Sixth International Conference on

Recent Advances in Geotechnical Earthquake Engineering and Soil Dynamics

August 1 – 6, 2016, IIT Roorkee Extension Centre, 20 Knowledge Park II, Greater Noida, India

Surface Wave Data Uncertainty and Its Consequence on Seismic Design Ground Motion
Narayan Roy Ravi S. Jakka H.R. Wason
Jadavpur University IIT Roorkee IIT Roorkee
Kolkata-700032, India Roorkee, India Roorkee, India

ABSTRACT

Surface wave methods which are increasingly being used to characterize a site may suffer from several uncertainties. Data
measurement uncertainty is one of the major uncertainty, which can lead to an inaccurate estimation of shear wave velocity profiles
and this may further affect the calculation of design ground motion. Data measurement uncertainty arises from the surrounding noises
which contaminate the recorded signals and make the record uncertain. To measure this variation, repetition tests of active MASW
have been performed in and around Roorkee city, India. The data of each repetition has been analyzed separately and a statistical
study has been performed to generate the standard deviation of coefficient of variations (COV) for all the sites After knowing the
maximum data variation, a numerical study is performed to assess the consequence of data uncertainty on seismic design ground
motion. Inversion is carried out with neighbourhood algorithm and profiles are generated below the maximum data variation. Now
few profiles are selected covering the whole range of data uncertainty bound and these profiles are subjected to 1D ground response
analysis using SHAKE2000 software. Finally, the results are presented in the form of variations in amplification spectra and response
spectra.

INTRODUCTION

Multichannel analysis of surface wave (MASW) method velocity (Bodet et al., 2009) may arise due to the testing set-up
which basically uses the dispersion properties of surface wave configuration, type of source and subsurface soil profiles. Data
is largely used to determine the soil stiffness variation with measurement uncertainty may arise from the noise in the
depth (Park et al., 1999; Miller et al., 1999). The dispersive recorded signals. This uncertainty plays an important role in
nature of Rayleigh wave allows different frequency wave to the resulting shear wave velocity profiles which may lead to
travel at different velocity in a vertically layered medium and different seismic site responses (Jakka et al., 2014). But very
penetrates up to different subsurface depths. Generally, few research so far has been carried out to quantify the
Rayleigh waves with large wavelength, i.e. low frequency, maximum data variation of surface wave tests. In this study,
penetrate higher depth and reflects high phase velocity in the an attempt has been made to quantify the maximum data
dispersion spectra. And the waves with low wavelengths, i.e. variation conducting several MASW tests in and around
high frequency waves, penetrates shallow depth reflecting the Roorkee city with several repetitions. Nine different sites are
properties of shallow sub surface (Fig. 1). selected to carry out extensive active MASW testing with
multiple repetition keeping the configuration same. Once we
Mainly two types of surface wave methods are used in site calculate the maximum data variation bound, a numerical
characterization. In active-source surface wave tests, waves study is performed to assess the effect of data uncertainty on
are generated using a seismic source (Stokoe et al., 1994; Park seismic design ground motion. A site specific study is also
et al., 1999), while continuous vibrations of earth or presented to validate our numerical study
microtremors are used in passive-source tests (Horike, 1985;
Louie, 2001). The uncertainty in surface wave methods EXPERIMENTAL STUDY
broadly classified into model based uncertainty and data
measurement uncertainty. In model based uncertainty, near- Sites for conducting active MASW testing are selected inside
field effect is an area of great concern and some studies have IIT Roorkee campus as well as surrounding the Roorkee city.
been carried out to mitigate this uncertainty. Near-field effects To measure the maximum data variation, repetitions of test
which result in underestimation of Rayleigh wave phase have been performed keeping the configuration same. For this

Paper No. abs120 1

 later is used to generate the different data bounds.

Fig. 1. Geometrical dispersion of surface wave in a layered

media (Rix, 1988)

