Vous êtes sur la page 1sur 12

Soil Dynamics and Earthquake Engineering 100 (2017) 345–356

Contents lists available at ScienceDirect

Soil Dynamics and Earthquake Engineering


journal homepage: www.elsevier.com/locate/soildyn

Amplification of strong ground motions at Heathcote Valley during the MARK


2010–2011 Canterbury earthquakes: Observation and 1D site response
analysis

Seokho Jeonga, , Brendon A. Bradleya,b,1
a
QuakeCoRE, University of Canterbury, Private Bag 4800, Christchurch 8140, New Zealand
b
Department of Civil and Natural Resources Engineering, University of Canterbury, Private Bag 4800, Christchurch 8140, New Zealand

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

Keywords: The strong motion station at Heathcote Valley School (HVSC) recorded intense ground accelerations (peak value
Heathcote Valley of 1.4 g in horizontal and 2.2 g in vertical component) during the 22 February 2011 Christchurch earthquake.
Christchurch earthquake More importantly, ground motions recorded at HVSC in numerous other events during the 2010–2011
Site effects Canterbury earthquake sequence also exhibited consistently larger peak ground accelerations compared with
Site response simulation
nearby strong motion stations, which suggests significant near-surface site amplification effects.
Ground motion
This paper presents a quantitative case study of near-surface site effects of Heathcote Valley during the
Canterbury earthquakes
2010–2011 Canterbury earthquake sequence, by means of one-dimensional nonlinear dynamic finite element
analyses based on the recorded ground motions and a rigorous site characterisation study. Various geophysical
and geotechnical in-situ tests are performed to establish a simplified three-dimensional representation of wave
velocities, mass densities and the geological structure (i.e. layering) of Heathcote Valley. Simulations are per-
formed using the velocity profile at the location of station HVSC with the finite element analysis program
OpenSees. Overall, simulations agree well with the recorded motions and suggest that ground motions at HVSC
are amplified in a wide band of frequencies. However, the one-dimensional simulations tend to underestimate
the site response at frequencies higher than the site fundamental frequency, likely due to its inability of mod-
elling surface waves caused by the inclined soil-rock interface. Comparison between the nonlinear and the
equivalent linear model shows that, although both approaches produce similar level of peak amplitude, the
equivalent linear model significantly underestimates the high frequency motions.

1. Introduction school station (HVSC) during the February 2011 event have received
particular attention, with recorded peak ground accelerations (PGA) of
The 2010–2011 Canterbury earthquake sequence—which includes 2.2 g in the vertical component and 1.4 g in the horizontal component
the September 2010 MW7.1 Darfield earthquake, the February 2011 [8,12,18]. More importantly, the intensities of ground motions at HVSC
MW6.2 Christchurch earthquake, and over 30 aftershocks with magni- were consistently higher than nearby stations throughout the sequence
tude larger than ML5.0—caused severe ground shaking in urban of earthquakes, which suggests strong near surface site effects [7].
Christchurch, which led to 185 casualties (during the February 2011 Heathcote Valley is located near the northern side of Port Hills
event) and extensive structural and infrastructure damage where the depth to the volcanic bedrock is expected to be shallow.
[11,25,18,13]. These repeated damaging events have provided an un- Based on 1D wave propagation theory the large impedance contrast
precedented level of high-quality ground motion data recorded ex- resulting from shallow soft soils overlying the competent Port Hills
tensively by the New Zealand strong motion station network GeoNet volcanics may cause strong amplification of ground motion amplitude
(https://www.geonet.org.nz/) at numerous locations in the Canterbury in high frequencies, not to mention the possibility of further aggrava-
region [8,5,9,7]. tion due to wave inteference caused by 2D/3D subsurface geometry.
Among the numerous ground motion data recorded during the In order to elucidate the principal phenomena resulting in the
earthquake sequence, the ground motions of the Heathcote Valley consistently large recorded ground motions, this study presents the


Corresponding author.
E-mail address: seokho.jeong@canterbury.ac.nz (S. Jeong).
1
Present address: UPS Visiting Professor, Stanford University, Stanford, CA

http://dx.doi.org/10.1016/j.soildyn.2017.06.004
Received 13 June 2016; Received in revised form 11 May 2017; Accepted 7 June 2017
0267-7261/ © 2017 Elsevier Ltd. All rights reserved.
S. Jeong, B.A. Bradley Soil Dynamics and Earthquake Engineering 100 (2017) 345–356

