1994 Northridge ground motion data set [4].
The aforementioned problem might exist partly because seismologists and engi-
neers lack sufficient understanding of the underlying causes in nonlinear soil. For
example, the influence of soil heterogeneity does not scale linearly, even when the
soil is perfectly linear [5]. In other words, a linear elastic medium with random
heterogeneity can change ground motion in a way similar to that caused by medium
N N
X (t ) = ℜ ∑j =1
Aje
iΩ j t
=ℜ ∑[ A sin(Ω t) + iA cos(Ω t)] ,
j =1
j j j j (8.1)
where ℜ denotes the real part of the value to be calculated, i = (–1)∫ is an imaginary
unit, amplitudes Aj are a function of time-independent frequency Ωj that is defined
over the window in which the data is analyzed, and the Fourier amplitude spectrum
is defined as
F (Ω) = ∑A .
j =1
j (8.2)
To apply this Fourier spectral analysis to estimate the influences of soil nonlinearity
in the seismic wave responses at soil site or simply site amplification, two sets of
recordings are typically needed [18], one at a soil site and the other at a referenced
site such as bedrock or outcrop. For a frequency, the Fourier-based factor of site
amplification (FF) for an earthquake event (either mainshock or aftershock) can then
be found by
Fs2,h1 + Fs2,h 2
FFs Ω = ( ) , (8.3)
Fr2,h1 + Fr2,h 2
where subscripts s and r denote respectively the soil and referenced sites, and
subscripts h1 and h2 denote the two horizontal components. Note that Equation 8.3
is one of many representatives for site-amplification factor that can be the ratio of
characteristics of seismic waves or spectral responses at a site versus referenced site.
Since the wave paths and earth structures except the soil layer are almost the
same for the soil and referenced sites, the factor for site amplification in Equation
8.3 eliminates approximately the influences of source from the earthquake event and
thus provides essentially the dynamic characteristics of the soil. In addition, the
recordings at the referenced site are generally believed to be the results of linear
wave responses and the recordings at the soil site subject to the large-amplitude
mainshock to be the results of nonlinear wave responses. Accordingly, comparing
the factors from the mainshock and the aftershock could help us explore and quantify
the influences of soil nonlinearity in site amplification.
While the Fourier-based approach given here and similar methods are widely
used, they have the following deficiencies in characterizing the nonstationarity of
the earthquake motion that is caused by source, different types of propagating waves,
and soil nonlinearity if the earthquake magnitude is large enough.
A Fourier-based approach defines harmonic components globally and thus yields
average characteristics over the entire duration of the data. However, some charac-
teristics of data, such as the downshift of soil resonant frequency at a nonlinear site,
may occur only over a short portion of a record. This is particularly true when the
intensity of the seismic input to a soil layer is not strong, such that the soil becomes
nonlinear over only a portion of the entire duration of motion and in only a certain
frequency band. As a result, the averaging characteristic in Fourier spectral analysis
makes it insensitive for identifying time-dependent frequency content. While a
windowed (or short-time) Fourier-based approach can be used to improve the above
analysis to a certain extent, it also reduces frequency resolution as the length of the
window shortens. Thus, one is faced with a tradeoff. The shorter the window, the
better the temporal localization of the Fourier amplitude spectrum, but the poorer
the frequency resolution, which directly influences the measurement of downshift
of soil resonance that typically arises in a low to intermediate frequency band.
More important, a Fourier-based approach explains data in terms of a linear
superposition of harmonic functions. Therefore, it is an appropriate, effective method
for characterizing linear phenomena such as waves with time-independent frequency,
rather than nonlinear phenomena with time-dependent frequency. An example of
time-independent and time-dependent frequency waves is a hypothetical wave record
y(t) = y1(t) + y2(t), where decaying waves y1(t) = cos[2πt + εsin(2πt)]e–0.2t have
time-dependent frequency of 1 + εcos(2πt) Hz, with ε denoting a constant factor,
and noise y2(t) = 0.05sin(30t) has time-independent frequency of 15 Hz. Note that
the waves shown in Figure 8.1 with ε = 0.5 are physically related to one type of
1.5
0.5
Amplitude
−0.5
−1
−1.5
0 1 2 3 4 5 6 7 8 9 10
Time (sec)
FIGURE 8.1 A hypothetical wave recording, consisting of nonlinear waves and noise with
frequencies 1 + 0.5cos(2πt) Hz and 15 Hz, respectively.
FIGURE 8.2 Fourier and marginal Hilbert amplitude spectra of the recording in Figure 8.1.
water waves that result from a nonlinear dynamic process and are also representative
of seismic responses at a nonlinear soil site (to be elaborated).
The time-dependent frequency waves can be expanded into and thus interpreted
by a series of time-independent frequency waves, as done by the Fourier spectral
analysis in which y(t), or y1(t) in particular, can be interpreted as to contain Fourier
components at all frequencies (see Equation 8.1, Figure 8.2, and Figure 8.3).