purpose, 35-40 repetitions of test have been conducted at Fig. 3. Receivers array at a testing site
every site. Active source test is conducted with McSeis-SXW
24 channel MASW set up. Figure 2 shows the recording Figure 4(a) presents the plot of all the COV variations of the
device used in the study. A sample recording array has been considered 9 sites. The plot clearly shows the variations of
shown in Fig. 3. A data length of 2 sec is selected to record the data are quite different and relatively higher for lower
data and number of data points are 2048. Sampling frequency frequencies than the COVs of higher frequencies. After
was kept at 1000 Hz so as to get good data quality at higher accumulating all the COV variations, the bounds are generated
frequencies also. A total of 24 receivers are used to capture the by the statistical analysis and have been presented in Fig. 4(b).
signal with a source-to-first receiver distance and receiver-to- Mean curve, upper bound curve and lower bound curve have
receiver distance as 2m at most of the sites. For each shot, the been generated, which are nothing but the mean, +1 standard
recorded trace is taken out for further processing of the data to deviation and -1 standard deviation of the COV variation plot.
generate the dispersion curve. Here, we can observe that there are two distinct regions in this
plot which define the variations of the generated three bounds.
We can divide the plot (Fig. 1b) into two separate parts: (i) 5
to ≤ 20 Hz, and (ii) >20 to 40 Hz. Part-I (5 to ≤ 20 Hz) shows
a sudden increase of the COV values and this gradually goes
on increasing with the reduction of frequencies.

Part-II (>20 to 40 Hz) is associated with lower variation of

COVs which is nearly constant irrespective of the frequencies.
Once we generated the data variation bounds from the
extensive field experiments, a numerical and site specific
study is performed to study the effect of data uncertainty on
design ground motion.

INVERSION WITH NEIGHBOURHOOD ALGORITHM

The neighborhood algorithm is used to carry out the inversion.

A pseudo-random samples set is generated after defining the
Fig. 2. Recording device used in data analysis variation of each model parameters (thickness and shear wave
velocity of each layer) in the parameter space. The dispersion
Dispersion curve is generated separately for each shot. From curves are developed from this set of samples by using the
the 37- 40 repetitions of data, 37- 40 different dispersion forward model algorithm which was originally proposed by
curves have been generated at each site to find out the Thomson (1950) and Haskell (1953), and later modified by
variations in the measured data. Then, a statistical analysis is Dunkin (1965) and Knopoff (1964). The forward algorithm
performed to generate the mean and standard deviation of the uses the fundamental mode of Rayleigh wave propagation by
data set, and finally COV plot with frequency is generated to modelling soil column as a stack of horizontal and
get a clear insight about the data variation at every site, which homogeneous layers. Once the theoretical dispersion curve is

Paper No. abs120 2

generated from the random samples given by the inversion. Once the target dispersion curve is generated, the
neighborhood algorithm, the misfit value is calculated. When error associated with each point is assigned from the upper
the experimental dispersion curves are provided with an bound curve of the estimated error from field testing. This
generated dispersion curve along with the associated error is
now the target dispersion curve (Fig. 5b) and it will allow the
inversion limit to consider within this error bound. Now the
inversion is run and all the profiles are generated within this
defined error bound. Now 60 profiles are selected covering the
whole range of error bound. To select these profiles, a proper
consistency is maintained while to select the velocity profiles.
Figure 6(a) shows all the selected velocity profiles after
inversion with Neighbourhood algorithm. Figure 6(b) shows
the dispersion curve of the same selected velocity profiles.

These selected velocity profiles are then subjected to 1D

seismic ground response analysis using SHAKE2000
software. The acceleration time history of January 26, 2001
Bhuj earthquake of magnitude 7.6 recorded at Ahmedabad
(Fig. 7a) has been used as input base motion in SHAKE
analysis. The response spectra of the Bhuj motion is shown in
Fig. 7(b).

Fig. 4(a). Plot of COV variation of all the nine sites, and (b)
Generated bounds of the variation after statistical analysis

uncertainty estimate, the misfit is calculated as follows: (a)

Misfit = (1)

where Xti is the theoretical and Xei is the experimental phase

velocity of the calculated curve at frequency fi, σi is the
uncertainty of the frequency samples, and n is the number of
frequency samples considered in the dispersion curve. The
detailed description about the procedure can be found out in
the literature (Wathelet et al., 2004).

(b)
NUMERICAL INVESTIGATION

A synthetic profile with three layers plus half-space is selected

for the numerical simulation. Figure 5(a) shows the profile Fig. 5(a) Profile considered for the numerical simulation, and
considered for the numerical simulation. Then, the dispersion (b) Generated target dispersion curve with associated error
curve of the considered profile is generated for the purpose of bound

Paper No. abs120 3

 (a)

The peak acceleration of the time history is 0.104g. G/Gmax

and Damping curves are taken from SHAKE data base for
sand avg. curve for all three layers as we are not considering
the effect of type of soil, rather our main aim is to find out the
effect of surface wave data measurement uncertainty on
seismic site response analysis. The results are presented in the
form of amplification spectra and response spectra. COV plots
for both amplification spectra and response spectra have also
been prepared to get more information about the relative
differences of each selected profiles. Figure 5 presents the
comparison of the amplification spectra (Fig. 8a) and the COV
variation with frequency (Fig. 8b) of those selected profiles.
(b)