result of site-specific site response simulations at Heathcote Valley. Table 1


Detailed site investigation data comprising pre-existing data and re- List of earthquakes used in the analyses, in chronological order.
cently performed geotechnical and geophysical testing are used to de-
HVSC LPCC
velop a realistic quantitative representation of the local geology of
Heathcote Valley. A brief summary of the geological and geotechnical Event date MW Rrupa PGAb PGVb Rrup PGA PGV
data collected for modelling is first presented. Simulations are per- (km) (g) (cm/s) (km) (g) (cm/s)
formed using a finite element code OpenSees [27], with deconvolved
04/09/2010 7.1 20.8 0.61 29 22.4 0.29 19
and amplitude-corrected ground motions recorded at the nearby Lyt- 19/10/2010 4.8 12.8 0.09 3.2 13.1 0.02 0.71
telton Port Company station (LPCC) as input motions at the bedrock 26/12/2010 4.7 4.7 0.11 2.9 7.7 0.02 0.65
level. Numerical simulations are compared with recorded ground mo- 22/02/2011 6.2 3.9 1.41 81 7.0 0.92 46
tions in terms of acceleration time series and the acceleration response 16/04/2011 5.0 7.3 0.68 32 5.2 0.29 8.5
13/06/2011 (a) 5.3 4.7 0.45 14 5.3 0.15 5.4
spectra, followed by a discussion on the performance and limitations of
13/06/2011 (b) 6.0 3.6 0.91 55 5.8 0.64 33
the model. Based on the result of numerical simulations, discussion is 21/06/2011 5.2 14.9 0.26 8.0 15.6 0.07 2.1
also given to the effect of the soil nonlinear response and the pressure 23/12/2011 (a) 5.8 9.9 0.31 12.7 11.4 0.24 7.6
dependency of dynamic soil properties on the intensity of the ground 23/12/2011 (b) 5.9 9.7 0.44 22 12.4 0.44 23
motions. a
The shortest source-to-site distance based on Beavan et al. [4].
b
Horizontal fault-normal component.
2. Site overview and considered earthquake events
easily be biased by the uncertainties in the reference input motions if
2.1. Site overview simulations are not performed with sufficient number of recorded
motions. However, selecting the ground motions for site response si-
Heathcote Valley is a suburb of Christchurch, approximately 8 km mulations always involves a trade-off between the number of events
southeast of the city centre, located at the foot of volcanic ridges that and the quality of recorded ground motions. In an urban environment
form the northern edge of the Port Hills. Fig. 1 shows the location of the like Christchurch, strong motion stations may record not only the
valley, the stations HVSC and LPCC, and the epicentres of the earth- ground shaking caused by earthquake events but also numerous other
quake events considered in this study. The HVSC strong motion station anthropogenic activities. Recorded ground motions for smaller earth-
is located at the Heathcote Valley school, near the edge of the valley, quakes often have poor signal-to-noise ratio, and the noise is often
where shallow firm colluvium sediments overlie the Port Hills volca- difficult to isolate.
nics. The thickness of surficial soils (i.e. depth to the volcanic rock) Based on above considerations, this study employs a set of ground
reaches 20–30 m in the centre of the valley. Loess is predominant in the motions from ten earthquake events that was utilised in a previous
surficial soils, and originates from glacial and river erosion of the study by Bradley [7]. The selected earthquake events have moment
Southern Alps during the Quaternary that is transported to Port Hills magnitudes MW 4.7–7.1, and produced significant ground motions in
through Aeolian processes [10]. The Lyttelton Port (LPCC) strong mo- the urban Christchurch area (and thus have good signal-to-noise ratios).
tion station is located on a volcanic rock site, covered with a thin layer These events are listed in Table 1 in chronological order, along with
(approximately 6 m) of loess [32], and approximately 3 km south- their corresponding moment magnitude (MW), source-to-site distance
southeast of HVSC. (Rrup), peak ground acceleration (PGA), and peak ground velocity (PGV)
recorded at stations HVSC and LPCC for each event. In Table 1, both
2.2. Events considered in this study PGA and PGV are for the horizontal fault-normal component, and Rrup is
the shortest distance from the station to the fault rupture surface. The
This study takes advantage of the large set of strong ground motions source-to-site distances Rrup are calculated based on the finite fault
recorded by GeoNet during the 2010–2011 Canterbury earthquakes, models by Beavan et al. [4] for the four largest events (04/09/2010,
which include four major earthquakes of MW ≥ 5. 9 and more than 300 22/02/2011, 13/06/2011(b), 23/12/2011(b)), and based on simplified
aftershocks of ML > 4.0 . Considering a larger number of events is de- approximations of the finite faults for the smaller events [7].
sirable for this type of study, because the simulated ground motions can

Fig. 1. Locations of Heathcote Valley (HVSC) and Lyttelton Port


(LPCC) strong motion stations and epicentres of the simulated
earthquake events. The inset shows a close-up of Heathcote Valley
and the surrounding area.

346
S. Jeong, B.A. Bradley Soil Dynamics and Earthquake Engineering 100 (2017) 345–356

Fig. 2. Recorded north-south component accelera-


tion time series for the MW7.1 2010 Darfield earth-
quake (04/09/2010) at HVSC and 4 closest strong
motions stations.

3. Observational evidence of large site amplification at Heathcote 3.2. Comparison of recorded motions and empirical predictions
Valley
Bradley [7] examined the observed ground motions at 20 strong
3.1. Comparison of recorded motions: HVSC vs nearby stations motion stations during the 2010–2011 Canterbury earthquake se-
quence, to demonstrate the site-specific systematic effects by relaxing
During the Canterbury earthquake sequence, the station HVSC re- the conventional ergodic assumption [2]. In that study, the non-ergodic
peatedly recorded acceleration time series that have amplitudes much empirical ground motion is expressed as:
higher than the nearby strong motion stations [7]. As an example, Fig. 2
lnSAes = fes (Site , Rup) + δBe + δWes
shows the recorded north-south component acceleration time series for
the MW7.1 2010 Darfield earthquake at HVSC and the four closest = fes (Site , Rup) + (δL2Ll + δBel0 ) + (δS 2Ss + δWes0 ) (1)
strong motion stations. The source-to-site distance, Rrup, for this event is
where lnSAes is the natural logarithm of the observed spectral accel-
20.8 km at HVSC and 22.4 km at LPCC, and the epicentre is approxi-
eration (SA), and the first, second, and third terms on the right hand
mately 43 km west of HVSC. The recorded acceleration time series at
side represent the median prediction of SA, the between-event residual,
HVSC clearly shows much higher amplitude compared with nearby
and the within-event residual, respectively.
strong motion stations that have similar source-to-site distances and
Fig. 4 shows the within-event residual, δWes and the systematic site-
back-azimuths, which demonstrates the near-surface site amplification.
to-site residual, δS 2SS (the mean of δWes , which represents the site-
It is also evident that the ground motions at HVSC and LPCC are richer
specific effects), of Heathcote Valley and all stations for all earthquake
in high frequency content, which is likely due to the response of the
events considered. The site-to-site residual at HVSC clearly demon-
shallow layer of sediments that overlies the competent Port Hills vol-
strates the amplification at vibration period T < 0.5 s specific at this
canic rock. Ground motions recorded at HVSC for other events also
station that is significantly greater than the empirical prediction.
show consistently higher PGA compared with other strong motions
The local geology of Heathcote Valley is characterised by a thin
stations. Fig. 3 shows the three component (fault-normal, fault-parallel,
layer of soils over the Port Hills volcanics. Based on 1D wave propa-
and vertical) acceleration time series recorded at HVSC and LPCC for
gation theory, it is expected that the dynamic response of this shallow
the events at 13/06/2011(b), 22/02/2011, and 04/09/2010, which
soil layer over volcanic rock would greatly amplify ground motions at
strongly suggests the amplification of ground motions due to local site
high frequencies. The following sections seek to characterise the geo-
effects at HVSC.
technical and geophysical properties of the site, and examine the extent