Alternatively, the expansion of y1(t), i.e.,
suggests that the Fourier transform of y1(t) consists primarily of two harmonic
functions centered respectively at 1 Hz and 2 Hz for ε << 1, and the widths of these
harmonic functions are proportional to the exponential parameter 0.2, which is
related to the damping factor. Note that Figure 8.1 and Figure 8.2 use ε = 0.5, which
is not a small number in comparison to unity, and thus they have the third observable
harmonic function at 3 Hz in Figure 8.2. Therefore, one can equally well describe
y1(t) by saying that it consists of just two frequency components, each component
having a time-varying amplitude that is proportional to e–0.2t. Indeed, if one were to
examine the local behavior of y1(t) in the neighborhood of a given time instant, say
FIGURE 8.3 Fourier components (fj, j = 1, 2, 3, 4, 5) of the recording in Figure 8.1 at selected
frequencies (i.e., 10 Hz, 5 Hz, 2 Hz, 1 Hz, and 0.5 Hz).
t0, this is precisely what one would observe. The Fourier-based analysis or interpre-
tation given here can also be seen in Priestley [19] and Zhang et al. [20], among
others.
Because the true frequency content of the waves y1(t) is bounded between 1 –
ε and 1 + ε, much less than 2 Hz, analysis of the above example suggests that Fourier
spectral analysis typically needs higher-frequency harmonics (at least 2 Hz for the
example) to simulate the nonlinear waveform of the data. Stated differently, Fourier
spectral analysis distorts the nonlinear data. Consequently, the Fourier-based
approach in Equation 8.3 twists the influences of soil nonlinearity in site amplifica-
tion. The above assertions are confirmed in Huang et al. [21] and Worden and
Tomlinson [22], among others, with the aid of solutions to classic nonlinear systems
such as the Duffing equation in general, and in Zhang et al. [23] with the nonlinear
site amplification in particular.
In theory, Fourier spectral analysis in general and Fourier-based approaches for
site amplification in particular can be further used for evaluating damping factor.
For example, the resonant amplification method or half-power method uses the
amplitude change or width of the peaks at a certain frequency in the Fourier ampli-
tude spectrum to find the damping factor of dynamic systems such as a soil layer
(e.g., [24]). However, the distorted Fourier amplitude spectrum for nonlinear data
will mislead the subsequent use for damping evaluation with nonlinear soil. For
example, the damping factor evaluated at the first and second peaks in the Fourier
amplitude spectrum in Figure 8.2 suggests that the damping is associated with
frequency at 1 Hz and 2 Hz. In fact, the damping of the hypothetical record is
dependent only on the true frequency content of the waves y1(t) bounded between
1 – ε and 1 + ε, or 0.5 Hz and 1.5 Hz with ε = 0.5 in Figure 8.1. Accordingly, a
Fourier-based approach would misrepresent the influences of damping factor as it
relates to soil nonlinearity.
n n
X (t ) = ℜ ∑ a (t)e
j =1
j
iθ j ( t )
=ℜ ∑[C (t) + iY (t)] ,
j =1
j j (8.5)
where Cj(t) and Yj(t) are respectively the jth IMF component of X(t) and its Hilbert
transform,
C j (t ′)
∫
1
Yj (t ) = P dt ′ ,
π t − t′
where P denotes the Cauchy principal value, and the time-dependent amplitudes
aj(t) and phases θj(t) are the polar-coordinate expression of Cartesian-coordinate
expression of Cj(t) and Yj(t), from which the instantaneous frequency is defined as
d θ j (t )
ω j (t ) = . (8.6)
dt
H (ω , t ) = ∑ a (t) ,
j =1
j (8.7)
and its square gives the temporal evolution of the energy distribution. The marginal
Hilbert amplitude spectrum, h(ω), defined as
h(ω ) =
∫ H (ω, t)dt ,
0
(8.8)
provides a measure of the total amplitude or energy contribution from each frequency
value, in which T denotes the time duration of the data.
In comparison with Equation 8.2, the Hilbert amplitude spectrum H(ω,t) provides
an extra dimension by including time t in motion frequency and is thus more general
than the Fourier amplitude spectrum F(Ω). While the marginal amplitude spectrum
h(ω) provides information similar to the Fourier amplitude spectrum, its frequency
term is different. The Fourier-based frequency (Ω) is constant over the sinusoidal
harmonics persisting through the data window, as seen in Equation 8.1, while the
HHT-based frequency ω varies with time based on Equation 8.6. As the Fourier
transformation window length reduces to zero, the Fourier-based frequency (Ω)
approaches the HHT-based frequency (ω). The Fourier-based frequency is locally
averaged and not truly instantaneous, for it depends on window length, which is
controlled by the uncertainty principle and the sampling rate of data.
