Journal of Earthquake Engineering

ISSN: 1363-2469 (Print) 1559-808X (Online) Journal homepage: http://www.tandfonline.com/loi/ueqe20

GROUND MOTION PARAMETRS FOR THE

2011GREAT JAPAN TOHOKU EARTHQUAKE

Bhargavi Podili & STG Raghukanth

To cite this article: Bhargavi Podili & STG Raghukanth (2017): GROUND MOTION PARAMETRS
FOR THE 2011GREAT JAPAN TOHOKU EARTHQUAKE, Journal of Earthquake Engineering,
DOI: 10.1080/13632469.2017.1342292

To link to this article: http://dx.doi.org/10.1080/13632469.2017.1342292

View supplementary material

Accepted author version posted online: 09

Aug 2017.

Submit your article to this journal

Article views: 18

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=ueqe20

Download by: [UC Santa Barbara Library] Date: 02 September 2017, At: 23:14
 GROUND MOTION PARAMETRS FOR THE
2011GREAT JAPAN TOHOKU EARTHQUAKE
Bhargavi Podili1 and STG Raghukanth2

1 Structural Engineering, Indian Institute of Technology, Chennai, India

t
ip
2 Department of Civil Engineering, Indian Institute of Technology Madras, Chennai, India

Correspondence to STG Raghukanth, Department of Civil Engineering, Indian Institute of

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

Technology Madras, IIT Madras, chennai, 600036 India. Email: [email protected]

Abstract:

us
The 2011 great Japan Tohoku earthquake (Mw 9.0) is not only the most devastating but also, one
an
of the best recorded earthquakes in the history of seismology. Strong motion characteristics of

this earthquake are studied using well established ground motion parameters. A total of 20
M

parameters are extracted from the data. The behaviour of these parameters with fault distance

and average shear wave velocity is examined and their values for stations MYG004 and
ed

TCGH16 are tabulated. Further, an attenuation relationship is developed using the 1172 surface

level strong motion records. Also, Principal Component Analysis is used in classification of the
pt

derived GMPs, which is further used to rate the damage potential of ground motion records.
ce

Keywords: Ground Motion Parameters, Fault distance, shear wave velocity, Evolutionary PSD,

Response spectra, and Principal component analysis

Ac

1
 1. INTRODUCTION

Great earthquakes that occur at subduction zones of destructive plate boundaries are called

Megathrust earthquakes. These are considered to be earth’s most powerful earthquakes in terms

t
of magnitude and destruction. According to the United States Geological Survey (USGS), there

ip
are five megathrust earthquakes with moment magnitude (Mw) greater than or equal to 9.0 in the

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

past 100 years. These include the 1952 Kamchatka earthquake (Mw-9.0), the 1960 Chile

us
earthquake (Mw-9.5), the 1964 Alaska earthquake (Mw-9.2), the 2004 Sumatra-Andaman

earthquake (Mw-9.3) and recently, the 2011 Tohoku earthquake (Mw-9.0). These earthquakes in
an
turn are responsible for huge tsunamis, which increased the damage further. India had evidenced

the effect of 2004 Sumatra earthquake at the Indo-Burmese subduction plate boundary and there
M
is possibility of such great earthquakes at the Indo-Eurasian plate boundary as well. Thus, it is

necessary to study these megathrust events to understand the nature of ground motions resulted
ed

in the process. Among all, the 2011 Tohoku earthquake is the most well recorded earthquake

with high density of strong motion records over a large spatial coverage. The study of ground
pt

motion records available for this event provides better understanding of megathrust earthquakes
ce

and in turn, would help in the structural design of important structures against the same.

Due to availability of large datasets, numerous investigations on various features of the Mw 9.0
Ac

Tohoku earthquake are conducted over the years. So far, these ground motion records were used

to develop various finite fault models [Shao et al. (2011), Suzuki et al. (2011), Faqi et al. (2012)

etc.], examine the source rupture process and site effects. Many theories are proposed regarding

the source and site characteristics, of which the widely accepted are multiple strong motion

2
 generation areas, nonlinear and local site effects. Kurahashi and Irikura (2011) simulated the

acceleration and velocity motions using a source model consisting of five Strong Motion

Generation Areas (SMGAs) and they have attributed the reason behind the extreme peak

accelerations to nonlinear soil behavior and non-uniform stress parameters in the strong motion

t
generation areas. Asano and Iwata (2012) have developed a source model composed of four

ip
SMGAs using empirical Green’s function and assumed that each wave packet in the acceleration

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

waveform is generated from the corresponding SMGA. Joshi et al. (2014) made use of the

us
KIKNet borehole records and simulated ground motions through source modeling via a modified

semi-empirical technique. Ghofrani et al. (2013) have performed stochastic finite fault
an
simulations for the Tohoku earthquake using the EXSIM algorithm given by Boore et al (2009)

and found out that the multi-event model with five SMGAs given by Kurahashi and Irikura
M
(2011) was successful in producing ground motions with spectral amplitudes both at low and

high frequencies. Kawase (2011) has determined strong motion characteristics such as Peak
ed

ground Acceleration (PGA), Peak ground Velocity (PGV), Horizontal-to-Vertical (H/V) spectral

ratios and PGA/PGV to examine site effects such as amplification due to soil nonlinearity and
pt

structural damage potential of the strong motions. Wu and Peng (2011) have further confirmed
ce

the non linear site response during the main shock using spectral ratio approaches. Ghofrani et al.

(2012) have made use of surface to borehole spectral ratios (S/B) as a direct measure of site
Ac

response and found out that site effects were strong at high frequencies for the Tohoku event.

They have developed an empirical model to express site amplification as a function of average

Shear wave Velocity (Vs30) and H/V spectral ratio. Midorikawa et al (2012) compared the strong

motion records of the Tohoku earthquake with other similar events in Chile and Peru (Mw8.4 to

3
 8.8) and pointed out that the attenuation and spectral characteristics were same for all the events

over a period range of 0-10 seconds. Goda et al. (2012) have conducted building damage surveys

post the event and concluded that the unique characteristics of the ground motions were due to

the effects of local asperities and nonlinear soil behavior. Stewart et al (2013) reported that the

t
scaling of site amplification with VS30 is weak to nonexistent for high-frequencies, but strong

ip
for low frequency ground motions.

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

The Tohoku earthquake is the first Mw 9.0 megathrust earthquake for which a huge dataset of

us
ground records are available and the ground motions recorded during this event have some

unique and remarkable features. The studies that have been carried out on ground motion
an
characteristics of the Tohoku earthquake are only on peak and spectral characteristics. It would

be interesting to find out the behavior of energy, intensity and duration parameters as well.
M

Therefore, in the present study, a systematic and detailed analysis of 20 Ground Motion

Parameters (GMPs) representing the amplitude, energy, intensity, frequency content, duration
ed

and evolutionary characteristics of the ground motion is performed. All these parameters are
pt

listed in Table 1, along with their corresponding applications in engineering. Ground propagation

effects such as geometric spreading and anelastic attenuation are studied through behavior of
ce

these GMPs with rupture distance. The effect of the volcanic arc on the ground motion

attenuation at back-arc is also addressed. The site effects are studied through variation of all the
Ac

GMPs with average shear wave velocity of the top 30m (Vs30). Also, H/V spectral ratios and

predominant periods are evaluated to observe the site amplification at each site. An evolutionary

power spectrum is calculated to observe the nonstationary energy distribution in time and

frequency. Further, Principal component analysis is used to classify the GMPs, based on the

4
 statistical correlation that exists between the parameters. The groups resulted from the

classification is then used to rate the damage potential of the ground motion records. Finally, a

ground motion prediction equation (GMPE) is derived for all the 20 GMPs and the functional

form is compared with other regionally established GMPEs.

t
ip
2. STRONG MOTIONS OF THE TOHOKU

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

EARTHQUAKE

us
Japan, due to its impeccable position near four of the active plate boundaries (Figure 1), is prone
an
to more destructive earthquakes than any other country. Yet, the catastrophic event that occurred

on March 11, 2011 is unique and unforgettable in the history of Japan. The epicenter of the
M

earthquake along with other seismo-tectonic features such as fault details and plate boundaries is

shown in Figure1. The earthquake is the result of reverse faulting near the subduction zone
ed

interface between Pacific and North American plates, making it one of the biggest megathrust

earthquakes experienced by Japan. According to the Japan Meteorological Agency (JMA), the
pt

rupture initiated 130km east of Sendai city (Tohoku region) with a focal depth of 24km beneath
ce

the Pacific Ocean, which eventually lead to a huge tsunami with wave height as high as 9.3m.

Figure 1 includes the fault model given by the Geospatial Information Authority (GSI) of Japan,
Ac

which consists of two rectangular faults, the northern segment (186x129km) and the southern

segment (194x88km) with slip amounts of 25m and 6m respectively.

5
 2.1 Ground Motion Data

The strong motion records available for Tohoku earthquake is the largest available database till

date for any megathrust earthquake. The records used here are the data collected through high

t
sensitivity seismographs of Kyoshin Network (KNet) and Kiban Kyoshin Network (KIKNet).

ip
KNet consists of more than 1000 observation stations with seismographs positioned at ground

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

surface with an average spacing of 25km, covering the whole nation. It has been operating since

May, 1996. KIKNet is composed of 688 stations with a pair of strong motion seismographs

us
located at the deep end of borehole and at ground surface. The average depth of the borehole
an
varies between 100-200m, whereas some stations lie as deep as 2km. The collected waveform

records are archived by data management centre of National Research Institute for Earth Science
M
and Disaster Prevention (NIED) and are available for public access through

http://www.kyoshin.bosai.go.jp.
ed

2.2 Metadata
pt

The fault model presented in Figure 1 is used to calculate rupture distances, which were further
ce

used to study the behavior of GMPs. This choice does not make any significant difference in the

final results as the location of the fault given by majority of other fault models is consistent with
Ac

the current case. Due to a difference in Q-material between the two regions separated by a chain

of active volcanoes called volcanic arc, the back-arc region of Japan is considered to have higher

attenuation rate compared to that of the fore-arc region. Skarlatoudis and Papazachos (2012)

have investigated that the back-arc and fore-arc observations exhibit smaller differences for

6
 shorter periods for shallow interface events of focal depth less than 60 km. Given the magnitude

of earthquake under consideration is high, the ground motions comprise of broader band of

frequencies even at larger distances. Therefore, the volcanic arc is constructed (Figure 1)

following the studies of Zhao et al (2006). All the KNet and KIKNet stations are divided in to

t
fore-arc and back-arc stations according to their location with respect to the volcanic arc. The

ip
ground vibrations of Tohoku earthquake are captured by a total of 701 KNet stations and 520

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

KIKNet stations. Figure 2 shows the distribution of all the KNet and KIKNet stations relative to

us
the volcanic arc.