(a)

Fig. 7(a). Bhuj earthquake motion used in the ground

response analysis, and (b) Response spectra of the same
motion at 5% damping

(b) The plots show considerable differences in terms of their peak

frequency and peak amplification. The curve marked in red
color in the plot is the mean curve of all the amplification
spectra. The coefficient of variation of the amplification
spectra of 60 profiles shows quite high variation below 6Hz,
and above 6Hz frequency the plot shows a little lower value of
COV. So, the variation of first peak in the amplification plot is
quite significant which can cause a considerable effect on the
calculated design ground motion and can affect the design of
the engineering structures. Figure 9(a) presents the
comparison of the response spectra of those selected 60
profiles and Fig. 9(b) is the COV plot of those response
spectra. Response spectra also show considerable differences.
Fig. 6(a). Selected 60 velocity profiles after the inversion with The COV plot shows a maximum value of 14% at peak
Neighbourhood algorithm, and (b) Dispersion curve of those spectral acceleration which occurs nearly at 4Hz frequency.
selected velocity profiles

Paper No. abs120 4

 (a) CONCLUSIONS

Extensive field experiments have been conducted to quantify

the data measurement uncertainty with different noisy
environment. Then, the consequence of this data measurement
uncertainty is studied to assess the effect on different ground
motion parameters. A large number of field tests have been
performed to assess the maximum data variation bound of the
measured dispersion curves with several repetitions. The data
variation shows two distinct regions in the Frequency-COV
plot. Below 20Hz frequency, the data scatter is very high and
an almost linear increase of COV is observed, but above 20Hz
frequency data shows a nearly constant value of COV. A
(b) numerical study is performed to assess the consequence of
data uncertainty on 1D seismic ground response analysis.
Finally, the variation in amplification spectra and response
spectra are quantified in terms of coefficient of variation.

REFERENCES

1. Bodet, L., Abraham, O. and Clorennec, D. (2009),

“Near-offset effects on Rayleigh-wave dispersion
measurements: physical modelling”, J. Applied
Geophys., 68, 95-103.
2. Dunkin, J.W. (1965), “Computation of modal
solution in layered elastic media at high frequencies”,
Fig. 8 (a). Comparison of amplification spectra of the 60 Bull. Seismol. Soc. Am., 55, 335-358.
selected profiles, and (b) COV plot of those selected profiles
3. Haskell, N.A. (1953), “The dispersion of surface
waves on multilayered media”, Bull. Seismol. Soc.
(a) Am., 43(1), 17-34.
4. Horike, M. (1985), “Inversion of phase velocity of
long-period micro tremors to the S-wave velocity
structure down to the basement in urbanized areas”,
J. Phys. Earth, 33, 59-96.
5. Jakka, R.S., Roy, N. and Wason H.R. (2014),
“Implications of surface wave data measurement
uncertainty on seismic ground response analysis”,
Soil Dyn. Earthq. Engg., 61-62, 239–245.
6. Knopoff, L. (1964), “A matrix method for elastic
wave problems”, Bull. Seismol. Soc. Am., 54, 431-
438.
(b) 7. Louie, J.N. (2001), “Faster, better: shear-wave
velocity to 100 meters depth from refraction
microtremor arrays”, Bull. Seismol. Soc. Am., 91(2),
347-364.
8. Miller, R.D., Xia, J., Park, C.B. and Ivanov, J.
(1999), “Using MASW to map bedrock in Olathe”,
Kansas Soc. Explor. Geophys., 433-436.
9. Park, C.B., Miller, R.D. and Xia, J. (1999), “Multi-
channel analysis of surface waves”, Geophysics,
Fig. 9(a). Comparison of response spectra of the 60 selected 64(3), 800-808.
profiles, and (b) COV plot of those response spectra of
10. Stokoe II, K.H., Wright, S.G., Bay, J.A. and Roesset,
selected profiles

Paper No. abs120 5

 J.M. (1994), “Characterization of geotechnical sites
by SASW” In: RD Woods (ed). Oxford & IBH
Publishing, New Delhi, 15-25, 1994.
11. Thomson, W.T. (1950), “Transmission of elastic
waves through a stratified solid medium”, J. Appl.
Phys., 21(1), 89-93.
12. Wathelet, M., Jongmans, D. and Ohrnberger, M.
(2004), “Surface-wave inversion using a direct search
algorithm and its application to ambient vibration
measurements”, Near Surface Geophys., 2(4), 211-
22.

Paper No. abs120 6