Recordings that are stationary and linear can typically be decomposed or rep-
resented by a series of time-independent frequency waves through the Fourier-based
approach in Equation 8.1. If the jth IMF component, i.e., Cj(t) in Equation 8.5,
corresponds to a Fourier component with a sine function at a time-independent
frequency, the Hilbert transform of the sine function, i.e., Yj(t) in Equation 8.5, can
be found to equal the cosine function at the same frequency in opposite sign. Because
the sign can be changed with adding a constant phase, the above analysis essentially
leads to the consistence between Fourier- and HHT-based approaches in general,
hs2,h1 + hs2,h 2
FH s ω = ( ) . (8.9)
hr2,h1 + hr2,h 2
n n
X (t ) = ℜ ∑
j =1
a j (t )e
iθ j ( t )
=ℜ ∑ Λ (t)e
j =1
j
− ϕ j ( t )+ iθ j ( t )
(8.10)
− ϕ j (t )
a j (t ) = Λ j (t )e . (8.11)
d ϕ j (t )
η j (t ) = . (8.12)
dt
With the aid of Equation 8.11, the Hilbert damping spectrum can be found as
n n
a j (t ) (t )
Λ
D(ω , t ) = ∑j =1
η j (t ) = ∑ − a (t) + Λ (t) .
j =1 j
j
j
(8.13)
T n T
(t )
a j (t ) Λ
d (ω ) =
∫ D(ω , t )dt = ∑∫
j =1 0
− + j dt = d a (ω ) + d Λ (ω ) .
a j (t ) Λ j (t )
(8.14)
0
Equation 8.14 indicates that the marginal Hilbert damping spectrum consists of
two terms: one is from the time-dependent amplitudes aj(t) that are related to
marginal and Hilbert amplitude spectra, and the other is from source-related intensity,
i.e., time-dependent amplitudes Λ j(t).
It is of interest to note that the definition of instantaneous damping factor in
Equation 8.12 and subsequent spectra in Equation 8.13 and Equation 8.14 are
different from those in Salvino [26] and Loh et al. [27]. For recordings of
impulse-induced or ambient linear vibration responses, some IMF components can
be extracted from the data that are related to certain vibration modes [28–30].
Consequently, Λj(t) are constant and ηj(t) are proportional to the damping ratio and
damped frequency. The modal damping ratio can then be found. This is essentially
the same as those in Salvino [26] and Loh et al. [27], if the latter can judicially
relate the IMF components to the vibration/wave modes.
For recordings to an earthquake, Λj are functions of time and unknown, which
are dependent upon the seismic source. Their influences in the site amplification,
however, can be removed if two recordings at soil and referenced sites are used.
Similar to the HHT-based factor of site amplification, the difference of marginal
Hilbert damping spectra at soil and referenced sites, or HHT-based factor of site
damping, could approximately eliminate the influences of the source that is associated
0.5
0
c1
−0.5
0.5
c2
0
−0.5
0.02
c3
0
−0.02
0.02
c4
0
−0.02 −3
×10
10
5
c5
0
0 1 2 3 4 5 6 7 8 9 10
Time (sec)
FIGURE 8.4 The five IMF components of the recording in Figure 8.1.
with Λj and thus provide essentially the characterization of the damping in the soil
site. The HHT-based factor of site damping can be found as
where use has been made in the last approximation of the fact that the source-related
damping terms at the soil and referenced sites are approximately equal, i.e., dsΛ (ω)
drΛ (ω).
Finally, comparing the HHT-based factors of site damping from the mainshock
and the aftershock could help quantify the influences of nonlinear soil damping in
site responses.
To illustrate the HHT-based characterization of nonlinearity, the hypothetical
record in Figure 8.1 is analyzed again. Figure 8.4 shows the five IMF components
decomposed from the data by EMD. The first and second components (c1 and c2)
capture the noise and primary waveform, while the other three (c3 to c5), with
negligible amplitudes, represent the numerical error in the EMD process. Comparing
Figure 8.3 and Figure 8.4 suggests that some IMF components can be not only more
physically meaningful than the Fourier components, they can also be in principle
used to explore the damping factor with the use of, e.g., the consecutive peak values
18 0.9
16 0.8
14 0.7
12 0.6
Frequency (Hz)
10 0.5
8 0.4
6 0.3
4 0.2
2 0.1
0
0 1 2 3 4 5 6 7 8 9 10
Time (sec)
and corresponding time elapse in the second IMF component, if that component is
related to a free-vibration response or forced response at a certain mode.
The Hilbert amplitude spectrum in Figure 8.5 shows a clear picture of tempo-
ral–frequency energy distribution of the data, i.e., primary waves with frequency
dependence modulated around 1 Hz and bounded by 0.5 Hz and 1.5 Hz, noise at
15 Hz, and the decaying energy of the primary waves with the color changing from
the red/yellow at the beginning to the dark blue at the end of the record. In contrast,
the Fourier amplitude spectrum in Figure 8.2 not only loses the information pertain-
ing to temporal characteristics of the motion, but, more important, it also distorts
the information of the record by introducing higher-order harmonics, notably at 2
Hz and 3 Hz. For comparison, the marginal amplitude spectrum of the recording is
also plotted in Figure 8.2, showing truthfully the energy distribution of the motion
in frequency.