In addition to the strong motion data, seismic velocity profiles for KNet stations of the top 10-
an
20m and for KIKNet stations up to the respective borehole depth are made available by NIED.

Figure S1 (Electronic Supplement) represents shear wave velocity profiles of MYG004 and
M

TCGH16. Average soil shear wave velocity of top 30m (Vs30) is calculated for all KNet and

KIKNet stations with the assumption that the lower most velocity extends to 30m for those
ed

stations whose velocity profiles are given only up to the depth of 20m or less. Based on Vs30
pt

estimates of each station, all the sites are categorized from hard rock to soft soil site classes as

per Table 2 (International Building Code (IBC), 2009). Figure S2 (Electronic Supplement) shows
ce

the distribution of KNet and KIKNet stations over different soil classes. It is noted that more

than 80 percent stations of both the networks are classified under soft soil site classes of C and
Ac

D.

7
 2.3 Preliminary Analysis

The ground motion acceleration time histories have many interesting features such as twin phase

phenomena, extensive recorded durations and extremely peak values. The raw data is subjected

t
to baseline correction, where the mean of each ground motion record is subtracted from the

ip
whole time history and filtering, where the data is band pass filtered over a frequency range of

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

0.05 – 25 Hz. The lower cut-off frequency is selected after main trials involving visual

inspection of Fourier amplitude spectrum, velocity and displacement time histories of the filtered

us
acceleration time histories. The upper cutoff frequency is taken as half of Nyquist frequency,
an
which is usually considered half of the sampling frequency (Boore and Bommer, 2005). The

value of Peak Ground Acceleration (PGA) exceeds 1g for more than 10 records and a maximum
M
value of 2.73g is recorded for station MYG004. . It is identified that stations adjacent to the first

fault F1 have predominant first phase, whereas stations adjacent to the second fault F2 have
ed

second phase as the predominant phase. Figure 3 shows the accelerograms recorded at MYG004

and TCGH16, which are the stations that have recorded peak values among KNet and KIKNet
pt

respectively. The twin phase phenomenon can be observed in acceleration time histories of

MYG004 (Figure 3). But the first phase is not predominant in the ground motions recorded at
ce

TCGH16 and it is also evident that the acceleration values recorded at surface is higher
Ac

compared to those recorded at the deep end of the borehole. It is noted from Figure S1 that both

the stations MYG004 and TCGH16 have similar range of average velocities and are assigned to

site class D with Vs30 estimates of 224.92m/s and 213.19m/s respectively. This article studies

8
 and explores the characteristics of the earthquake ground motions with emphasis on the records

of these two stations.

3. ESTIMATION OF GROUND MOTION

t
ip
PARAMETERS (GMPs)

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

An acceleration time history provides complete information about the earthquake ground motion

us
at a given location. But the ground motion is highly stochastic and non-stationary, due its

dependence on complex source and medium characteristics. Therefore, it is difficult to model an

an
accelerogram for future earthquakes. Moreover, most of the linear structural analysis methods in

engineering consider peak values and response spectra as input parameters. Therefore,
M

parameters quantifying unique characteristics of the acceleration time histories are adapted as

representatives of the ground motion. Unlike the accelerogram, these are stable and thus, can be
ed

modeled mathematically. Over the years, many parameters that can represent one or more

characteristics of the earthquake ground motion have been proposed. After examining several of
pt

those well established ground motion parameters (Kramer, 1996), a total of 20 GMPs are
ce

estimated for the available strong motion data. These parameters vary from a simple peak value

of the acceleration time history to correlation between time and frequency in the Evolutionary
Ac

Power Spectral Density function. The variation of these parameters with fault distance over

different soil classes is examined and their values for stations MYG004 and TCGH16 are

tabulated for further comparison. The following subsections deals with the method of estimation

of each of the 20 parameters.

9
 3.1 GMPs derived from acceleration time history:

In engineering seismology, the most renowned and traditional characteristic of a ground motion

time history is its peak value. Therefore, first and fore-most, Peak Ground Acceleration (PGA),

t
Peak Ground Velocity (PGV) and Peak Ground Displacement (PGD) values are estimated.

ip
Velocity and Displacement time histories of MYG004 and TCGH16 are shown in Figure 3,

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

along with the corresponding acceleration time histories. Mean values of EW and NS

components of the accelerogram are considered as the Horizontal component for all ground

us
records. The station MYG004 has recorded a maximum PHA of 2.02g and a PHV of 78.5 cm/s.
an
All the near field ground motion records have recorded exceptionally high acceleration, velocity

and displacement values. The effect of near field source characteristics can be observed in the
M
variation of PGA values from Figure 4. Figure 5 shows the variation of these GMPs with fault

rupture distance and for different soil classes. It is noted that all the three GMPs follow similar
ed

pattern of higher values at lower distances with differential slopes.

The peak values of the vertical component of ground motion time histories exceeded that of
pt

horizontal components for some strong motion records. Figure 6 shows the variation of ratio of
ce

Vertical-to-Horizontal components (V/H) of peak acceleration, velocity and displacement to that

of horizontal components for different soil classes and rupture distances. The V/H ratios for the
Ac

far field records of hard rock strata exceeds one and the vertical components of peak velocity and

displacements are twice to that of horizontal components for some of these records.

To include the effect of duration on peak values, ratio of peak values of velocity to acceleration

(PGV/PGA) is calculated. The ratio approaches unity for far field records and there is a steep

10
 decrease in these values when the distance decreases. The near field acceleration time histories

of KNet indicate that the ground motions were recorded in multiple phases. Moreover, most of

the records have a long duration of 300 seconds. Therefore, a GMP depicting duration of ground

motion based on the energy thresholds of the acceleration time history is calculated. Significant

t
duration (Tsig) is taken as the time interval between thresholds 5% (t1) and 95% (t2) of the

ip
squared acceleration time history (Trifunac and Brady, 1975). In this study, Tsig is used instead of

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

total duration in calculating all the other GMPs that depends on duration of the ground motion.

us
Figure 7 show the variation of these GMPs with rupture distance and for different soil classes.

Even though the values of PGV/PGA and significant duration shares similar trend with rupture
an
distance, maximum values of PGV/PGA are recorded for hard rock strata instead of soft soil

classes.
M

In order to check the dominance of peak acceleration, a time independent effective acceleration

called Root-Mean-Square acceleration (aRMS) is evaluated. An effective acceleration of 0.235g is

ed

obtained for horizontal component of MYG004, which has recorded a PHA of 2.02g.
pt
ce

(1)
1
= [ ( )]
Ac

All the above GMPs capture peak values, which are instantaneous effects of the ground motion.

11
 Arias Intensity (Ia), which is the energy dissipated during ground shaking, is estimated to capture

the cumulative effects and thus the potential destructiveness of the earthquake. A maximum

value of 8505 cm/s was recorded at MYG004 over a significant duration of 82.7 seconds.

t
ip
(2)

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

< > = </ > [ ( )]

2

us
Here, g is the acceleration due to gravity and a(t) is the acceleration time history.
an
Another parameter that includes the cumulative effects of ground motion duration is cumulative
M
absolute velocity (CAV). CAV is developed to indicate the onset of structural damage to

engineered structures by the electric power research institute (EPRI, 1988), USA.
ed
pt

(3)
= | ( )|
ce
Ac

Figure 8 show the variation of these GMPs with rupture distance and for different soil classes. It

is observed that aRMS, Ia and CAV shows similar pattern as that of PGA, PGV and PGD.

12
 3.2 GMPs derived from Response Spectra (Intensity Parameters):

In seismic analysis and design, a single parameter that has been used in wide application as an

earthquake damage index is intensity. Spectrum intensity represents the intensity of shaking of a

t
structure due to an earthquake at a given site. Response spectra plots are obtained for all the

ip
recorded ground motions by calculating maximum response of a single degree of freedom

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

(SDOF) system to each component of the corresponding ground motion over a period range of 0

to 10s. Newmark’s linear acceleration method is used to derive spectral displacement from the

us
basic equation of motion of the SDOF system and corresponding values of pseudo-spectral
an
velocity, pseudo-spectral acceleration are obtained. Sample plots of acceleration and velocity

response spectra for ground motions (Figure 3) of MYG004 and TCGH16 are shown in Figures
M
9a and Figure 9b respectively.

The stations MYG004, MYG006 and MYG007 are all located at same distances from the fault
ed

with different site soil classes (Table 3). But the PGA of MYG004 with site class D is almost

four times to that of the other two stations with site classes C and E respectively. Therefore, H/V
pt

response spectra are obtained for three KNet stations MYG004, MYG006 and MYG007, in order
ce

to analyze the factors behind high amplitude values. Figure 10 shows the H/V response spectra

plots along with the corresponding site's fundamental frequency (F0) for EW and NS components
Ac

of the ground motion. It is noticed that, maximum H/V spectral ratios of MYG004 are recorded

for frequencies matching with the site's fundamental frequency. The frequency corresponding to

maximum H/V ratio of MYG004-NS is same as the site's F0. Therefore, the reason behind high

amplitude values (PGA of 2.73g) of the NS component of MYG004 is attributed to the resonance

13
 of ground motions with the first mode of natural frequency of the site. It should be noted here

that the fundamental frequency of the site varies with different modes. Therefore, consideration

of only the first mode in identifying resonance phenomena did not show any patterns among

other stations with unusually higher amplitudes (PGA > 1g). Apart from amplifications due to

t
resonance, site amplification is usually observed for soft soil classes.

ip
cr
In order to examine site amplification, Horizontal-to-Vertical ratio of response spectra for all the
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

ground motions is evaluated. The average H/V response spectra of stations MYG004 and

us
TCGH16 is shown in Figure 11. The period corresponding to maximum H/V value is taken as

the Predominant period (Tp) of the site. Figure 12 shows the variation of maximum H/V value
an
and Tp with shear wave velocity. It can be observed that there is a linear trend in between

maximum H/V values and Vs30, where higher values are recorded for lower Vs30 values i.e.,
M

sites with soft soil classes. Also, the linear trend between Tp and Vs30 is dominant for lower

periods, whereas there is no significant variation at higher periods. This suggests lower
ed

frequencies are predominant during the site amplification due to nonlinear behavior of the soft
pt

soil sites.
ce

To characterize the effect of these strong ground motions on both short and long period

structures, Acceleration Spectrum Intensity (ASI) and Housner’s Velocity Response Spectrum
Ac

Intensity (VSI) are computed using pseudo spectral acceleration and pseudo spectral velocities

respectively for a 5% critical damping ratio. Figure 13 shows the variation of these parameters

with rupture distance and for different soil classes. Another important GMP that has been used as

a damage index in hazard engineering is Characteristic Intensity (Ic). Ic is linearly related to

14
 structural damage index (Park et al., 1985). Figure 14 shows the variation of Ic with rupture

distance and for different soil classes.

t
ip
. (4)
( )= ( = 0.05, )

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

. (5)
( )= ( = 0.05, )

us
=( ) . . (Tsig )0.5 an (6)

Overall, the values of parameters quantifying amplitude, energy and intensity are very high. In
M

fact, the PHA recorded for this earthquake is the maximum value recorded for any other

earthquake till date. Table 4 show the values of all these parameters calculated for MYG004 and
ed

TCGH16. With such high amplitude and intensity values, the ground motions are expected to
pt

create a major havoc. But, the after studies show that the areas with these maximum recorded

values did not get affected by the ground motions as expected. Also, majority of the destruction
ce

in the 2011 event is not caused by the earthquake ground motions, but by the Tsunami that

followed. Therefore, we are bound to investigate the other characteristic that defines the ground
Ac

motion time histories, i.e. the frequency content.