Fig. 3. Comparison of the three component time series at HVSC and LPCC for events 13/06/2011, 22/02/2011, and 04/09/2010, which produced some of the largest ground amplitudes
recorded at the sites.

347
S. Jeong, B.A. Bradley Soil Dynamics and Earthquake Engineering 100 (2017) 345–356

Fig. 4. (a) Within-event residuals for individual events (grey lines), and the site specific effect δS 2S (bold black line) at HVSC; (b) δS 2S all 20 stations in urban Christchurch. (modified
from Bradley [7]).

to which 1D wave propagation theory can capture the observed ground Table 2
motions at HVSC. Locations and the depths of refusal of sCPT sites.

Site code Latitude [°] Longitude [°] Refusal depth [°]

4. Site characterisation sCPT1 −43.579214 172.708726 13.5


sCPT1a −43.578493 172.708742 6.5
sCPT2 −43.581878 172.708107 20.9
To characterise the important dynamic properties of Heathcote
sCPT3 −43.580573 172.708386 34.8
Valley soils and the underlying volcanics, various geophysical and sCPT4 −43.579154 172.706423 19.8
geotechnical tests have been conducted. These tests consist of fifteen sCPT6 −43.580782 172.705771 21.1
seismic piezozone tests (sCPT), five active source surface wave tests sCPT7 −43.579168 172.704937 9.2
(MASW), and fifteen ambient vibration tests for horizontal-to-vertical sCPT8 −43.575958 172.702934 23.4
sCPT9 −43.584877 172.595905 2.1
(H/V) spectral ratio. The detailed plan of these test locations relative to
sCPT10 −43.582946 172.594754 8.8
the valley geometry and the HVSC station is shown in Fig. 5. sCPT11 −43.579716 172.707944 26.0
sCPT12 −43.576130 172.707020 24.9
sCPT13 −43.577447 172.707942 27.1
sCPT14 −43.578346 172.710923 7.4

4.1. Seismic cone penetration test

To characterise the shear wave velocity (VS) and thickness of the


loess, fifteen seismic piezocone tests (sCPT) were conducted by Fugro
Geotechnical (NZ) at locations shown in Fig. 5, using a 20-tonne CPT rig
with their 15 cm2 seismic piezocone. Table 2 shows the locations of the
sCPT sites and the depths of refusal. Shear waves were generated using
a sledgehammer and a steel capped beam, and the shear wave velocities
are computed at the alternating intervals of 0.5 and 1.5 m, using two
geophone receivers installed in the cone.

Fig. 5. Locations of site characterisation tests at Heathcote Valley comprising sCPT, H/V, Fig. 6. Measured shear wave velocity as function of depth from 15 sCPT tests and the
and MASW. The location of the strong motion station (HVSC) and surface topography are power-law equation fitted to the data. Nsites indicates the number of available data at each
also noted. depth interval. Data in the upper five metres are excluded from fitting as noted in the text.