Figure 8.6 shows the marginal Hilbert damping spectrum, calculated without
using any smooth function, which was not so done in the calculation of the above
marginal Hilbert amplitude spectrum with the use of Hilbert-Huang Transformation
Toolbox [31]. While oscillation of the curve in Figure 8.6 is originally due to the
numerical calculation of a· j(t) and a· j(t)/aj(t) that subsequently influences the compu-
tation of the spectra in Equation 8.13 and Equation 8.14, the estimated mean damping
factor at frequency 0.5 to 1 Hz is around the true value of 0.2. Compared with no
100
10−1
10−2
100 1
10
Frequency (Hz)
FIGURE 8.6 Marginal Hilbert damping spectrum of the recording in Figure 8.1.
damping in the record at frequency larger than 1.5 Hz, the very small mean damping
factor in Figure 8.6 (about 0.04, lower at a factor of five than the damping factor of
waves) due to the aforementioned unavoidable, cumulated numerical error is still
acceptable. It is believed that the oscillated curve in Figure 8.6 can be improved by
replacing it with the instantaneous mean curve. The mean curve can be found by
many methods, one of which is the use of the summation of all the IMF components,
excluding the first IMF component, that are extracted from the data of the oscillated
curve with one sifting process. The large damping factors at around 1.5 Hz are due
to the numerical error caused by the transition of two damping factors from 0.2 to
0.04, although such abrupt damping change is unlikely in practice. The large damp-
ing factor below 0.5 Hz is caused by the high-order, low-frequency IMF components
(primarily from the third to fifth IMFs). The error in overestimating damping factor
at very low frequency can be theoretically minimized if high-order, low-frequency
IMF components with very small amplitudes are judicially not used in the damping
calculation.
0.2
Acceleration (g)
0.1
−0.1
−0.2
0 5 10 15 20 25 30 35 40
0.05
Acceleration (g)
−0.05
−0.1
0 5 10 15 20 25 30 35 40
Time (sec)
FIGURE 8.7 (a, top) NS-acceleration recording and (b, bottom) its first IMF of the Nisqually
mainshock at SDS (soft soil).
148 m/s. The LAP is located over a stiff soil with Vs30 = 367 m/s. To examine the
soil nonlinearity from recordings at the two stations, recordings at SEW are used
as referenced ones, because Vs30 = 433 m/s at SEW is within the range of Vs30
values for typical rock sites in the western United States. Previous Fourier-based
studies (e.g., [17]) suggest that SDS experienced strong soil nonlinearity during the
mainshock while the LAP did not, with recordings at SEW as a reference.
0.2
Acceleration (g)
−0.2
−0.4
0 5 10 15 20 25 30 35 40
0.04
Acceleration (g)
0.02
−0.02
−0.04
−0.06
0 5 10 15 20 25 30 35 40
Time (sec)
FIGURE 8.8 (a, top) EW-acceleration recording and (b, bottom) its first IMF of the Nisqually
mainshock at SDS (soft soil).
spikes. It should be noted that there exist other tools to identify the spikes. For
example, Hou et al. [32] used a wavelet-based approach to characterize the spikes
from nonlinear vibration recordings in the vicinity of a damaged-structure location
subject to a severe earthquake. The disadvantages of these approaches, however,
reside with the subjective selection of frequency band in a Fourier-based approach
and subjective selection of another wavelet in a wavelet-based approach, among
others.
This study presents the effectiveness of the HHT-based approach in identifying
the spikes. Figure 8.7b depicts the first IMF component of the NS-component of
motion, clearly showing the two largest spikes between 20 sec and 25 sec. While
the spikes in the recording of Figure 8.7a can be visualized without using any tools,
the first IMF component in Figure 8.7b is shown simply for validation of the
HHT-based approach in identifying the spikes. Indeed, the EW-component of the
same recording in Figure 8.8a does not clearly show the spikes between 20 sec and
25 sec. The corresponding first IMF component in Figure 8.8b is, however, able to
reveal them explicitly, suggesting that the first IMF component is effective at detect-
ing the high-frequency spikes. This study also analyzed the horizontal recordings
of the ML3.4 aftershock at the same location in Figure 8.9 and Figure 8.10, in which
one does not observe the spikes in the S-coda waves in the corresponding first IMF
0.5
Acceleration (g)
−0.5
−1
0 5 10 15 20 25 30 35 40
1
Acceleration (g)
−1
−2
0 5 10 15 20 25 30 35 40
Time (sec)
FIGURE 8.9 (a, top) NS-acceleration recording and (b, bottom) its first IMF of the Nisqually
aftershock at SDS (soft soil).