15
 3.3 GMPs derived from Power Spectral Density (Frequency

Parameters):

To comprehend the frequency content of the recorded ground motion, Power Spectral Density

t
ip
(PSD) of individual time series, G(ω) is obtained using Welch’s estimation method and spectral

properties are interpreted through respective GMPs (Vanmarcke, 1976): Predominant frequency

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

(Fp), Central frequency (Ω) and Shape factor (q). Equation 7 gives the nth spectral moment of

us
PSD, G(ω); whereas ωN is the maximum frequency of the ground motion time history. , and

gives the zeroth, first and second spectral moments respectively. PSD plots for acceleration
an
time histories of MYG004 and TCGH16 are shown in Figure 15. The dominance of high

frequencies can be observed in PSD plots of MYG004. The PSD plots of TCGH16 show a
M

similar variation of density values with frequencies for both the surface and borehole records,

except for a significant amplification in the values of surface records over a frequency range of
ed

0.8-10Hz.
pt

Here, Predominant frequency (Fp) is taken as the frequency of vibration corresponding to the

maximum value of the PSD; and Central frequency (Ω), and Shape factor (q) represents spectral
ce

centroid and bandwidth of the ground motion spectrum, respectively.

Ac

(7)
= ( )

16
 (8)
=

(9)
= 1−

t
ip
The variation of these GMPs with rupture distance and for different soil classes is shown in

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

Figure 16. The dependence of frequency parameters is evaluated against rupture distance in order

us
to understand the propagation effects on the ground motions It is evident from the plots of

predominant frequency and central frequency of PSD that lower frequencies dominate far field
an
records, whereas higher frequencies dominate near field ground motion records. However, the

distance - frequency parameter fit has highest variance compared to any of the distance -
M
amplitude parameter relations. It is also observed that the attenuation is higher in the frequency

parameters for back-arc stations, to that of the attenuation of amplitude and duration parameters.
ed

The predominant frequency of MYG004 is 5.27Hz, which implies that a ground acceleration of

2.7g is in effect for a total of 0.19s. Shape factor, similar to PGV/PGA, is increasing with rupture
pt

distance and have recorded maximum values for hard ground strata. Furthermore, the values of
ce

shape factor approaches unity for larger rupture distances, revealing the broad band nature of far

field ground motions.

Ac

17
 3.4 GMPs derived from Evolutionary Power Spectral Density (EPSD):

The ground motion under consideration is both nonlinear and non-stationary in nature.

t
ip
Therefore, study of the evolutionary spectrum with instantaneous values of amplitudes and

cr
frequencies will provide a better characterization of ground motion. There exist many Time-
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

Frequency (TF) analysis methods such as Short Time Fourier Transform, Wavelet Transform,

us
Wigner Ville distribution etc., for the analysis of non-stationary signals. In this study, Hilbert

Huang Transform (HHT) with an Ensemble Empirical Mode Decomposition (EEMD) technique
an
is used for obtaining Evolutionary Power Spectral Densities (EPSD) for all the ground motion
M
time histories. The main reason behind adopting EEMD is that it extracts Instantaneous

frequencies (IF) and the corresponding amplitudes in a more physically meaningful Time-
ed

Frequency-Amplitude space.

Empirical Mode Decomposition (EMD) is an adaptive TF analysis method without any

pt

predetermined basis functions, which decomposes the complex signal into different simple
ce

empirical modes called Intrinsic Mode Functions (IMFs). Huang et al. (1998) have defined any

oscillatory mode satisfying the below two properties as an IMF: The number of zero crossings
Ac

and the number of extrema must either be equal or almost differ by one; At any point, mean

value of the envelopes defined by local maxima and local minima is zero. All the IMFs are

orthogonal to each other and each IMF represents unique range of amplitude and frequency. The

18
 only drawback using this method is mode mixing due to signal intermittency, where the

frequency modulated IMFs show irregular scales of amplitude modulation.

Wu and Huang (2009) have proposed ensemble EMD (EEMD) to overcome this effect. EEMD

defines the IMF components as the mean of an ensemble, which is derived from modified signals

t
ip
with an added White Noise (WN) of finite amplitude. The fundamental principle behind this

cr
approach is that the added WN would help the signal populate the whole Time-Frequency-
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

Amplitude space uniformly with different scales of the constituting components, avoiding mode

us
mixing. Therefore, the first step of EEMD is adding WN to the ground motion time history and

obtaining the new modified signal. The ensemble number (N) and the noise to data standard
an
deviation ratio, which corresponds to amplitude (An) of the WN to be added, are predefined such

that the modified signal represents more than 90 percent variance of the original signal. Then,
M

IMFs are extracted from each modified signal using EMD. The basic idea behind this approach is

to remove the local median from the signal through a sifting process. The procedure for
ed

extracting IMFs from the data involves three steps. The first step is to connect all the local
pt

maxima/minima of the signal with a cubic function as upper/lower envelopes. Then, the first

component ‘h’ is obtained by taking difference between the data and local mean of the upper and
ce

lower envelopes. The final step involves treating ‘h’ as the new data and repeating the first two

steps until the envelopes are symmetric with respect to zero mean and the final result is
Ac

considered to be the first IMF. The above sifting process is continued until the residue rn

becomes a monotonic function, from which no IMF can be extracted. Different WN sequences

are used in obtaining N modified signals and IMFs are extracted in each case following the

above procedure. An ensemble mean is performed over respective IMFs of all the modified

19
 signals and it is verified that the sum of resultant IMFs gives back the original signal. The

standard deviation of error (e) in the decomposition caused by the added WN is given by

equation (3.11). It is noted that the final error is relatively small for lower amplitudes of the WN

over a large ensemble number.

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

( )= + (10)

us
(11)
= an
The standard deviation ratio and ensemble number are selected efficiently such that the sum of
M

resultant IMFs gives the original data with a variation not exceeding 5%. Figure S3 (Electronic

Supplement) represents the IMFs obtained for ground motion time histories of MYG004 and
ed

TCGH16. The IMF1 of MYG004 represents 49% of the original time history and only the first
pt

three IMFs are required to reconstruct 93% of the ground motion. Also, due to the reason that the

last four IMFs represent less than 0.1% of the ground motion time history, these are neglected in
ce

further calculations. It should be noted that IMF1 is more rich in amplitude and frequency and

both the amplitude and frequency of the IMFs decreases from IMF1 to IMF12. The EW1
Ac

component of TCGH16 i.e., the borehole ground motion record shows a different nature in the

distribution of variance among different IMFs. In this case, IMF3 is more dominant with 26% of

variance and a total of 5 IMFs are required to reconstruct at least 90% of the time history.

20
 Locally, each IMF contains lower amplitude as well as frequency components than the one

extracted just before.

The next step is to obatin the evolutionary power spectrum through instantaneous values of

frequency and amplitude. In order to calculate instantaneous frequencies (ωj) and the

t
ip
corresponding amplitudes (Cj), an analytical function is constructed by forming a conjugate pair

cr
of IMF and hIMF, where hIMF is the Hilbert Transform of IMF.
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
( )= ( ) + ℎ ( )
an
(12)
M
( )= [ ( )] + [ℎ ( )] (13)
ed

ℎ ( )
( )= [ ] (14)
( )
pt
ce

1
( , )= [ ( )] (15)
2
Ac

A final surface plot of Evolutionary Power Spectral Density (EPSD), which is the energy

distribution in a time-frequency space is obtained by combining the values of energy in time for

21
 all the extracted IMFs. The distribution of energy in time over different frequencies in the EPSD

plots of MYG004 and TCGH16 can be observed in Figures 17 (a), (b) and (c). Therefore, the bi-

variant EPSD accounts for both the temporal and spectral non-stationarities of the ground

motion. Three distinct energy phases can be observed in the EPSD plot of MYG004. If each

t
energy phase is to be produced by a strong motion generation area, then this supports the theory

ip
[Kurahashi and Irikura (2011), Asano and Iwata (2012)] that the source is composed of more

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

than one strong motion generation area. The energy in the second phase has higher amplitudes

us
and a large frequency bandwidth. It is also observed that the borehole acceleration time history

of the station TCGH16 is rich in frequency content than that of the surface acceleration time
an
history.

Since it is difficult to evaluate variation of the complex EPSD with rupture distance and shear
M

wave velocity, a total of six GMPs characterizing EPSD are selected, such that these parameters

can represent the evolutionary power spectrum. The six GMPs include Total Energy (Eacc) of the
ed

EPSD, first and second moments of the frequency and time axes: Spectral centroid (Ew), Spectral
pt

standard deviation (Sw) and Temporal centroid (Et), Temporal standard deviation (St) and finally,

the Correlation coefficient of time and frequency (ρ(t,ω)).

ce
Ac

(16)
= ( , )

22
 ( , ) (17)
( )=
( , )

[ − ( )] ( , ) (18)
( )=
( , )

t
ip
( , ) (19)
( )=
( , )

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

[ − ( )] ( , ) (20)
( )=

us
( , )

[ − ( )][ − ( )] ( , )
( , )= (21)
( ) ( )
an
( , )
M

Here, (t2-t1) is Significant duration as defined by Tsig; Nyq is the Nyquist frequency and G (t,ω) is
ed

EPSD of the corresponding time history. The Total Energy (Eacc) of the ground motion time

history is equal to the area under EPSD, which in turn is similar to Arias Intensity (Ia). Also, both
pt

the Spectral centroid (Ew) and the Central frequency (Ω) derived from PSD represents same
ce

characteristic of the ground motion i.e., mean frequency.