348
S. Jeong, B.A. Bradley Soil Dynamics and Earthquake Engineering 100 (2017) 345–356

Fig. 6 shows the obtained sCPT velocity profiles of individual sites vertical (H/V) spectral ratios, obtained from the ambient vibrations
with grey lines. Because the termination depth of CPT differs location- measured on the ground surface using a single three-axis seismometer,
by-location, the total number of measurement at each depth increment can provide important dynamic properties of the site, such as the fun-
decreases as the depth increases. At every depth interval, the variable damental vibration frequency, f0.
Nsites noted in Fig. 6 indicates the number of sites with available data. Ambient vibrations were measured at 26 sites at Heathcote Valley
Fig. 6 shows that the loess VS is strongly depth dependent, a typical using broadband seismometers for the duration of half an hour.
characteristic of non-plastic granular materials. To idealise the mea- Measured ambient vibration records are sub-divided into 180 s win-
sured velocity profiles in the numerical model, a power law equation dows. H/V spectral amplitude ratios are then computed, for each in-
was fitted to the measured velocities as shown in Fig. 6 and the fol- dividual window, using the geometric mean of the two horizontal
lowing equation: components and the vertical component Fourier spectral amplitudes,
smoothed using the Konno-Ohmachi smoothing window with the
VS = 207z 0.25 (2)
smoothing parameter, b = 40 [20]. The above-mentioned processing is
where z is the depth below ground surface in metres and VS is in m / s . performed using the Geopsy software suite [31].
Overall, Eq. (2) is very close to the median of measured velocities. It is Among the 26 sites, 15 sites shown explicitly in Fig. 5 are located
worthwhile to note that the measured velocities in the top five metres within the area of interest and were utilised in this study. Fig. 8 shows
are excluded from the model fitting, because most sCPT sites have examples of the H/V spectral ratios computed from the measured am-
paved surface or other type of foreign materials at the top few metres, bient vibrations at HV_H2 and HV_H18, both of which have a clear
and there are also difficulties in using sCPT to obtain VS at such shallow dominant spectral ratio peak. Since the near surface geology of
depths. Heathcote Valley can be characterised by a single layer of shallow soils
on the Port Hills volcanics with a large impedance contrast, it is ex-
4.2. Active source surface wave analysis pected that the peak response in H/V ratios would be dominated by the
response of the shallow soils. Table 3 summarises the site locations and
the H/V site frequencies, f0HV , the depths to the bedrock, and the
In addition to sCPT, we also performed active source surface wave
average shear wave velocity of the soil layer, estimated by solving the
tests (MASW) at five sites within the valley (see Fig. 5). For each survey,
following equation in terms of depth H assuming VS can be approxi-
24 vertical and 24 horizontal geophones were used at either 1.5 m or
mated by Eq. (2):
2 m spacing, depending on the space available, with a 5 kg sledge
hammer as the active source. Surface wave dispersion curves were 1 H dz
obtained by the frequency domain beamforming technique [17], which f0HV
=4 ∫0 VS (z ) (3)
were used for the inversion analyses using Geopsy [31]. For the in-
version, we constrained the velocity of the soil to be represented by a Among the fifteen H/V sites, the fundamental frequencies of eleven
power law equation (we did not enforce the coefficients) based on our sites are compared with sCPT, in which eight of them are compared
prior knowledge of the local geology. Fig. 7 shows the velocity profiles directly with nearby sCPT measurements and the fundamental fre-
obtained by inversion at sites MASW1 and MASW3 compared with quencies of three additional sites (HV_H3, HV_H17, and HV_H26) are
nearby sCPT data. Overall, velocity profiles obtained by MASW com- compared with the interpolation of nearby sCPT measurements. The
pared favourably with sCPT, and more importantly, MASW was able to fundamental frequencies, f 0CPT , from sCPT are computed using the re-
identify the depth to the volcanic rock, which was also broadly con- fusal depth and Eq. (2). Fig. 9 compares the ratios of the fundamental
sistent with sCPT measurements. frequencies, f0HV / f 0CPT , demonstrating that the fundamental frequencies
from the two different techniques are broadly consistent.
4.3. Ambient noise horizontal-to-vertical spectral ratio Since the site frequency from H/V ratio would represent the local
average, and the site frequency from the sCPT is from a fixed location,
Previous studies [28,22,20] have shown that the horizontal to the variability of the ratio is relatively large. For example, it is possible

Fig. 7. Examples of velocity profiles inverted from the Rayleigh wave dispersion curves compared with co-located sCPT measurements: (a) MASW1 and (b) MASW3.

349
S. Jeong, B.A. Bradley Soil Dynamics and Earthquake Engineering 100 (2017) 345–356

Fig. 8. Examples of ambient vibration H/V spectral ratios: (a) HV_H2 and (b) HV_H18.

Table 3
Summary of H/V test results used in this study.

Site code Latitude [°] Longitude [°] f0HV [Hz] Estimated Average VS
Depth [m] [m/s]

HV_H1 −43.582154 172.707713 5.0 15.5 309


HV_H2 −43.58057 172.708228 3.3 26.8 354
HV_H3 −43.58036 172.706931 4.7 16.9 316
HV_H4 −43.579156 172.706311 4.2 19.4 327
HV_H5 −43.575865 172.711026 2.1 48.5 410
HV_H6 −43.575808 172.708661 1.6 69.9 449
HV_H7 −43.576039 172.706802 1.7 63.8 439
HV_H8 −43.576055 172.704874 2.3 44.2 401
HV_H9 −43.575957 172.702907 5.7 12.9 295
HV_H15 −43.583127 172.706349 8.4 7.8 260 Fig. 10. Fence diagram showing the 3D VS model of Heathcote Valley. Northings and
HV_H16 −43.580725 172.705650 4.7 16.9 316 Eastings are in New Zealand Transverse Mercator 2000 (NZTM2000) projection. The scale
HV_H17 −43.578733 172.710107 5.2 14.8 306 in all three axes are identical.
HV_H18 −43.577706 172.707796 2.0 50.7 415
HV_H19 −43.577431 172.710723 3.7 23.3 342
HV_H26 −43.579811 172.709192 3.8 22.3 338
estimated depth of volcanic rock within the valley, using the Kriging
algorithm [26]. Fig. 10 shows a fence diagram of the 3D VS model of
Heathcote Valley obtained by the interpolation, in which the pressure-
dependent shear wave velocity of the soil is approximated by Eq. (2),
which was solely derived from sCPT data.
The VS of the volcanic rock at Heathcote Valley is estimated from
the velocity profiles obtained by MASW, and it is assumed to vary be-
tween VS (Rock ) = 800 − 1500 m / s , which is consistent with a previous
study done near HVSC and LPCC [32] and borehole logs performed at
different locations on Port Hills [24]. We assumed that the top 20 m of
the weathered volcanic layer with VS = 800 m / s mantles the intact rock
with VS = 1500 m / s . It is expected that the assumed rock VS profile
would have a large uncertainty. However, a parametric study showed
that the possible variations in the rock VS would cause only moderate
difference in the Fourier spectral amplification factor, without affecting
the fundamental site frequency.