0.5
Acceleration (g)
−0.5
−1
0 5 10 15 20 25 30 35 40
6
Acceleration (g)
−2
−4
0 5 10 15 20 25 30 35 40
Time (sec)
FIGURE 8.10 (a, top) EW-acceleration recording and (b, bottom) its first IMF of the
Nisqually aftershock at SDS (soft soil).
0.18
101 0.16
0.14
0.12
Frequency (Hz)
0.1
100 0.08
0.06
0.04
0.02
10−1
0 5 10 15 20 25 30 35 40
Time (s)
FIGURE 8.11 Hilbert amplitude spectra of NS-acceleration recording of the 2001 Nisqually
earthquake mainshock at SDS (soft soil).
• In the low-frequency range (below 2.5 Hz), Figure 8.16a shows a down-
shift profile in both frequency and amplitude from the aftershock to
mainshock that is similar to Figure 8.15a in the low to intermediate
frequency range, but the former shows a smaller shift (about 0.1 Hz in
0.1
101
0.09
0.08
0.07
Frequency (Hz)
0.06
100 0.05
0.04
0.03
0.02
0.01
−1
10
0 5 10 15 20 25 30 35 40
Time (s)
FIGURE 8.12 Hilbert amplitude spectra of EW-acceleration recording of the 2001 Nisqually
earthquake mainshock at SDS (soft soil).
frequency and a factor of 0.2) than the latter (about 0.7 Hz and a factor
of 0.43).
• In the intermediate to high frequency range, there is almost no difference
in the two factors between the mainshock and the aftershock.
Comparison of the HHT-based factors at SDS and LAP suggests that SDS had
severe soil nonlinearity during the mainshock and LAP had slight soil nonlinearity.
Site LAP can also be regarded as having no soil nonlinearity under the mainshock
if the aforementioned small downshift in both frequency and amplitude in the
low-frequency range is the result of variation of data collection and sampling.
−4
Nisqually Aftershock – 2-D Hilbert Spectrum - SDSns ×10
6
101
5
4
Frequency (Hz)
3
100
10−1
0 5 10 15 20 25 30 35 40
Time (s)
FIGURE 8.13 Hilbert amplitude spectra of NS-acceleration recording of the 2001 Nisqually
earthquake aftershock at SDS (soft soil).
−4
Nisqually Aftershock – 2-D Hilbert Spectrum - SDSew ×10
5.5
101 5
4.5
3.5
Frequency (Hz)
3
0
10 2.5
1.5
0.5
10−1
0 5 10 15 20 25 30 35 40
Time (s)
FIGURE 8.14 Hilbert amplitude spectra of EW-acceleration recording of the 2001 Nisqually
earthquake aftershock at SDS (soft soil).
the mainshock. The downshift in both amplitude and frequency from the HHT-based
factors between mainshock and aftershock is clearly seen in Figure 8.17b.
To further illustrate the characteristics of the HHT approach, this study compares
the HHT-based factors of site amplification at SDS shown in Figure 8.15a with the
Fourier-based ones shown in Figure 8.15b (i.e., Figure 8.7 in Frankel et al. [17]).
In the low-frequency range, Figure 8.15b shows a frequency-downshift profile
from the aftershock to mainshock that is similar to Figure 8.15a, but the former
shows a smaller shift (about 0.2 Hz) than the latter (about 0.7 Hz). Because of the
averaging characteristic in Fourier spectral analysis, as indicated in Section 8.3, the
frequency downshift measured from the HHT-based factors in Figure 8.15a may
give a more truthful indication of the soil nonlinearity than that measured from
Fourier-based factors in Figure 8.15b. In addition, the factor in the low-frequency
range in Figure 8.15a is generally somewhat larger than the factors in Figure 8.15b.
In the intermediate-frequency range, Figure 8.15b shows an amplitude-reduction
profile from the aftershock to mainshock that is similar to Figure 8.15a, but the
1
10
Spectral Ratio
0
10
Aftershock
Mainshock
−1
10
10−1 100 10
1
Frequency (Hz)
1
10
Spectral Ratio
100
Aftershock
Mainshock
10−1
10−1 100 10
1
Frequency (Hz)
FIGURE 8.15 (A) HHT-based factor for site amplification at SDS (soft soil) for mainshock
and aftershock of the 2001 Nisqually earthquake. (B) Fourier-based factor for site amplifica-
tion at SDS (soft soil) for mainshock and aftershock of the 2001 Nisqually earthquake.
1
10
Spectral Ratio
0
10
Aftershock
Mainshock
10−1
10−1 100 101
Frequency (Hz)
101
Spectral Ratio
100
Aftershock
Mainshock
10−1
10−1 100 101
Frequency (Hz)
FIGURE 8.16 (A) HHT-based factor for site amplification at LAP (stiff soil) for mainshock
and aftershock of the 2001 Nisqually earthquake. (B) Fourier-ased factor for site amplification
at LAP (stiff soil) for mainshock and aftershock of the 2001 Nisqually earthquake.