Ac

Figure 18 shows the variation of these EPSD parameters with distance from the fault rupture and

for different soil classes. The Eacc values obtained using EPSD are upper bound to that of Ia

values computed from acceleration time histories. The values of Ew are also upper bound to that

of Ω values obtained using Welch’s PSD. From the plots of Ew and Et, it is observed that the

values of these parameters deviates from the linear trend for far field records at lower

23
 frequencies and higher periods respectively. The behavior of ρ(t,ω) is to be noted here. Initially,

there is a positive correlation between time and frequency for near field records, which reduces

and transforms to negative for larger distances when total duration is used in calculating these

parameters. But, this particular pattern in the values of correlation coefficient with rupture

t
distance is scattered due to consideration of Tsig in the calculation of ρ(t,ω). The spectral and

ip
temporal parameters along with Eacc and correlation coefficient can be used to simulate the EPSD

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

as a lognormal density function (Pousse et al. [2006] and Yamamoto [2011]), which can further

us
be used to develop ground motions (Liang et al.[2007]). All these GMPs are evaluated for the

entire 1172 surface and 502 borehole records and the corresponding values are presented for
an
stations MYG004 of KNet and TCGH16 of KIKNet in Table 4.
M
4. PRINCIPAL COMPONENT ANALYSIS (PCA)
ed

The behavior of different ground motion parameters with rupture distance and shear wave

velocity has been analyzed in the previous section. But, it is well known that some of these
pt

parameters are related to one another and may represent a similar feature of the ground motion. It

will be interesting here to find the correlation coefficient (CC) between these parameters. In fact,
ce

Arias Intensity (Ia) and Total Energy (Eacc) of EPSD has the highest CC of 0.99. The Central

frequency (Ω) of PSD and the Spectral centroid (Ew) of EPSD are highly correlated with a CC of
Ac

0.98. The CC of PGA with aRMS, Ia, CAV, Ic, ASI, VSI is of the order 0.90. Assuming the

parameters as Gaussian random variables, the CC value should be higher than 0.062 (2/√ , N

being the sample size) to be significant. Since all the GMPs have significant CCs with respect to

one another, Principal Component Analysis (PCA) is performed on the horizontal component of

24
 GMP values. The 20 GMPs considered in this study represents the ground motion statistically in

a 20 dimensional space. Now, PCA is used to rotate this 20D GMP space to a completely

uncorrelated space of principal components. Foremost, the 20D GMP space is to be standardized

for an unbiased analysis. Covariance matrix of the standardized values and its Eigen functions

t
are calculated using traditional techniques to obtain coefficients of the principal components.

ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
< >( ) −
< > Zi = ,
(22)
an
i = 1,2, … 20
M
ed

[ ]= , , = 1,2, … 20 (23)
pt
ce

Here, mi and σi are mean and standard deviation values of the corresponding GMPi, Zi is the final

standardized 20D space and Qij is the covariance matrix of Zi. Each Eigen vector (Uj) of Qij
Ac

corresponds to coefficients of each principal component and the Eigen value (λ) gives the

respective variance. Thus, a total of 20 orthogonal principal components are obtained in the order

of decreasing variance. The percentage variance of each principal component represents the

contribution of it in representing the original 20D space. From the values of percentage

25
 variances, it is observed that the first and second principal components together represent 75.6

percent of the original 20D space. Thus, the ground motion can be economically represented by

the two dimensional space of first two principal components.

Figure 19 shows the 2D plane with coefficients of these two principal components for the 20

t
ip
parameters plotted against each another. The result is significant, as all the 20 parameters have

cr
been sorted into three different quadrants of the 2D plane. This is similar to the three way
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

classification based on the characteristics of the ground motion i.e., amplitude, frequency content

us
and duration. The GMPs of amplitude space i.e., the first quadrant of the 2D plane exhibit very

high correlations in comparison to GMPs of other quadrants. The overlap between Arias
an
Intensity (Ia) and Total Energy (Eacc) of EPSD is as expected. The second quadrant contains all

the GMPs representing duration of ground motion. Additionally, Shape factor (q) and Spectral
M

standard deviation (Sw) are also classified in to this duration space. Even though these two

parameters are obtained from PSD and EPSD respectively, their behavior with rupture distance
ed

and shear wave velocity hints a pattern similar to those of the duration space. All the other
pt

parameters derived in the frequency domain are grouped into the fourth quadrant i.e., the

frequency space. Extension of this analysis over GMPs calculated for large datasets of ground
ce

motion records covering different earthquakes will provide a more efficient classification.
Ac

It is observed that the coefficients of first principal component are negative for the duration

parameters and the second principal component shows negative values for frequency parameters.

Therefore, it will be interesting to develop a rating scale for the ground motion records using

these principal components. It is evident from the post-earthquake analysis works of Goda et al.

26
 (2013) that all the near field records did not cause a similar damage even though the numerical

values of respective GMPs are similar. The values of first and second principal components for

all the strong motion records are plotted against each other (Figure 20) in order to examine the

damage rate of each ground motion. Since all these records are relative to each other and there is

t
no absolute datum to indicate zero damage, a ground motion record with zero parameter values is

ip
considered as a representation of Zero Damage Record. In addition, various cases of hypothetical

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

Single Parameter Records are considered for each of the 20 ground parameters, in which the

us
value of GMP is varied from zero to the respective maximum value at subsequent steps and

corresponding principal component coefficients are obtained. From figure 20, it can be observed
an
that the Single Parameter Records of duration and frequency parameters are varying similarly

with respect to the Zero Damage Record, with a linear trend in opposite directions. If a straight
M
axis is drawn interpolating these records and passing through the zero damage record, then the

single parameter records of amplitude parameters are varying orthogonal to this Frequency-
ed

Duration axis. The ground records closer to this axis are found out to have minimum values of

amplitude, energy and intensity. Also, it is found out that the farthest ground motion record from
pt

this axis is of the station MYG004, which has recorded maximum values for all the amplitude
ce

parameters. Figure S4 (Electronic Supplement) shows the variation of distance of ground records

from this Frequency-Duration axis (dF-D) with distance from fault rupture (Rrup). This behavior is
Ac

similar to that of the pattern followed by coefficients of ground motion records for the first

principal component. Therefore, it is safe to assume that this attribute, the distance from the

frequency-duration axis (dF-D) indicates the damage capacity of a strong motion record. The

variation of this distance for all ground records in the latitude-longitude space is shown in Figure

27
 21. It can be observed that those near field records which have closer range of peak acceleration

and energy values does not show the same pattern with respect to the distance from frequency-

duration axis (dF-D).

t
5. GROUND MOTION PREDICTION EQUATION

ip
(GMPE)

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
The next step is to develop an empirical equation for estimating the ground motion parameters.

Even though it is reasonable to develop prediction equations for principal components such that
an
all the parameters can be derived using minimal coefficients, the involvement of original mean

and standard deviation in inverse PCA makes this task unfeasible. Moreover, the principal
M

components thus obtained are physically unexplainable. Therefore, an empirical prediction

equation that estimates ground motion parameters directly is preferred. Meanwhile, many
ed

attenuation relations for traditional parameters like PGA, PGV and PGD are available for the

Japan region. But, it will be interesting to develop a new prediction equation for complex
pt

parameters as well such as RMS acceleration, Arias Intensity, Spectral centroid etc., in addition
ce

to the conventional parameters. Since this is an event specific analysis, random effects such as

the dependencies between moment magnitude and rupture distance does not affect the
Ac

attenuation equation, neither a mixed effects model nor a multi-stage regression analysis is

considered in the present study. Therefore, after examining different functional forms of the

available ground motion attenuation relations for subduction zones (Douglas et al, 2012), the

following simplified GMPE is selected for all the 20 Ground Motion Parameters:

28
 ( )= + + + + ( 30) + +Ԑ (24)

0, −

t
=
1

ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

The predictor variables Fault rupture distance (Rrup) and Shear wave velocity (Vs30), which is the

us
average velocity in the top 30 meters of the soil strata, accounts for path effects such as anelastic

attenuation, geometric spreading and local site effects respectively. The coefficient C3 in the
an
attenuation functional form addresses magnitude saturation at near fault distances. The location

of station relative to the volcanic arc (Larc) addresses the difference in attenuation characteristics
M
of fore-arc and back-arc stations. Larc is considered ‘0’ for all the fore-arc stations and ‘1’ for all

the back-arc stations, such that the coefficient C5 accommodates the higher rate of attenuation of
ed

back-arc stations. These types of functional forms are popular among various engineering

applications such as hazard analysis due to their simplicity. Non-linear least square regression
pt

analysis is used to calibrate the above prediction equation with the GMPs obtained from the
ce

recorded surface strong motion data. The residual error (ε) of the GMPE is normally distributed

with zero mean and standard deviation of σ(ε). Figure S5 (Electronic Supplement) shows the
Ac

variation of residuals from the regression equation with predictor variables: Rupture distance and

Shear wave velocity. It can be observed from the plots that the residuals follow no trend with

respect to predictor variables. Therefore, the functional form considered is able to capture the

trends in the GMPs sufficiently. The coefficients of predictor variables and the standard error

29
 obtained for the respective GMPs are reported in Table 5. The resulting estimates provide mean

values of the GMPs and σ(ε) represents variation of these estimates around the mean values.