5. 1D site response simulation


Fig. 9. Comparison of one-dimensional fundamental site frequencies obtained from H/V
ratios and sCPT.
5.1. Model description
that the cone stops penetrating at a boulder before it reaches the bed-
rock which would overestimate the site frequency (negative ratio in To quantify the effect of the site amplification on the observed
Fig. 9). Another possibility is that the bedrock depth at a sCPT site is ground motions at HVSC, we performed a series of 1D finite element
slightly deeper than the surrounding area; then f 0CPT would be slightly analyses based on the model shown in Fig. 10, using OpenSees. The
lower than f0HV (positive ratio in Fig. 9). dynamic response characteristics of loess is simulated by the pressure
dependent multi yield (PDMY) plasticity model [33], developed speci-
fically for modelling the constitutive behaviour of cohesionless soils. A
4.4. Simplified 3D model of Heathcote Valley few applications of the PDMY plasticity model for the dynamic response
simulation of dry cohesionless soils can be found in [29], [1], [15].
The sediment thickness data, obtained from sCPT, H/V spectral The shear wave velocity profile of the site used in the simulation is
ratio, and MASW were spatially interpolated to develop a model for shown in Fig. 11. The friction angle, ϕ, and the apparent cohesion

350
S. Jeong, B.A. Bradley Soil Dynamics and Earthquake Engineering 100 (2017) 345–356

motions usually involves selecting an appropriate reference station, and


the subsequent deconvolution analyses to invert the surface motion to
the incident bedrock motion at a target depth. The reference station
would be preferably not too far from the target site, to minimise the
error in the estimated incident motion that can be caused by the path
effects. It is also preferable that the recorded motion at the reference
station is free of nonlinear site response, because the deconvolution
analyses is based on either linear elastic or equivalent linear scheme.
For this study, we chose to use the recorded motions at LPCC, which
is located at the Lyttelton Port (3 km away from Heathcote Valley; see
Fig. 1) on a shallow 6 m layer of surficial fill soils (VS = 300 m / s )
overlying the volcanic rock (VS = 1500 m / s ) [32]. To obtain the in-
cident rock motions, recorded suface motions at LPCC were decon-
volved from its 1D elastic site response using the Thomson-Haskell
transfer matrix method [16].
Fig. 11. Shear wave velocity profile at HVSC. Even though the two stations are relatively close to each other, most
of the events considered in this study are also relatively close to
Heathcote Valley. Therefore, it is expected that the amplitudes of esti-
intercept, c, required for modelling soil behaviour, which cannot be mated incident rock motions would be still somewhat affected due to
directly determined by the in-situ tests performed for this study, are the different propagation distances. This potential source of error in the
adopted from published lab test data of loess samples obtained at amplitude of estimated input motions was mitigated by applying cor-
Clifton, a neighborhood in Port Hills 4 km northeast of Heathcote rection factors obtained based on a New Zealand-specific empirical
Valley [24]. The nonlinear shear behaviour is described by a shear ground motion model [6].
stress-strain backbone curve modelled by the modified hyperbolic
model [19,14], and the unloading/reloading behaviour is modelled by 6. Validation of 1D site response model
the Masing rule. The apparent cohesion intercepts, c, obtained from lab
tests were usually lower than c = 10 kPa . However, the model in this 6.1. Comparison of acceleration time series
study assumes c = 30 kPa , because using a too low value of c with
PDMY model may cause premature failure of near surface soils when Fig. 12 shows the comparison of recorded and simulated horizontal
combined with the sloping ground and the strong ground shaking. The acceleration time series at HVSC along the valley transverse direction
model was tested with a number of different values for c (5, 10, 30 and (N75E), for the four most notable events: MW 7.1 September 2010,
100 kPa) to ascertain the assumed value for c is acceptable, and it was MW 6.2 February 2011, MW 6.0 June 2011, and MW 5.9 December 2011.
confirmed that the result is not very sensitive to the considered varia- The time axes for each plot in Fig. 12 starts from the earthquake origin
tions. Linear elasticity is assumed for volcanic rocks, and the elastic time. It is expected that the simulation of an individual event would be
shear moduli, GS, are directly derived from the equation GS = ρVS2 and affected by the uncertainty associated with the estimated input motion,
Eq. (2), assuming the mass density of rock as ρ = 2.4 Mg / m3 . Table 4 because the input motion is obtained from the surface motion recorded
summarises the material parameters used for simulations. at LPCC via deconvolution analyses. Nonetheless, the simulated mo-
The periodic boundary condition (also known as the wrap-around tions overall agree well with observed motions, especially considering
boundary condition) is imposed on the left and right hand side of the the simplicity of the model and the aforementioned uncertainties. Since
model, to simulate the laterally infinite halfspace. An absorbing we obtain the input motions from the recorded motions at LPCC, the
boundary condition at the base of the model (a depth of 120 m ), is differences in arrival time between recorded and simulated motions
modelled with standard viscous dashpots [23]. The input motions for originates from the difference in arrival times at HVSC and LPCC as a
the analyses, estimated based on the procedure described in the next result of different path effects.
section, are applied as equivalent nodal forces at the model base,
computed by the following equation: 6.2. Comparison of acceleration response spectra
˙ E
F = ρVS uA (4)
Fig. 13 shows the comparison of recorded and simulated accelera-
where AE is the tributary area of the two elements that connect to the tion response spectra for the same four events as Fig. 12. The com-
base nodes, u̇ is the particle velocity of the incident waves, and ρ and VS parison is also satisfactory for the response spectra, in which the si-
are as previously defined. mulations have similar spectral shapes to the observations; however it
is clear in Fig. 13(d) that the simulated response spectra of MW 5.9
5.2. Determination of input ground motions December 2011 event is significantly overestimated in the entire period
range. This is most likely associated with the aforementioned un-
In the absence of a borehole station at the bedrock level, selecting certainly in the amplitude of estimated input motions, because the site
appropriate input motions is one of the most challenging tasks for va- response is expected to affect the surface ground motion only for a
lidation of the site response analyses. In such cases, estimating the input certain period range (i.e. ground motions at sufficiently long periods

Table 4
Summary of material parameters used for the analyses.