2
10
Spectral Ratio
1
10
0
10
Aftershock
Mainshock
−1 0
10 10 101
Frequency (Hz)
2
10
Spectral Ratio
1
10
0
10
Aftershock
Mainshock
−1 0
10 10 101
Frequency (Hz)
FIGURE 8.17 (A) HHT-based factor for site amplification with the use of recordings in
window 0 to 10 sec at SDS (soft soil) for mainshock and aftershock of the 2001 Nisqually
earthquake. (B) HHT-based factor for site amplification with the use of recordings in window
10 to 20 sec at SDS (soft soil) for mainshock and aftershock of the 2001 Nisqually earthquake.
Aftershock
Mainshock
10−1
Difference of Damping Factor
10−2
0 2 4 6 8 10 12 14 16 18 20
Frequency (Hz)
FIGURE 8.18 Difference of marginal Hilbert damping spectra at SDS and SEW for main-
shock and aftershock of the 2001 Nisqually earthquake.
former shows a smaller reduction (about a factor of 0.2) than the latter (about a
factor of 0.43).
In the high-frequency range, Figure 8.15b shows a significant increase in the
Fourier-based factor from the aftershock to mainshock, while Figure 8.15a does not.
Following the discussion in Section 8.3 and numerical illustration in Figure 8.1 and
Figure 8.2, the high-frequency content in the Fourier amplitude spectra may be
influenced by higher-order harmonics used to represent a nonlinear waveform, which
consequently increases the Fourier-based factor of the mainshock in the high-fre-
quency range.
This study also compares HHT- and Fourier-based factors at LAP in Figure 8.16a
and b. Almost no fundamental difference in the factors is observed in terms of overall
profile, amplitude value, frequency downshift, and amplitude reduction between the
mainshock and aftershock, indicating that the two approaches are essentially consistent
with each other in estimating linear or approximately linear site amplification.
reveals that the site damping during the mainshock is much larger than that in the
aftershock at frequency 0.5 to 5 Hz, suggesting that strong soil nonlinearity occurred
during the mainshock in this frequency band. The increased damping will decrease
the amplified magnitude of seismic wave responses through the nonlinear soil and
thus reduce the site amplification factor. This can be confirmed from Figure 8.15a,
which shows that the HHT-based factor for site amplification is observably reduced
for the mainshock from the aftershock in the similar frequency band of 0.5 to 7 Hz.
Figure 8.18 also shows that the second largest increased site damping for the
mainshock is in the frequency band 8 to 16 Hz. However, the HHT-based factors
for site amplification in Figure 8.15a do not appear to change between the mainshock
and aftershock. This can be explained as follows: On the one hand, the increased
site damping for the mainshock over the frequencies 8 to 16 Hz will reduce the
seismic wave responses in the same frequency band. Note that the soil nonlinearity
typically occurs under large-amplitude seismic waves incident to the soil layer, and
that the amplitude of the motion changes with time. This suggests that the soil
nonlinearity is likely most prevalent in the strong S-wave motion or following surface
or S-coda waves, but not in the P-wave motion during the mainshock. Figure 8.7a
and Figure 8.8a reveal that except for the abnormal high-frequency spikes between
20 sec and 25 sec, the mainshock accelerations after the S-wave arrival at about 10
sec contain less high-frequency motion than does the aftershock in general, and in
the high-frequency band of about 8 to 16 Hz in particular. This fact is consistent
with the increased damping of the nonlinear soil during the mainshock in the
high-frequency band of 8 to 16 Hz, which damped out the high-frequency wave
responses in the recordings.
On the other hand, the soil nonlinearity indeed introduces large-amplitude
high-frequency spikes from 20 to 25 sec in the mainshock recordings in Figure 8.7a
and Figure 8.8a.
Together with the fact that the HHT-based factors in Figure 8.15a show the
averaged characteristics of soil linearity and nonlinearity, these observations suggest
that the overall high-frequency (around 8 to 16 Hz) content of the mainshock may
be equivalent to that of the aftershock. This yields the same factors in Figure 8.15a
in the frequency range 8 to 16 Hz.
For further clarification, the HHT-based factor for site damping at LAP is
examined. Figure 8.19 shows that the profiles of soil damping are essentially no
different between the mainshock and aftershock, suggesting that site LAP is linear
during the mainshock. This can be further verified from the HHT-based site-ampli-
fication factors shown in Figure 8.16a.
Aftershock
Mainshock
10−1
Difference of Damping Factor
10−2
0 2 4 6 8 10 12 14 16 18 20
Frequency (Hz)
FIGURE 8.19 Difference of marginal Hilbert damping spectra at LAP and SEW for main-
shock and aftershock of the 2001 Nisqually earthquake.