The coefficient C0 in the regression equation, represents the strength of the strong motions and it

can be observed that the horizontal components of the amplitude parameters have higher values

t
ip
than the vertical components. The coefficients C1 corresponds to geometric spreading and the

cr
coefficients C2 corresponds to anelastic attenuation of the ground motions. The very small values
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

of C1 in comparison to that of C2, C4 and C5 values suggest that the effect of geometric spreading

us
is more than that of anelastic attenuation. The values of amplitude and frequency parameters of

ground motion decreases with distance and due to this reason, GMPs such as PGA, PGV, PGD,
an
aRMS, Fp and Ew etc., have negative coefficients for C2. But due to longer periods at large

distances, parameters such as PGV/PGA, q, Tsig, Et and St have positive values for C2. A similar
M

trend to that of C2 is observed in the values of C5. Parameters such as q, PGV/PGA, Et and ρ(t,ω)

have positive values for C5 and all the amplitude and frequency parameters have negative values.
ed

This pattern is due to the reason that C5 represents the anaelastic attenuation similar to that C2.
pt

The coefficient C3 accounts for short distance saturation, which depends on the magnitude of the

earthquake. It was considered the form c.10d.M (Kanno et al., Si and Midorikawa) and c.ed.M
ce

(Zhao et al., Abrahmason et al.), where M is the magnitude of the earthquake and c, d are

coefficients of the regression equation. Since, the present study is event specific, the coefficient
Ac

C3 is obtained through nonlinear least square regression analysis for all the GMPs. It is observed

that all the frequency parameters have highest values for C3 to that of amplitude and duration

parameters. This suggests that the effect of distance saturation is higher for frequency

parameters. The coefficients of Vs30, C4 are positive for Spectral parameter ‘q’. This makes

30
 sense as the frequency band width increases from soft soil class to hard rock strata. Similarly, the

positive values of C4 for ρ(t,ω) is due to the reason that only frequencies of narrow band exists

among soft soil classes. Therefore, ρ(t,ω) increases with Vs30. It can also be observed that the

magnitude of C4 is much higher for horizontal components of amplitude parameters. Therefore,

t
the coefficient C4 indicate a probable site amplification in the horizontal components, which

ip
supports the theory of using H/V spectrum as an indicator of site amplification given by

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

Nakamura (1989).

us
Even though Arias Intensity (Ia) and Total Energy of EPSD (Eacc) show maximum spread from

the mean values, it is verified that the prediction equation captures the trend in the variation of
an
these parameters with distance from the fault adequately. Besides, the systematic variation of

these predicted values with soft soil class to hard rock strata can be observed in Figures 8 and 18
M

respectively. The scatter in the predicted values of Spectral centroid (Ew) derived from EPSD is

much less compared to that of Central frequency (Ω) derived from PSD (Figures 16, 18). All the
ed

estimates of frequency and duration parameters, with the exception of Predominant frequency
pt

(Fp), deviates very less from the mean values compared to that of amplitude parameters. Overall,

all the estimates satisfies the R-squared test with the standard deviation ratio of calculated values
ce

to the respective residual errors much less than one, confirming the goodness of estimates.

Moreover, the validity of the functional form is verified by eliminating any pattern over the
Ac

predictor variable and residual error plots.

31
 5.1 Comparison with regional GMPEs

There are many well established GMPEs that are applicable to subduction zones: Abrahamson et

al. (2016), Atkinson and Boore (2003), Zhao et al. (2006) etc. Since the models proposed by Si

t
and Midorikawa (1999), Kanno et al. and Zhao et al. are regionally established GMPEs for

ip
Japan, only the above three models are considered for the purpose of comparison. Zhao et al.

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

(2006) has given the GMPE for four different site classes, which were classified using site

amplifications based on H/V response spectral ratio. The model is more generic, including the

us
coefficients applicable only for crustal events, subduction slab events and interface events. In
an
order to compare this GMPE with that derived in the present study, site class III is considered as

equivalent to soft soil class with a Vs30 of 300m/s. Kanno et al (2006) obtained Vs30 values for
M
KNet through deriving a correlation between Vs20 and Vs30 values of KIKNet stations. In

addition, a site correction term was given through regression analysis on the residuals (between
ed

observed and predicted values of PGA, PGV) averaged at intervals of every 100 m/s in Vs30.

Here, the GMPE is derived for Root-Mean-Square (RMS) of the two horizontal components.
pt

Therefore, a correction term is applied to this model, which is obtained by taking ratio of RMS

value and mean of the two horizontal (EW, NS) components. The model for PGA and PGV
ce

given by Si and Midorikawa (1999) is derived for larger of the two horizontal components.
Ac

Therefore, a similar correction term is applied, which is obtained by taking ratio of larger value

and mean of the two horizontal (EW, NS) components. Also, the GMPEs for PGA and PGV are

inherently derived for soil ground and stiff ground respectively. Therefore, the comparison of

GMPEs for PGA and PGV is adopted for Vs30 values of 300m/s and 600m/s.

32
 Figures 22 and 23 provide all the GMPEs considered for comparison along with the present

model. The attenuation forms for fore-arc and back-arc stations are given separately. It can be

observed that the GMPE provided by Zhao et al. (2006) is sufficiently predicting PGA, but

deviates at larger distances. Kanno et al. and Si and Midorikawa are over-estimating PGA values

t
at distances greater than 200 Km. Also, the GMPE given by Kanno et al. is underestimating near

ip
field PGA values. In case of PGV predictions, the model given by Si and Midorikawa is over-

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

estimating whereas the model given by Kanno et al. is effective with the near field estimations

us
and is over predicting PGV values for larger distances. Since the model derived in this article is

event specific, it manages to capture the trend in the variation of different GMPs with distance
an
from the fault and average shear wave velocity.
M
6. SUMMARY AND CONCLUSION
ed

Strong motions resulting from the 2011 Tohoku earthquake have many remarkable features.

Ground motion parameters of this earthquake reflect the scale of the event and complexity
pt

underlying the rupture process. The station MYG004, with a fault rupture distance of 119.4km

and site class D, has recorded the highest ever Peak Ground Acceleration (PGA) of 2.73g. The
ce

same station has recorded a Peak Ground Velocity (PGV) of 78.5cm/s and a Peak Ground

Displacement (PGD) of 61.5cm. It is apparent from Figure 5 that peak values of acceleration and
Ac

velocity are recorded for near field ground motions of soft soil strata. These extreme values are

attributed to the source effects of the earthquake. The vertical-to-horizontal (V/H) ratio of peak

values, which reaches as high as 2 for far field records (Figure 6), is an evidence of the

megathrust nature of the earthquake. Another interesting feature of these acceleration time

33
 histories is the twin phase phenomenon. Even though, all the recorded ground motions did not

have this property, most of the near field records with maximum Peak acceleration values

especially that of KNet have shown distinct phases (Figure 3). There is an average 50 second

time difference between the peak values of two phases for the time histories that are recorded for

t
an overall duration of 300 seconds. This particular feature is attributed to the multiple ruptures in

ip
the source mechanism as given by the Geospatial Information authority of Japan. The significant

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

duration (Tsig) of MYG004 is 82.7s, which resulted in a root-mean-square acceleration (arms) of

us
0.235g, which is far less compared to the peak horizontal acceleration (PHA) of 2.02g.

Therefore, even though the ground motions are recorded for longer durations and even though
an
extreme peak values are recorded, no significant damage is noticed. The Arias Intensity (Ia) and

Cumulative Absolute Velocity (CAV) values obtained for near field records are 5 times larger to
M
that of those recorded for the Mw 8.8 Chile earthquake (2010). Intensity related parameters such

as Acceleration Spectrum Intensity (ASI), Velocity Spectrum Intensity (VSI) and Characteristic
ed

Intensity (Ic) follow the same trend with exceptionally high values. Further, on analyzing H/V

spectral ratios, the reason behind high amplitude values (PGA of 2.73g) of the NS component of
pt

MYG004 is attributed to the resonance of ground motions with the first mode of natural
ce

frequency of the site. Therefore, in accordance with the enormity of these parameters, the area

around MYG004 and other near field stations with large peak and intensity parameters is bound
Ac

to be devastated. But, structural damage survey post the event show that these areas are relatively

less damaged compared to those stations that have recorded lower values.

Therefore, spectral characteristics of the ground motion are studied through analysis of Power

Spectral Density (PSD) functions. Central frequency (Ω), Predominant frequency (Fp) and shape

34
 factor (q) are derived from power spectra of corresponding ground motions. The ground motions

of most of the near fields are dominated by high frequencies, where the station MYG004 has

recorded predominant and central frequencies as 9.5Hz and 8Hz respectively. The broad band

nature of far field records compared to that of near field records can be observed in the variation

t
of the shape factor (q) with rupture distance. The study of predominant periods corresponding to

ip
maximum H/V response spectra indicate that lower frequencies are predominant during the site

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

amplification due to nonlinear behavior of the soft soil sites. Also, sudden variation in the

us
behavior of frequency parameters at larger rupture distances indicate that far field records are

dominated by long periods.

an
Since Evolutionary Power spectrum conveys more information on non-stationary ground

motions, Ensemble Empirical Mode Decomposition (EEMD) along with Hilbert transform is
M

used in deriving Evolutionary Power Spectral Density (EPSD) functions. This energy

distribution is studied through six parameters characterizing EPSD: Total energy of the ground
ed

motion time history (Eacc), Temporal centroid (Et), Spectral centroid (Ew), Temporal standard
pt

deviation (St), Spectral standard deviation (Sw) and correlation between time and frequency

(ρ(t,ω)). It is observed that the behavior and scale of Eacc is on par with that of Arias intensity
ce

and Ew delivers mostly upper bound values to that of central frequency (Ω). Also, Et deviates

from the linear trend at lower periods for larger distances and the behavior is similar to that of
Ac

significant duration with respect to fault rupture distance and shear wave velocity (Figure 18).

Since all the GMPs have significant correlation coefficient (CC) values with respect to one

another, Principal Component Analysis (PCA) is used to rotate the original 20 dimensional space

35
 of ground motion parameters to a new uncorrelated space of Principal Components (PCs).

Further, the dimensionality of the 20D GMP space is reduced by considering the first two PC’s,

which have a combined variance of 75.6 percent. Furthermore, all the GMPs are categorized into

three groups of amplitude, frequency content and duration spaces (Figure 19), by plotting the

t
coefficients of these first two principal components. It is noticed that parameters such as VSI and

ip
ASI, Ia and Eacc, CAV and PGA, Ew and Ω are highly correlated, even though they are defined in

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

different contexts. Further, a zero damage record along with various single parameter records in

us
the principal component space is used to construct a Frequency-Duration axis (Figure 20). The

distance from this Frequency-Duration axis (dF-D) is taken as an attribute to rate the damage
an
potential of ground motion records (Figure 21).