Material type Shear wave velocity Mass density Poisson ratio Friction angle Cohesion intercept Thickness
VS [m/ s ] ρ [Mg / m3 ] ν ϕ[○] c [kPa] t [m]

Soil (PDMY) Eq. (2) 1.8 0.25 36 30 30


Rock (Linear Elastic) 800 2.4 0.25 – – 20
Rock (Linear Elastic) 1500 2.4 0.25 – – –

351
S. Jeong, B.A. Bradley Soil Dynamics and Earthquake Engineering 100 (2017) 345–356

Fig. 12. Comparison of acceleration time series for events: MW7.1 04/09/2010, MW6.2 22/02/2011, MW6.0 13/06/2011(b), MW5.9 and 23/12/2011(b).

would not be affected by the site response). If the input motions are of Heathcote Valley and the study by Bard and Bouchon [3], it seems
properly estimated, the simulated and observed response spectra are likely that the Rayleigh waves of various vibration frequencies caused
expected to be very close to each other at long periods, which was not by the smoothly varying sediment thickness would effectively have
the case for the December 2011 events and a few others. filled up the troughs between the harmonics in the spectral ratio.

6.4. Residuals of acceleration response spectra


6.3. HVSC/LPCC spectral ratios
As discussed previously, the validation of the site response simula-
The 1D site effects are often represented by the transfer function, tion is complicated because of the inherent uncertainties involved in the
the spectral ratio of the surface to the rock outcrop motion [21]. As we estimated input motions. However, it is expected that this limitation
do not have measured rock outcrop motions or borehole motions in this can be mitigated with a sufficient number of recorded events. Fig. 15
study, we instead used the HVSC/LPCC spectral ratio, defined as the shows the residual of the response spectral acceleration, defined as:
ratio of smoothed Fourier spectral amplitudes [20] of the HVSC and
r = log(SAObserved ) − log(SASimulation ) (5)
LPCC motions, to compare the observed and simulated site amplifica-
tion at HVSC. Fig. 14 show the overall good agreement between the Residuals for individual events are plotted with grey lines. The mean of
observed and simulated spectral ratios, especially for the frequency and the residuals for all events considered is plotted with black solid line,
the amplitude of the fundamental mode response. Note that the am- and the dashed lines represent the 68% confidence interval (i.e. 16th
plitude of the first harmonic mode in the model is also reasonably close and 84th percentile).
to the observed spectral ratio. However, there seems to be a significant The figure shows that, on average, the observed and simulated re-
difference in spectral ratio amplitude in between the fundamental and sponse spectra agree very well with each other in long vibration periods
the first harmonic mode. (i.e. T > 2 sec ), albeit with a large uncertainty. We emphasise that the
For incident SV-waves, Bard and Bouchon [3] showed that the first large uncertainty at long periods is most likely the result of input mo-
two Rayleigh modes are excited as soon as the frequency of excitation tion characterisation uncertainty as discussed earlier, and this un-
exceeds the S-waves fundamental frequency of the corresponding flat certainty indeed demonstrates the importance of running simulations
layer. As the valley has continuously varying depth, this means that the for a large number of earthquake events. Although there is relatively
Rayleigh waves generated at different part of the basin edge would have large variability, the mean residual overall lies closer to zero compared
slightly different frequency content. The observed spectral ratios in with the empirical prediction shown in Fig. 4 and this shows the benefit
Fig. 14 do not show the clear troughs between the harmonics, which is a of explicitly modelling the near surface site response over the use of
characteristic of 1D transfer functions. Considering the basin geometry empirical prediction.

352
S. Jeong, B.A. Bradley Soil Dynamics and Earthquake Engineering 100 (2017) 345–356

Fig. 13. Comparison of acceleration response spectra for the four main earthquake events: MW7.1 04/09/2010, MW6.2 22/02/2011, MW6.0 13/06/2011(b), and MW5.9 23/12/2011(b).

Fig. 15. Median and the 68% confidence interval of the spectral acceleration residuals for
Fig. 14. Comparison of simulated and recorded HVSC/LPCC spectral ratios. Spectral ra- all considered events. Residuals for each of the 10 individual events are plotted with grey
tios of the individual events are plotted with grey lines. lines.

353
S. Jeong, B.A. Bradley Soil Dynamics and Earthquake Engineering 100 (2017) 345–356

Fig. 16. Shear stress-strain curves of soil elements for events: 04/09/2010 and 22/02/2011.

Despite the favourable overall bias seen in Fig. 15, there is still a relatively wide range of wavelengths. This is likely due to the pressure
general tendency for the simulation to underestimate the response for dependency of the Port Hills loess, which results in very low elastic
T < 0.5 s (i.e. a positive residual), with the simulated spectral accel- moduli (and strength) near surface; if the dynamic soil constitutive
erations on average 40% less than the observations. This bias is likely behaviour is relatively independent of the depth, shear strains would
due to the number of limitations of the 1D site response model. As have been much more concentrated near the soil-rock interface, which
discussed previously, the complex 2D/3D site effects caused by the in turn would reduce the high frequency amplification.
inclined soil-rock interface near the basin edge likely would have It is well understood that, as the ground motion becomes intense,
contributed to the bias. the amount of dissipated energy per cycle becomes larger. Fig. 16 de-
monstrates that the energy is dissipated much more quickly at HVSC
during the February 2011 event, compared with the September 2010
7. Role of soil constitutive behaviour on the observed strong event. The increased dissipation may significantly reduce the amplitude
motions of ground accelerations near the ground surface for strong earthquake
events.
7.1. Nonlinear hysteretic response of the soils
7.2. Nonlinear vs. equivalent linear analysis
Natural soils are known to exhibit nonlinear hysteretic response
when subjected to strong shaking that affects the characteristics of the
The equivalent linear approach [30] for the 1D site response ana-
ground motions recorded at the surface. Fig. 16 shows examples of the
lysis became very popular in 1970's and is still widely used in practice.
shear stress-strain curves of soil elements from simulations of the Sep-
We compared our pressure-dependent nonlinear model with the widely
tember 2010 and the February 2011 events. The simulation shows that
used equivalent linear approach, for the four recorded events: 04/09/
the September event induces moderate strain up to γ = 0.06%. However,
2010, 22/02/2011, 13/06/2011(b), and 23/12/2011(b). Fig. 18 shows
the ground strain due to the February event was significant and reached
the comparison of acceleration time series, which demonstrates that
γ = 0.3% at a depth of z = 7.1 m .
both approaches yield comparable level of peak ground accelerations
Interestingly, the level of maximum strain from the simulation was
for the considered events.
relatively well distributed across the entire depth of the soil except
However, the HVSC/LPCC spectral ratios, shown in Fig. 19, illus-
where it is very shallow, as shown in Fig. 17, which suggests that the
trates that the equivalent linear approach significantly underestimates
effect of nonlinear behaviour would affect the ground motions in a
the high frequency response of the site, while the nonlinear model does
a much better job in predicting the high frequency response. This is a
well known problem of the equivalent linear approach, caused by the
assumption of time-invariant moduli and damping ratios that are
iteratively updated such that they correspond to the two-third of the
maximum shear strain; this approach tends to underestimate the small
amplitude high-frequency motions for the price of reasonable approx-
imation of the peak response.