The HHT-based factor for site damping can be extracted from the HHT-based
factor for site amplification and used as an alternative index to measure the influences
of soil nonlinearity in seismic wave responses at sites.
It should be pointed out that the results from this study rely mainly on the use
of advanced signal processing techniques to explore and then quantify the signature
of soil nonlinearity from recordings. They must, therefore, be validated by
model-based simulation. Recently, significant advances have been made in borehole
data collection and simulation techniques. These include data from 17 borehole
arrays in Southern California [33] and from the Port Island vertical array for the
1995 Hyogoken Nanbu earthquake, geophysical data including the S-wave velocity
profile in the top layer(s) at key strong motion station sites (Rosrine 2002 at
http://geoinfo.usc.edu/rosrine, USGS open file reports), and simulated broadband
ground motion for scenario earthquakes including nonlinear soil effects [34]. The
above information should allow the further validation of the observations and results
from this study.
ACKNOWLEDGMENTS
The author would like to express sincere gratitude to Norden E. Huang at NASA;
Stephen Hartzell, Authur Frankel, and Erdal Safak at USGS; Lance VanDemark
from Colorado School of Mines; and Yuxian Hu from China Seismological Bureau
and Jianwen Liang from Tianjin University of China for providing data, Fourier
analysis and calculations, and more important, constructive suggestions. This work
was supported by the National Science Foundation with Grant Nos. 0085272 and
0414363, and by the US-PRC Researcher Exchange Program administered by Mul-
tidisciplinary Center for Earthquake Engineering Research. The opinions, findings,
and conclusions expressed herein are those of the author and do not necessarily
reflect the views of the sponsors.
REFERENCES
1. Frankel, A., C. Mueller, T. Barnhard, D. Perkins, E. Leyendecker, N. Dickman, S.
Hanson, and M. Hopper (2000), “USGS national seismic hazard maps,” Earthquake
Spectra, v. 16, pp. 1–19.
2. Kramer, S.L. (1996) Geotechnical Earthquake Engineering, Prentice-Hall, Inc. Upper
Saddle River, NJ.
3. Borcherdt, R.D. (2002) “Empirical evidence for site coefficients in building code
provisions,” Earthquake Spectra, Vol. 18, pp. 189–217.
4. Hartzell, S. (1998) “Variability in nonlinear sediment response during the 1994
Northridge, California, earthquake,” Bull. Seism. Soc. Am. Vol. 88, 1426–1437.
5. O’Connell, D.R.H. (1999) “Influence of random-correlated crustal velocity fluctua-
tions on the scaling and dispersion of near-source peak ground motions, Science, Vol.
283, pp. 2045–2050, March 26.
6. Yoshida, N. and S. Iai (1998) “Nonlinear site response and its evaluation and predic-
tion,” The Effects of Surface Geology on Seismic Motion, (Irikura, Kudo, Okada and
Sasatani, eds.), Balkema, Rotterdam, pp. 71–90.
7. Huang, N.E., S. Zheng, S.R. Long, M.C. Wu, H,H. Shih, Q. Zheng, N-C. Yen, C.C.
Tung, and M.H. Liu (1998). “The empirical mode decomposition and Hilbert spec-
trum for nonlinear and nonstationary time series analysis." Proc. R. Soc Lond., A 454
903–995.
8. Hardin, B.O. and V.P. Drnevich (1972a) “Shear modulus and damping in soils:
measurement and parameter effects,” J. Soil. Mech. Foundations Div. ASCE, 98,
603–624.
9. Hardin, B.O. and V.P. Drnevich (1972b) “Shear modulus and damping in soils: design
equation and curves measurement and parameter effects,” J. Soil. Mech. Foundations
Div. ASCE, 98, 667–692.
10. Erdik, M. (1987) Site response analysis. In Strong Ground Motion Seismology (Erdik,
M. and Toksoz, M., eds.), D. Reidel Publishing Company, Dordrecht, 479–543.
11. Vucetic, M. and R. Dobry (1991) “Effect of soil plasticity on cyclic response, J.
Geotech. Eng., 117, 89–107.
12. Silva, W., S. Li, B. Darragh, and N. Gregor (1999). “Surface geology based motion
amplification factors for the San Francisco Bay and Los Angeles areas,” A Pearl
Report.
13. Field, E.H., P.A. Johnson, I.A. Beresnev, and Y. Zeng (1997) “Nonlinear
ground-motion amplification by sediments during the 1994 Northridge earthquake,
Nature, 390, 599–602.
14. Beresnev, I.A., E.H. Field, P.A. Johnson, and K.E.A. Van Den Abeele (1998) “Mag-
nitude of nonlinear sediment response in Los Angeles basin during the 1994
Northridge, California, earthquake, Bull. Seis. Soc. Am., Vol. 88, pp. 1097–1084.