Finally, a new and simple empirical prediction equation is derived for all of these GMPs using
M

least square regression analysis with rupture distance (Rrup) and Shear wave velocity (Vs30) as

predictor variables. The goodness of estimates is verified through R-squared tests and
ed

elimination of any pattern over plots of predictor variables and the resultant error. Further, the
pt

estimates from the prediction equation are compared against regional ground motion prediction

equations of Japan such as Si and Midorikawa (1999), Zhao et al. (2006), and Kanno et al.
ce

(2006). Due to the effect of magnitude saturation at lower distances, all the GMPEs that were

derived prior to this study are under-predicting near field characteristics. Since all the GMPs are
Ac

derived for a single event, the predicted values would be governed by the source mechanism and

other properties underlying the ground motion generation of the 2011 Tohoku earthquake. Since

this is the first earthquake of its kind (Megathrust event with Mw ≥ 9.0), for which recorded

datasets are available, GMPE for GMPs is given without inclusion of the event term. It is true

36
 that without the event term, the model is not reliable to predict GMPs for other regions. But, the

aim of this article is to study the ground motions produced by the event in terms of Ground

Motion Parameters and to understand the propagation and site effects on each GMP via rupture

distance and shear wave velocities. A detailed study on the source characteristics and regional

t
trends of GMPs for other great earthquakes would cement the application of this GMPE to other

ip
megathrust events in other parts of the world. Therefore, the behavior of GMPs derived in this

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

study can be used as a reference in seismic analysis and design of structures for other great

us
earthquakes occurring in similar tectonic environments around the world.

REFERENCES
an
1. Arias A. [1970], “A measure of earthquake intensity”, Seismic Design for Nuclear Power
M
Plants. Hansen RJ (Ed.), MIT Press, Cambridge: 438–483.

2. Atkinson, G. M., and Boore, D. M., [2003], “Empirical ground-motion relationships for
ed

subduction-zone earthquakes and their application to Cascadia and other regions”, Bulletin of the

Seismological Society of America 93, 1703–1729.

pt

3. Asano, K. and Iwata, T. [2012], “Source model for strong ground motion generation in
ce

the frequency range 0.1-10 Hz during the 2011 Tohoku earthquake”, Earth Planets Space, 64,

1111–1123.
Ac

4. Abrahamson, N., Gregor, N., and Addo, K. [2016], “BC Hydro Ground Motion

Prediction Equations for Subduction Earthquakes”, Earthquake Spectra: February 2016, Vol. 32,

No. 1, pp. 23-44.

37
 5. Boore, D.M., Bommer, J.J. [2005]. “Processing of strong-motion accelerograms: needs,

options and consequences”. Soil Dynamics and Earthquake Engineering 25 (2005) 93–115.

6. Boore, D. M., Thompson, E. M., and Cadet, H. [2011], “Regional Correlations of VS30

and Velocities Averaged Over Depths Less Than and Greater Than 30 Meters”, Bulletin of the

t
ip
Seismological Society of America, Vol. 101, No. 6, pp. 3046–3059, doi: 10.1785/0120110071.

cr
7. Cornell, C. A. [1968], "Engineering seismic risk analysis", Bulletin of the Seismological
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

Society of America, 58(5), 1583-1606.

us
8. Douglas, J., F. Cotton, C. Di Alessandro, D.M. Boore, N.A. Abrahamson and S. Akkar
an
[2012], “Compilation and critical review of GMPEs for the GEM-PEER Global GMPEs

Project”, 15th World Conference on Earthquake Engineering, Lisboa.

M
9. Electrical Power Research Institute (EPRI), [1988], “A Criterion for Determining

Exceedance of the Operating Basis Earthquake”, EPRI NP-5930, Palo Alto, CA.
ed

10. Faqi, D., Xiong, X. and Yong, Z. [2012], “Static slip model of the Mw 9.0 Tohoku

(Japan) earthquake: Results from joint inversion of terrestrial GPS data and seafloor
pt

GPS/acoustic data”, Chinese Science bulletin, Vol.57, No.16: 1990-1997, doi: 10.1007/s11434-
ce

012-5014-5.

11. Geospatial Information Authority (GSI) of Japan, http://www.gsi.go.jp/cais/topic110422-

Ac

index-e.html

12. Ghofrani, H., K. Goda and G.M. Atkinson [2012], “Ground Motions and Site Effects

from the 2011 M9.0 Tohoku Japan Earthquake”, 15th World Conference on Earthquake

Engineering, Lisboa.

38
 13. Ghofrani, H., G. M. Atkinson, K. Goda, and K. Assatourians [2013], “Stochastic finite-

fault simulations of the 2011 Tohoku, Japan earthquake”, Bulletin of the Seismological Society of

America, Vol.103, No. 2B, pp. 1307–1320, doi: 10.1785/0120120228

14. Goda, K., Pomonis, A., Chen Chian, S., Offord, M., Saito, K., Sammonds, P., Fraser, S.,

t
ip
Raby, A. and Macabuag, J. [2013], “Ground motion characteristics and shaking damage of the

11th March 2011 Mw9.0 Great East Japan earthquake”, Bull Earthquake Eng, 11: 141-170, doi:

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

10.1007/s10518-012-9371-x.

us
15. Gutenberg, b. and Richter, C.F. [1956], "Earthquake magnitude: intensity, energy and

acceleration", Bulletin of the Seismological Society of America, 46, 104-145.

an
16. Housner GW. [1952], “Spectrum intensity of strong motion earthquakes”, Proceedings of
M
the Symposium on Earthquakes and Blast Effects on Structures, Earthquake Engineering

Research Institute, California: 20–36.

ed

17. Housner, G. W. and Jennings, P. C. [1964], "Generation of artificial earthquakes",

Journal of the Engineering Mechanics Division, ASCE, 90, 113-150.

pt

18. Huang, N. E., Shen, Z., Long, S. R., Wu, C. M., Shih, H. H., Zheng, Q., Yen, N. C.,
ce

Tung, C. C. and Liu, H. H. [1998], “The empirical mode decomposition and the Hilbert spectrum

for nonlinear and nonstationary time series analysis”. Proc. R. Soc. Lond. A, 454: 903–995.
Ac

19. Iyengar, R.N., Prodhan, K.C. [1983], “Classification and Rating of Strong Motion

Earthquake records”. Earthquake Engineering and Structural Dynamics, Vol.11, 415-426.

20. Japan Meteorological Agency, http://www.jma.go.jp/jma/en/2011 Earthquake.html

39
 21. Joshi, A., Sandeep and Kamal [2014], "Modeling of strong motion generation areas of the

2011 Tohoku, Japan earthquake using modified semi-empirical technique", Nat Hazards, 71:

587-609.

22. Kanno, T., Narita, A., Morikawa, N., Fujiwara, H. and Fukufhima, Y.. [2006], “A New

t
ip
Attenuation Relation for Strong Ground motion in Japan based on recorded data”, Bulletin of the

Seismological Society of America, Vol.96, No. 3, pp. 879–897, doi: 10.1785/0120050138

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

23. Kurahashi, S. and Irikura, K. [2011], “Source model for generating strong ground

us
motions during the 2011 off the Pacific coast of Tohoku Earthquake”, Earth Planets Space, 63,

571–576.
an
24. Kramer SL. [1996], ‘Geotechnical Earthquake Engineering’, Prentice-Hall, New Jersey.
M
25. Kawase, H. [2014], ‘Studies on the 2011 Off the Pacific coast of Tohoku earthquake’.

26. Liang, J., Chaudhuri, S. R. and Shinozuka,M. [2007], “Simulation of nonstationary

ed

stochastic processes by spectral representation” J. Eng. Mech., 133(6): 616–627.

27. Midorikawa, S., Miura, H. and Atsumi, T. [2012], “Characteristics of strong ground
pt

motion from the 2011 gigantic Tohoku, Japan earthquake”, Proceedings of the International
ce

Symposium on Engineering Lessons Learned from the 2011 Great East Japan Earthquake,

28. National Research Institute for Earth Science and Disaster Prevention (NIED) network,
Ac

Japan, https://www.kyoshin.bosai.go.jp

40
 29. Nakamura, Y. and Saito, A. [1983], “Estimations of Amplification Characteristics of

Surface Ground and PGA using Strong Motion Records (in Japanese)”, Proc. 17th JSCE

Earthquake Eng. Symp., 25-28.

30. National Institute of Advanced Industrial Science and Technology, [2012], Active Fault

t
ip
Database of Japan, Feb 28, 2012 version, https://gbank.gsj.jp/activefault.html.

cr
31. Okada, Y., Kasahara, K., Hori, S., Obara, K., Sekiguchi, S., Fujiwara, H., and
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

Yamamoto, A. [2004], “Recent progress of seismic observation networks in Japan-Hi-net, F-net,

us
K-NET and KiK-net”, Earth Planets Space 56, xv–xxviii.
an
32. Newmark, N. M. and Hall, W. J. [1973], "Procedures and criteria for earthquake resistant

design", Building Sci. Series, No.46, National Bureau of Standards, U.S. Dept. of Commerce,
M
Washington, D. C.

33. Park YJ, Ang AHS, Wen YK. [1985], “Seismic damage analysis of reinforced concrete
ed

buildings”, Journal of Structural Engineering, ASCE 111(4): 740–757.

34. Parvez, I. A., Vaccari, F. and Panza, G. F. [2003], "A deterministic seismic hazard map
pt

of India and adjacent areas", Geophys. J. Int., 155, 489-508.

ce

35. Pousse, G., Bonilla, L. F., Cotton, F. and Margerin, L. [2006], “Nonstationary stochastic

simulation of strong ground motion time histories including natural variability: Application to
Ac

the K-net Japanese database”, Bulletin of the Seismological Society of America, 96(6): 2103–

2117.

36. Priestly, M. J. N. [1993], "Myths and Fallacies in Earthquake Engineering - COnflicts

Between Design and Reality", Bull. NZNSEE, 26, 3, pp 329-341.

41
 37. Quek, S.T., Teo, Y.P. and Balendra. T. [1990], "Non-stationary structural response with

evolutionary spectra using seismological input model", Earthquake Engineering and Structural

Dynamics, 19, 275-288.

38. Raghukanth, S. T. G. and Sangeetha, S. [2013]. “Empirical mode decomposition of

t
ip
earthquake accelerograms”, Adv in adaptive data analysis, Vol. 4, No. 3 (2012) 1250023.

cr
39. Rosenblueth, E. [1964], "Probabilistic design to resist earthquakes", Journal of the
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

Engineering Mechanics Divison, ASCE, 90, 189-219.

us
40. Si, H. and Midorikawa, S., “New Attenuation Relationships for Peak Ground
an
Acceleration and Velocity considering effects of fault type and site condition”, Journal of

Structural and Construction Engineering, A.I.J., 523, 63-70, 1999 (in Japanese).
M
41. Shao, G., X. Li, C. Ji. and T. Maeda [2011], “Focal mechanism and slip history of 2011

Mw 9.1 off the Pacific coast of Tohoku earthquake, constrained with teleseismic body and
ed

surface waves”, Earth Planets Space, 63 (7), 559-564.