8. Conclusions

Motivated by the severity of the recorded ground shaking at


Heathcote Valley during the 2010–2011 Canterbury earthquake se-
quence, we conducted a series of one-dimensional nonlinear site-spe-
Fig. 17. Maximum shear strain profiles of the soil elements for all considered events. cific response analyses based on detailed site characterisation studies to
Profiles for the four main events (04/09/2010, 22/02/2011, 13/06/2011(b), and 23/12/ investigate the role of the dynamic response of the near surface soils on
2011(b)) are plotted with thick lines. the observed ground motions.

354
S. Jeong, B.A. Bradley Soil Dynamics and Earthquake Engineering 100 (2017) 345–356

Fig. 18. Comparison of acceleration time series from equivalent linear and nonlinear analyses for events: 04/09/2010, 22/02/2011, 13/06/2011(b), and 23/12/2011(b).

show that the dynamic response of the surficial soil layer overlying
volcanic rock at Heathcote Valley strongly amplified ground motions in
a wide band of frequencies. The residuals of spectral accelerations de-
monstrate that these simple 1D site response analyses perform sig-
nificantly better than the New Zealand-specific empirical ground mo-
tion model by Bradley [7]. Simulated and observed HVSC/LPCC Fourier
spectral ratios indicate that the fundamental site frequency at HVSC is
approximately f0 = 2.8 Hz . However, simulations still underestimate
the amplification at frequencies higher than the fundamental site fre-
quency, especially in between the harmonics of the 1D model Fourier
spectral ratio. Considering the geometry of the sedimentary basin at
Heathcote Valley, it is likely that the recorded ground motions at HVSC
would have been further amplified by the surface waves generated at
the valley. However this cannot be confirmed by the current 1D site
response model and requires a more complex model.
Comparison between the nonlinear and the equivalent linear model
Fig. 19. Effect of soil nonlinearity on the HVSC/LPCC Fourier spectral ratio. Fourier shows that although both approaches produce similar level of peak
amplitudes are smoothed using the method by Konno and Ohmachi [20] with the amplitude, the equivalent linear model significantly underestimates the
smoothing coefficient, b = 40. high frequency motions.

We compiled a number of site characterisation studies performed at


Heathcote Valley, which were used to synthesise a 3D geological model. Acknowledgements
The results of 15 sCPT show that the shear wave velocity of the soils at
Heathcote Valley can be effectively modelled by a simple power-law The high performance computing used in this research was provided
equation as function of depth. H/V spectral ratios were computed from through the New Zealand National e-Science Infrastructure, NeSI, merit
measured ambient ground vibrations at 26 sites within the valley, and allocation. Financial support for this work from the Royal Society of
fundamental site frequencies obtained from H/V spectral ratios agree New Zealand Rutherford Discovery Fellowship, New Zealand
well with sCPT. Earthquake Commission and QuakeCoRE: The Centre for Earthquake
Overall, 1D numerical analyses agree well with observations and Resilience is greatly appreciated. This is QuakeCoRE publication