15. Joyner, W.B. (1999). “Equivalent-linear ground-response calculations with fre-
quency-dependent damping,” Proceedings of the 31st Joint Meeting of the US-Japan
Panel on Wind and Seismic Effects (UJNR), Tsukuba, Japan, May 11–14, 1999, pp.
258–264.
16. Bonilla, L.F., D. Lavallee, and R.J. Archuleta (1998). “Nonlinear site response:
laboratory modeling as a constraint for modeling accelerograms,” The Effects of
Surface Geology on Seismic Motion, (Irikura, Kudo, Okada and Sasatani, eds.),
Balkema, Rotterdam.
17. Frankel, A.D., D.L. Carver, and R.A., Williams (2002) “Nonlinear and linear site
response and basin effects in Seattle for the M6.8 Nisqually, Washington, Earthquake,”
Bull. Seism. Soc. Am., v. 92, pp. 2090–2109.
18. Safak, E. (1997). “Models and methods to characterize site amplification from a pair
of records,” Earthquake Spectra, Vol. 13, pp. 97–129.
19. Priestley, M.B. (1981). Spectral Analysis and Time Series, Vol. 2, Multivariate Series,
Prediction and Control, Academic Press, London, 1981.
20. Zhang, R., S. Ma, E. Safak, and S. Hartzell (2003b) “Hilbert-Huang transform
analysis of dynamic and earthquake motion recordings,” ASCE Journal of Engineer-
ing Mechanics, Vol. 129, No. 8, pp 861–875.
21. Huang, N.E., Z. Shen, and R.S. Long, (1999). “A new view of nonlinear water waves:
Hilbert Spectrum.” Annu. Rev. Fluid Mech. 31,417–457.
22. Worden, K. and Tomlinson, G.R. (2001) Nonlinearity in Structural Dynamics: Detec-
tion, Identification and Modeling, Institute of Physics Publishing, Bristol and Phila-
delphia.
23. Zhang, R., L. VanDemark, J. Liang, and Y.X. Hu (2003) “Detecting and quantifying
nonlinear soil amplification from earthquake recordings,” Advances in Stochastic
Structural Dynamics (Zhu, Cai and Zhang, eds.), pp. 543–550.
24. Clough, R.W. and J. Penzien (1993). Dynamics of Structures, Second Edition,
McGraw-Hill, Inc., New York.
25. Zhang, R., S. Ma, and S. Hartzell (2003) “Signatures of the seismic source in
EMD-based characterization of the 1994 Northridge, California, earthquake record-
ings,” Bulletin of the Seismological Society of America, Vol. 93, No. 1, pp 501–518.
26. Salvino, L.W. (2001) “Evaluation of structural response and damping using empirical
mode analysis and HHT,” in CD ROM, 5th World Multiconference on Systemics,
Cybernetics and Informatics, July 22–25, 2001, Orlando, Florida.
27. Loh, C.H., T.C. Wu and N.E. Huang (2001). Application of the empirical mode
decomposition-Hilbert spectrum method to identify near-fault ground-motion char-
acteristics and structural responses, Bull. Seism. Soc. Am. 91, 1339–1357.
28. Yang, J.N., Y. Lei, S. Pan and N. Huang (2002a). “System identification of linear
structures based on Hilbert-Huang spectral analysis. Part I: Normal modes,” Earth-
quake Eng. Structural Dynamics, Vol. 32, pp. 1443–1467.
29. Yang, J.N., Y. Lei, S. Pan and N. Huang (2002b). “System identification of linear
structures based on Hilbert-Huang spectral analysis. Part II: Complex modes,” Earth-
quake Eng. Structural Dynamics, Vol. 32, pp. 1533–1554.
30. Xu, Y.L., S.W. Chen, and R. Zhang (2003) “Modal identification of Di Wang building
under typhoon York using HHT method,” The Structural Design of Tall and Special
Buildings, Vol. 12, No. 1, pp. 21–47.
31. Hilbert-Huang Transformation Toolbox (2000). Professional Edition V1.0, Princeton
Satellite Systems, Inc., Princeton, New Jersey.
32. Hou, Z., M. Noori and R. St. Amand (2000) “Wavelet-based approach for structural
damage detection,” ASCE Journal of Engineering Mechanics, 126(7), pp. 677–683.
33. Archuleta, R.J. and J.H. Steidl (1998). “ESG studies in the United States: results
from borehole arrays,” The Effects of Surface Geology on Seismic Motion, (Irikura,
Kudo, Okada and Sasatani, eds.)., Balkema, Rotterdam, pp. 3–14.
34. Hartzell, S., A. Leeds, A. Frankel, R.A. Williams, J. Odum, W. Stephenson and W.
Silva (2002). “Simulation of broadband ground motion including nonlinear soil
effects for a magnitude 6.5 earthquake on the Seattle fault, Seattle, Washington,” Bull.
Seism. Soc. Am. Vol. 92, No. 2, pp. 831–853.