42. Suzuki, W., S. Aoi, H. Sekiguchi, and T. Kunugi [2011], “Rupture process of the 2011
pt

Tohoku-Oki mega-thrust earthquake (M9.0) inverted from strong-motion data”, Geophysics

ce

Research Letters, Vol.38, L00G16, doi: 10.1029/2011GL049136, 2011.

43. Skarlatoudis, A.A., and Papazachos, C. B. [2012], “Preliminary Study of the Strong
Ac

Ground Motions of the Tohoku, Japan, Earthquake of 11 March 2011: Assessing the Influence of

Anelastic Attenuation and Rupture Directivity”, Seismological Research Letters, Vol 83, No. 1,

doi: 10.1782/gssrl.831.1.119.

42
 44. Stewart, J. P., Midorikawa, S., Graves, R. W., Khodaverdi, K., Kishida, T., Miura, H.,

Bozorgnia, Y., and Campbell, K. W. [2013], “Implications of the Mw 9.0 Tohoku-Oki

Earthquake for Ground Motion Scaling with Source, Path, and Site Parameters”, Earthquake

Spectra, Vol 29, No. S1, pages S1–S21.

t
ip
45. Site classification for seismic design, International Buliding code, 2009: 340-343.

cr
46. Trifunac MD, Brady AG. [1975], “A study on the duration of strong earthquake ground
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

motion”, Bulletin of the Seismological Society of America 65(3): 581–626.

us
47. Vanmarcke, E. H. [1976], “Structural response to earthquakes”, Seismic Risk and
an
Engineering Decisions, ch 8, C Lomnitz and E Rosenblueth, Editors, Elsevier Publishing Co.,

Amsterdam, 287-337.
M
48. Von Thun, J. L., Rochim, L. H., Scott, G. A., and Wilson, J. A. [1988], "earthquake

ground motions for design and analysis of dams", Earthquake Engineering and Soil Dynamics II
ed

- Recent Advance in Ground Motion Evaluation, Geotechnical Special Publication 20, ASCE,

New York, 463-481.

pt

49. Wu, Z. and Huang, N. E. [2009], “Ensemble emprical mode decomposition: A noise-
ce

assisted data analysis method”, Adv. Adapt. Data Anal., 1(1): 1–41.

50. Wu, C. and Peng, Z. [2011], “Temporal changes of site response during the 2011 Mw 9.0
Ac

off the Pacific coast of Tohoku earthquake”, Earth Planets Space, 63, 791–795.

51. Yamamoto, Y. [2011], “Stochastic Model for Earthquake Ground Motion Using Wavelet

Packets”, Ph.D. Thesis, Department of Civil and Environmental Engineering, Stanford

University, Stanford, California.

43
 52. Zhao, J. X., , Zhang, J., Asano, A., Ohno, Y., Oouchi, T., Takahashi, T., Ogawa, H.,

Irikura, K., Thio, H. K., Somerville, P. G., Yasuhiro Fukushima, and Yoshimitsu Fukushima.

[2006], “Attenuation Relations of Strong Ground Motion in Japan Using Site Classification

Based on Predominant Period”, Bulletin of the Seismological Society of America, Vol. 96, No. 3,

t
pp. 898-913, doi: 10.1785/0120050122.

ip
53. Zhu, T.J., Tso, W.K. and Heidebrecht, A.C. [1987], "Effects of peak a/v ratio on

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

structural damage", Journal of Structural Engineering Division, ASCE, 1019-1037.

us
an
M
ed
pt
ce
Ac

44
 Table 1. Details of Ground Motion Parameters used in the current study

S.No Ground Motion Parameter Application in Engineering

t
1. Ground Motion damage Index (Gutenberg and Richter, 1956)

ip
Seismic Risk analysis (Cornell, 1968)

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

Peak Ground Acceleration

Design spectra (Newmark and Hall, 1973)

us
(PGA)

Seismic Hazard maps (Parvez et al. 2003)

an
Force based Seismic Design: Design value
M

2. Ground Motion damage Index (Rosenblueth,1964)

ed

Peak Ground Velocity Seismic Risk analysis (Cornell, 1968)

(PGV)
Design spectra (Newmark and Hall, 1973)
pt

Seismic Hazard maps

ce

3. Ground Design spectra (Newmark and Hall, 1973)

Ac

Peak

Displacement (PGD) Displacement based Seismic Design (Priestly, 1993)

45
 4. PGV/PGA Damage measure (Zhu et al, 1988)

5. Damage Index - in combination with PGA or PGV (Trifunac

Significant Duration (Tsig)
and Brady, 1975)

t
ip
6. Root-Mean-Square

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

Damage Index (Housner and Jennings, 1964)

Acceleration (arms)

us
7. Arias Intensity (Ia) Damage Index (Arias, 1970)
an
8. Cumulative Absolute
Damage Index (EPRI, 1988)
M
Velocity (CAV)
ed

9. Acceleration Spectrum
Seismic Design (Von Thun et al., 1988)
Intensity (ASI)
pt
ce

10. Velocity Spectrum

Seismic Design (Housner, 1952)
Intensity (VSI)
Ac

11. Characteristic Intensity (Ic) Damage Index (Park et al., 1985)

12. Predominant Frequency Site classification

46
 (Fp)

13. Central Frequency (Ω)

Stochastic response - Random Vibration techniques

(Vanmarcke, 1976)

t
14. Shape factor (q)

ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

15. Total Energy (Eacc)

us
16. Spectral Centroid (Ew) an
17. Spectral standard deviation

(Sw)
M
Constructing EPSD - computing responses to non-stationary

loads (Quek et al. 1990)

18. Temporal Centroid (Et)
ed

19. Temporal standard

pt

deviation (St)
ce

20. Correlation between time

Ac

and frequency (ρ(t,ω))

47
 Table 2. Site class definitions based on IBC, 2009.

Site
Soil profile name Soil shear wave velocity, Vs (m/s)
Class

t
ip
A Hard rock Vs >1500

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

B Rock 760< Vs ≤1500

C Very dense soil and soft rock

us
360< Vs ≤760
an
D Stiff soil profile 180≤ Vs ≤360
M

E Soft soil profile Vs <180

ed
pt
ce
Ac

48
 Table 3. Comparison of ground motion parameters among stations MYG004, MYG006 and
MYG007

Station Rrup Vs30 (m/s) Site Location Dur. Tsig (s) (Hz) PGA (g)

MYG00 119.4 224.9 D Fore Arc 300 64 6.30 2.02

t
4

ip
MYG00 118.0 170.0 E Fore Arc 300 105 2.29 0.50

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

MYG00
119.0 537.0 C Fore Arc 300 102 4.90 0.50

us
7
an
M
ed
pt
ce
Ac

49
 Table 4. Ground motion parameters obtained for strong motion records of KNet station
MYG004 and KIKNet station TCGH16: H-horizontal and V-vertical components

TCGH16 TCGH16
Ground MYG004
(surface) (borehole)

t
motion Units

ip
Parameter
H V H V H V

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
PGA g 2.02 1.92 1.02 0.82 0.17 0.14
an
PGV cm/s 78.48 39.24 68.67 39.24 21.582 14.715
M
PGD cm 16.45 9.07 17.97 4.92 12.40 5.91
ed

PGV/PGA s 0.039 0.0197 0.071 0.035 0.125 0.110

pt

Tsig s 82.78 64.07 42.72 16.16 63.72 83.82

ce

arms g 0.235 0.139 0.195 0.162 0.029 0.018

Ac

Ia cm/s 8505.3 2128.8 2756.6 725.94 90.25 47.09

CAV cm/s 17216 6875.8 9162.5 3492.4 1932.6 1540.2

50
 ASI cm/s 1530.4 716.13 1167.4 676.9 147.2 89.27

VSI cm 61.50 19.95 61.22 28.11 6.90 4.39

t
ip
25.76e+ 11.35e+
Ic 9.41e+3 4.19e+3 0.88e+3 0.54e+3
3 3

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
Fp s-1 5.27 8.32 1.98 4.36 0.52 0.54
an
Ω Hz 6.33 9.87 3.01 5.95 6.16 6.95
M
q - 0.45 0.37 0.46 0.55 0.75 0.82
ed

Eacc cm/s 10526 2358.3 2791.9 833.8 91.2 47.9

pt

Ew Hz 9.306 11.326 7.047 8.462 8.274 8.534

ce

Sw Hz 10.095 8.82 10.655 10.127 10.89 11.371

Ac

Et s 40.54 41.13 21.96 6.96 27.47 38.10

St s 24.29 17.04 8.53 3.30 12.90 16.76

51
 ρ(t,ω) - -0.009 -0.225 -0.059 0.126 0.113 0.038

MYG004: Rrup =119.3km; Vs30 =224.92 m/s; site class =D

t
TCGH16: Rrup =112.0km; Vs30 =213.19 m/s; site class =D

ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

52
 Table 5. Coefficients of the prediction equation (4.1) for the 20 Ground Motion Parameters: H-
horizontal and V-vertical components

GROUND

MOTION
Units C0 C1 C2 C3 C4 C5 σ(Ԑ)

t
PARAMETE

ip
R

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

H 46.902 0.004 -7.585 260.10 -0.621 -0.475 0.617

us
PGA g
4
V 25.571 0.003 -4.736 112.80 -0.346 -0.379 0.58
an
H 18.623 -0.001 -2.159 115.80 -0.634 -0.304 0.544
PGV cm/s
M
V 16.187 -0.001 -2.103 222.75 -0.262 -0.159 0.453

6
H 31.063 0 -3.881 760.58 -0.452 -0.199 0.505
ed

PGD cm
V 25.028 -0.001 -3.28 850.71 -0.128 0.012 0.416
pt

H -44.215 -0.005 6.658 459.48 -0.011 0.169 0.473

ce

PGV/PG
s 1
A
V -17.588 -0.004 2.818 75.507 0.084 0.218 0.468
Ac

H -2.819 -0.002 1.487 114.56 -0.116 -0.03 0.341

Tsig s
9
V -0.297 -0.002 1.108 18.181 -0.093 -0.068 0.334

53
 H 38.593 0.004 -6.584 214.48 -0.653 -0.447 0.594
arms g
2
V 21.138 0.003 -4.305 96.62 -0.341 -0.319 0.534

H 85.248 0.006 -12.188 244.88 -1.422 -0.924 1.21

t
Ia cm/s

ip
9
V 51.63 0.004 -7.833 124.63 -0.775 -0.707 1.074

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

H 46.286 0.002 -5.652 298.24 -0.775 -0.468 0.661

CAV cm/s

us
V 30.667 0.001 -3.67 183.01 -0.437 -0.376 0.589
an
H 162.34 0.011 -22.757 746.86 -0.724 -0.579 0.68