355
S. Jeong, B.A. Bradley Soil Dynamics and Earthquake Engineering 100 (2017) 345–356

number 0085. The ambient vibration data for H/V spectral ratios were [14] Hardin BO, Drnevich VP. Shear modulus and damping in soils: design equations and
curves. Soil Mech Found Div 1972;98:667–92.
collected in collaboration with Liam M. Wotherspoon (University of [15] Hashash YM, Dashti S, Romero MI, Ghayoomi M, Musgrove M. Evaluation of 1-D
Auckland) and Clint M. Wood (University of Arkansas), whose assis- seismic site response modeling of sand using centrifuge experiments. Soil Dyn
tance is greatly appreciated. Earthq Eng 2015;78:19–31. http://dx.doi.org/10.1016/j.soildyn.2015.07.
003%5Cn. [〈http:// linkinghub.elsevier.com/retrieve/pii/S0267726115001761〉].
[16] Haskell NA. The dispersion of surface waves on multilayered media. Bull Seismol
References Soc Am 1953;43(January (1)):17–34. [〈http:// www.bssaonline.org/content/43/1/
17.short〉].
[17] Johnson DH, Dudgeon DE. Array signal processing: concepts and techniques. 1st ed.
[1] Al Atik L, Sitar N. Seismic earth pressures on cantilever retaining structures. J
Prentice Hall; 1993.
Geotech Geoenviron Eng 2010;136(10):1324–33.
[18] Kaiser A, Holden C, Beavan J, Beetham D, Benites R, Celentano A, et al. The M w 6.
[2] Anderson JG, Brune JN. Probabilistic seismic hazard analysis without the ergodic
2 Christchurch earthquake of February 2011: preliminary report. NZ J Geol
assumption. Seismol Res Lett 1999;70(Jaunary (1)):19–28. [〈http:// srl.
Geophys 2012;55(1):67–90.
geoscienceworld.org/content/70/1/19.extract〉].
[19] Kondner RL. Hyperbolic stress-strain response: cohesive soils. J Soil Mech Found
[3] Bard P-Y, Bouchon M. The seismic response of sediment-filled valleys. Part 2. The
Div, ASCE 1963;89:115–43.
case of incident P and SV waves. Bull Seismol Soc Am 1980;70(5):1921–41.
[20] Konno K, Ohmachi T. Ground-motion characteristics estimated from spectral ratio
[〈http:// www.bssaonline.org/content/70/5/1921.abstract〉].
between horizontal and vertical components of microtremor. Bull Seismol Soc Am
[4] Beavan J, Motagh M, Fielding EJ, Donnelly N, Collett D. Fault slip models of the
1998;88(February (1)):228–41. [〈http:// bssa.geoscienceworld.org/content/88/1/
2010–2011 Canterbury, New Zealand, earthquakes from geodetic data and ob-
228.abstract〉].
servations of postseismic ground deformation. NZ J Geol Geophys
[21] Kramer SL. Geotech Earthq Eng 1996;6.
2012;55(3):207–21.
[22] Lachet C, Bard P-Y. Numerical possibilities and theoretical investigations on the and
[5] Bradley BA. Ground motions observed in the Darfield and Christchurch earthquakes
limitations of Nakamura's Technique. J Phys Earth 1994;42:377–97.
and the importance of local site response effects. NZ J Geol Geophys
[23] Lysmer J, Kuhlemeyer RL. Finite dynamic model for infinite media. J Eng Mech Div
2012;55(3):279–86. http://dx.doi.org/10.1080/00288306.2012.674049%5Cn.
1969;95(4):859–78. [〈http:// cedb.asce.org/cgi/WWWdisplay.cgi?16353〉].
[6] Bradley BA. A New Zealand-specific pseudospectral acceleration ground-motion
[24] Massey CI, Della-Pasqua FN, Lukovic B, Yetton MD, Archibald GC, Ries W. 2014a.
prediction equation for active shallow crustal earthquakes based on foreign models.
Canterbury earthquakes 2010/11 Port Hills slope stability: Risk assessment for
Bull Seismol Soc Am 2013;103(3):1801–22. [〈http:// earthquakespectra.org/doi/
Clifton Terrace. Tech. Rep. 2014/76, GNS Science.
10.1193/053013EQS137M〉].
[25] Massey CI, McSaveney MJ, Taig T, Richards L, Litchfield NJ, Rhoades DA, et al.
[7] Bradley BA. Systematic ground motion observations in the canterbury earthquakes
Determining rockfall risk in Christchurch using rockfalls triggered by the
and region-specific non-ergodic empirical ground motion modeling. Earthq Spectra
2010–2011 canterbury earthquake sequence. Earthq Spectra 2014;30(1):155–81.
2015;31(August (3)):1735–61.
[26] Matheron G. Principles of geostatistics. Econ Geol 1963;58(December (8)):1246–66.
[8] Bradley BA, Cubrinovski M. Near-source strong ground motions observed in the 22
[〈http:// econgeol.geoscienceworld.org/content/58/8/1246.abstract〉].
February 2011 Christchurch earthquake. Seismol Res Lett 2011;82(November
[27] Mazzoni S, McKenna F, Scott MH, Fenves GL. OpenSees command language manual
(6)):853–65. [〈http:// srl.geoscienceworld.org/cgi/doi/10.1785/gssrl.82.6.853〉].
2007 〈http:// opensees.berkeley.edu/OpenSees/manuals/usermanual/
[9] Bradley BA, Quigley MC, Van Dissen RJ, Litchfield NJ. Ground motion and seismic
OpenSeesCommandLanguageManual.pdf〉.
source aspects of the canterbury earthquake sequence. Earthq Spectra
[28] Nakamura Y. A method for dynamic characteristics estimation of subsurface using
2014;30(1):1–15.
microtremor on the ground surface. Tech. rep. 1989.
[10] Brown L, Weeber J. Geology of the Christchurch Urban Area. Institute of
[29] Prevost JH. A simple plasticity theory for frictional cohesionless soils. Soil Dyn
Geological & Nuclear Science Ltd. 1992.
Earthq Eng 1985;4(1):9–17.
[11] Cubrinovski M, Bradley B, Wotherspoon L, Green R, Bray J, Wood C, et al.
[30] Schnabel PB, Lysmer J, Seed HB. Shake: a computer program for earthquake re-
Geotechnical aspects of the 22 February 2011 Christchurch earthquake. Bull NZ Soc
sponse analysis of horizontally layered sites 1972.
Earthq Eng 2011;44(4):205–26.
[31] Wathelet M. An improved neighborhood algorithm: parameter conditions and dy-
[12] Fry B, Benites R, Reyners M, Holden C, Kaiser A, Bannister S, Gerstenberger M,
namic scaling. Geophys Res Lett 2008;35(9):1–5.
Williams C, Ristau J, Beavan J. Strong shaking in recent New Zealandearthquakes.
[32] Wood CM, Cox BR, Wotherspoon LM, Green RA. Dynamic site characterization of
Eos 92 (41), 2011, 349 351.
Christchurch strong motion stations. Bull NZ Soc Earthq Eng 2011;44(4):195–204.
[13] Gledhill K, Ristau J, Reyners M, Fry B, Holden C. The darfield (canterbury) earth-
[33] Yang Z, Elgamal A, Parra E. Computational model for cyclic mobility and associated
quake of september 2010: preliminary seismological report. Bull NZ Soc Earthq Eng
shear deformation. J Geotech Geoenviron Eng 2003;129(December (12)):1119–27.
2010;43(4):215–21.

356

Vous aimerez peut-être aussi