ASI cm/s 2 4
M
V 93.673 0.008 -13.653 473.40 -0.484 -0.397 0.577

3
ed

H 42.074 0.001 -5.281 403.45 -0.81 -0.337 0.598

VSI cm
V 55.308 0.001 -7.375 706.93 -0.411 -0.251 0.528
pt

5
H 68.264 0.005 -9.346 232.51 -1.038 -0.685 0.894
ce

Ic 2
V 42.751 0.003 -6.031 112.63 -0.558 -0.513 0.795
Ac

Fp H s-1 128.23 0.009 -18.366 872.11 -0.271 -0.199 1.137

7 8

54
 V 47.038 0.007 -7.672 228.66 -0.307 -0.546 1.204

H 743.08
25.907 0.001 -3.806 0.237 -0.255 0.54
Ω Hz 9

t
ip
V 53.286 0.004 -7.652 699.31 -0.034 -0.399 0.606

cr
1
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

H -10.332 -0.001 1.387 723.44 0.077 0.039 0.203

us
q - 4

V -2.658 0 0.396 13.083 0.035 0.045 0.165

an
H 82.148 0.005 -11.724 249.41 -1.418 -0.944 1.226
M

Eacc cm/s

V 123.01
49.653 0.003 -7.579 -0.77 -0.732 1.087
ed

H 43.983 0.002 -6.167 809.96 -0.078 -0.14 0.738

pt

Ew Hz 3
ce

V 81.062 0.006 -11.674 729.65 -0.194 -0.379 0.781

7
Ac

H 47.988 0.004 -6.995 719.60 0.013 -0.055 0.761

9
Sw Hz
724.41
V
51.74 0.004 -7.446 -0.081 -0.263 0.72
5

55
 H 380.07
-12.225 -0.003 2.648 -0.063 0.093 0.333
Et s 5

V -1.426 -0.002 1.076 42.123 -0.047 0.015 0.305

t
ip
306.34
H
-13.952 -0.003 2.938 -0.113 -0.025 0.387

cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

5
St s

us
V -2.936 -0.002 1.268 37.539 -0.072 -0.046 0.351
an
H 735.89
12.887 0.002 -1.958 0.011 0.027 0.122
1
M
*
ρ(t,ω) -
387.79
V
ed

7.912 0.002 -1.33 1 0.011 0.044 0.134

pt

*The coefficients for ρ(t,ω) are derived for non logarithmic values.[i.e., the prediction equation
in this case is GMP=c0 + c1Rrup+ c2 ln(Rrup+ c3) + c4 ln(Vs30) + c5 Larc+ ε]
ce
Ac

56
 Figure 1. Seismo-tectonic map showing fault lines of Japan and active plate boundaries around
the country. The 2011 Mw 9.0 Tohoku earthquake occurred due to the energy release near plate
boundary (red line) of Pacific and North American plates and epicentre is estimated to be located
130km east of Sendai city (Japan Meteorological Agency, 2011). F1 and F2 are the respective
northern and southern segments of the fault model given by Geospatial Information Authority of
Japan (GSI).

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

57
 Figure 2. Distribution of KNet (blue) and KIKNet (red) stations that have recorded the Tohoku
earthquake, Japan. Seismographs of KNet are located at ground surface and KIKNet consist of a
pair of seismographs located at surface and at deep end of the borehole.

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

58
 Figure 3. Acceleration, Velocity and Displacement time histories showing peak values for
ground motion records of KNet station MYG004 [EW, NS, UD] and KIKNet station TCGH16
[Borehole records: EW1, NS1, UD1; Surface records: EW2, NS2, UD2]

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
.
M
ed
pt
ce
Ac

59
 Figure 4. Variation of Peak Horizontal Acceleration, PHA (g) for ground motion records of
KNET and KIKNet [surface] stations. The station MYG004 with a fault rupture distance of
119.4km recorded a PHA of 2.02g.

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

60
 Figure 5. Variation of (a) Horizontal and (b) Vertical components of GMPs: Peak Ground
Acceleration (PGA), Peak Ground Velocity (PGV), Peak Ground Displacement (PGD) with
distance from the fault rupture (Rrup) showing recorded values (marker) and predicted values
(line) for different soil classes [A (Dark red), B (red), C (yellow), D (green), E (blue)]

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

61
 Figure 6. Variation of V/H (Vertical-to-Horizontal component) ratios of PGA, PGV and PGD
with Peak Horizontal values for different soil classes [A (Dark red), B (red), C (yellow), D
(green), E (blue)] and epicentral distances (increasing marker size with distance)

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

62
 Figure 7. Variation of (a) Horizontal and (b) Vertical components of GMPs: PGV/PGA,
Significant duration (Tsig) with distance from the fault rupture (Rrup) showing recorded values
(marker) and predicted values (line) for different soil classes [A (Dark red), B (red), C (yellow),
D (green), E (blue)]

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

63
 Figure 8. Variation of (a) Horizontal and (b) Vertical components of GMPs: R-M-S
Acceleration (aRMS), Arias Intensity (Ia), Cumulative Absolute Velocity (CAV) with distance
from the fault rupture (Rrup) showing recorded values (marker) and predicted values (line) for
different soil classes [A (Dark red), B (red), C (yellow), D (green), E (blue)]

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

64
 Figure 9a. Spectral acceleration (Sa) plots for ground motion components of stations (a)
MYG004 [EW, NS, UD] and (b) TCGH16 [Borehole: EW1, NS1, UD1; Surface: EW2, NS2,
UD2] over a 5% critical damping ratio.

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

65
 Figure 9b. Spectral velocity (Sv) plots for ground motion components of stations (a) MYG004
[EW, NS, UD] and (b) TCGH16 [Borehole: EW1, NS1, UD1; Surface: EW2, NS2, UD2] over a

t
5% critical damping ratio.

ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

66
 Figure 10. H/V response spectral ratio plots for ground motion components of stations
MYG004, MYG006 and MYG007 over a 5% critical damping ratio

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

67
 Figure 11. H/V response spectral ratio plots for ground motion components of stations MYG004
and TCGH16 [Borehole, Surface] over a 5% critical damping ratio

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

68
 Figure 12. Variation of Rupture distance (Rrup) and average shear wave velocity up to top 30 m
of the soil column (Vs30) with maximum Horizontal-to-Vertical (H/V) response spectra and
corresponding Predominant period (Tp).

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

69
 Figure 13. Variation of (a) Horizontal and (b) Vertical components of GMPs: Acceleration
Spectrum Intensity (ASI), Velocity Response spectrum Intensity (VSI) with distance from the
fault rupture (Rrup) showing recorded values (marker) and predicted values (line) for different
soil classes [A (Dark red), B (red), C (yellow), D (green), E (blue)]

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

70
 Figure 14. Variation of (a) Horizontal and (b) Vertical components of GMPs: Characteristic
Intensity (Ic) with distance from the fault rupture (Rrup) showing recorded values (marker) and
predicted values (line) for different soil classes [A (Dark red), B (red), C (yellow), D (green), E
(blue)]

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

71
 Figure 15. Power Spectral Density plots for ground motion components of stations (a) MYG004
[EW, NS, UD] and (b) TCGH16 [Borehole: EW1, NS1, UD1; Surface: EW2, NS2, UD2].

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

72
 Figure 16. Variation of (a) Horizontal and (b) Vertical components of GMPs: Predominant
Frequency (Fp), Central Frequency (Ω), Shape factor (q) with distance from the fault rupture
(Rrup) showing recorded values (marker) and predicted values (line) for different soil classes [A
(Dark red), B (red), C (yellow), D (green), E (blue)]

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

73
 Figure 17. Evolutionary Power Spectral Density (EPSD) plots for ground motion time histories
(EW component) of stations (a) MYG004 (b) TCGH16-Surface (c) TCGH16-Borehole.

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

74
 Figure 18. Variation of (a) Horizontal and (b) Vertical components of GMPs: Spectral Centroid
(Ew), Spectral standard deviation (Sw), Temporal centroid (Et), Temporal standard deviation (St)
and Correlation coefficient (ρ(t,ω)) with distance from the fault rupture (Rrup) showing recorded
values (marker) and predicted values (line) for different soil classes [A (Dark red), B (red), C
(yellow), D (green), E (blue)]

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

75
 Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

Figure 18b:
Ac
ce
pt
ed

76
M
an
us
cr
ip
t
 Figure 19. GMP classification based on Principal Component Analysis:

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M

(Top Right, Anti-clockwise) The first quadrant constitutes of GMPs representing peak values,
ed

energy and intensity parameters. All the GMPs characterizing duration of ground motion are

grouped in to the Second quadrant and the frequency content parameters of PSD and EPSD are
pt

classified in to the Fourth quadrant (with the exception of Sw). The corresponding quadrants are

named Amplitude, Duration and Frequency Spaces respectively.

ce
Ac

77
 Figure 20. Representation of ground motion records in the Principal Component Space. Here, P1
indicates single parameter records of GMPs in the Amplitude space, P2 represents that of
duration space and P4 are the single parameter records of Frequency parameters. Since the
amplitude parameters are varying orthogonal to that of the interpolated Frequency-Duration axis,
ground motion records closer to this axis are considered to be least damaging and vice versa.

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

78
 Figure 21. Variation of distance from Frequency-Duration axis (dF-D) for all the ground motion
records: MYG004 is the farthest station, whereas most of the far-field stations are at negligible
distances from the F-D axis.

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

79
 Figure 22. Comparison of GMPEs of Peak Horizontal Acceleration (PHA) obtained for the
present model with other empirical GMPEs: Zhao et al., (2006), Kanno et al., (2006) and Si and
Midorikawa (1999), for a Mw 9.0 earthquake.

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

80
 Figure 23. Comparison of GMPEs of Peak Horizontal Velocity (PHV) obtained for the present
model with other empirical GMPEs: Kanno et al., (2006) and Si and Midorikawa (1999), for a
Mw 9.0 earthquake.

t
ip
cr
Downloaded by [UC Santa Barbara Library] at 23:15 02 September 2017

us
an
M
ed
pt
ce
Ac

81