Site Response Analysis Including Earthquake Input Ground Motion and Soil Dynamic Properties Variability
Site Response Analysis Including Earthquake Input Ground Motion and Soil Dynamic Properties Variability
Site Response Analysis Including Earthquake Input Ground Motion and Soil Dynamic Properties Variability
F. Lopez-Caballero
Ecole Centrale de Paris, France
C. Gelis
Institut de Radioprotection et de Surete Nucleaire, France
J. Regnier
CETE-Mediterrannee, France
L.F. Bonilla
Universite Paris Est - IFSTTAR, France
SUMMARY
Earthquake engineering analyses often concentrate on the variability of soil properties when computing site-
specific ground motion. Conversely, the earthquake source is modelled as a simple modulated noise that fits a
target response spectrum. In this study we propose to evaluate the site-specific ground-motion variability
including the overall effect of the input ground motion as well as the dynamic parameters of the soil column. In
this case, we show that including realistic input ground motions may strongly control the total variability of soil
response and computed ground motion. This is very important when assessing site-specific probabilistic seismic
hazard studies where rock uniform hazard spectra (UHS) estimates are usually computed. Such studies have an
inherent uncertainty related to all possible scenarios that contribute to the seismic hazard, which should be taken
into account.
INTRODUCTION
Local site effects have long been recognized as an important factor contributing to variations in strong
ground motions. Their study is one of the most important goals of earthquake engineering. Seismic
hazard evaluations are calculated over broad geographical areas; however, as more ground motion data
are collected, the local geology condition is emerging as one of the dominant factors controlling the
variation in ground motion and determination of the site-specific seismic hazard for a given
earthquake.
Site response analyses are usually performed in a deterministic way. This means that soil elastic and
dynamic properties remain constant through the analysis. At most, several input ground motions are
used when computing nonlinear soil computations to assess the variability of the obtained time
histories. This is because the soil dynamic properties play an important role, especially when the
material enters into a nonlinear regime. Recorded data generally show high frequency de-amplification
and a shift of the soil resonant frequency to lower frequencies.
In this paper, we study the effect of input motion and soil parameters variability on numerical site
response evaluation. In particular, we show the effect of nonlinear soil response on the dispersion of
computed ground motion.
In this work, the relevant input parameters are the shear wave velocity profile and the shear modulus
reduction curves. The studied response parameters concern both the acceleration level and spectral
response at free field. According to Griffiths and Fenton (2001), Popescu et al. (2006) among others,
there is no clear evidence pointing to any specific model for the probability density function (pdf) of
soil properties. However, they proposed to use non-negative functions as Beta, Gamma or lognormal
for many material properties. The probabilistic shear-wave velocity profiles generated for Latin
hypercube simulations are based on a baseline shear-wave velocity profile. The baseline shear-wave
velocity profile used in this study is based on the model proposed for the IWTH08 KiK-net station
(see section 3) and it is assumed to be characterized statistically by a lognormal distribution at any
given depth. The baseline shear-wave velocity profile defines the mean values of Vs and in order to
take into account the uncertainty several values of the coefficient of variation (CV) varying from 10 to
30% are used. Figure 1 displays one of the obtained uncertainty shear-wave profile for CVVs=20%. In
this figure, the median, the ± one standard deviation and the range of Vs profiles determined by Latin
hypercube sampling are showed. These summarized curves involve 100 sample computations. The
range of Vs profiles represents the limits of the probabilistic profiles. It is important to note that the
median response obtained is in agreement with the baseline shear-wave velocity profile, meaning that
the statistical model converge at least at first order.
Concerning the probabilistic shear modulus degradation (i.e. G-γ) curves, according to the used
backbone stress-strain model (i.e. the hyperbolic model) the nonlinear relation is controlled by the γref
parameter following the hyperbolic model (Konder and Zelasko, 1963). Thus the randomness in the
dynamic properties of the soil is introduced through this parameter. It is assumed that γref is
characterized statistically by a lognormal distribution with a coefficient of variation (CV) varying
from 20 to 30%. Figure 2a shows the mean, the ± one standard deviation and the range of G/Gmax
curves determined by LHS. The choice of simulation (i.e. only one random parameter) implies a
variation of the CV of G/Gmax for each γ level. It means that for lower γ values the randomness of these
curves is due principally to Vs dispersion and for higher strain levels the randomness is a combination
of both Vs and γ dispersion.
a) b)
In the case of strong motion propagating on soft soils, the shear strain becomes significant and
nonlinear soil behavior may take place (Iai et al., 1995; Ishihara, 1996). In this study, we adopted the
nonlinear soil rheology proposed by Towhata and Ishihara (1985) and Iai et al. (1990). This is a plane
strain model that is relatively easy to implement and needs only the angle of friction and the cohesion
when pore pressure is not taken into account, which is the case in this paper. The material strength is
computed following a Coulomb’s criterion, and the stress-strain relation follows the hyperbolic model.
Bonilla et al. (2005) modified the nonlinear constitutive model of Towhata and Ishihara (1985) and Iai
et al. (1990) so that hysteresis cycles are assured by applying the Generalized Masing Rules operator.
In order to take into account low strains damping (viscoelastic part) and hysterestic attenuation
(nonlinear part), we follow Assimaki et al. (2010) approach where the total energy dissipated in the
soil is equal to the sum of attenuation related to small shear strain damping modeled with the
technique of Liu and Archuleta (2006) and hysteretic damping accounted for through the nonlinear
constitutive model.
We calculated for each site the empirical linear borehole response (namely the Fourier spectral ratio
between the surface and the borehole). We use earthquake data with PGA at depth lower than 10 gals.
For each recording at each site we select the signal (beginning of the P-waves arrivals until the end of
the coda wave) along with the pre-event noise. We calculate the Fourier transform of the recordings at
depth and at the surface for the 3 components of motion and compute the quadratic mean of the
horizontal spectral ratio of the surface to depth spectrum (Régnier et al, 2012). Then for each site the
mean and 95% confidence limit of the borehole spectral ratio were computed. We choose a given site
that accomplishes the following criteria:
1. The empirical linear site response is close to the 1D configuration.
2. The linear site response variability is weak (the variability correspond to the inter-events
variability)
3. The station has also recorded strong events and has non-negligible nonlinear soil behavior.
The station IWTH08 fulfils all the requirements previously defined. As displayed in figure 3, this
station is characterized by a strong amplification at 2.9 Hz associated to the velocity contrast located at
50 m depth and having broadband amplification from 6Hz up to 12Hz. The H/V curve and the
empirical borehole site response indicate the same first peak and the same trend for the broadband
amplification which suggest that the down going wave field did not pollute significantly the borehole
recording. Furthermore, the numerical simulation is in very good agreement with the empirical
evaluation indicating that the shear wave velocity profile at this station is valid and the site
configuration is 1D. Finally, the empirical linear site response has a very low inter-event variability,
which was already highlighted by Baise et al (2011). Three events with surface PGA higher than 100
gals were recorded at this station. The comparison between linear and nonlinear site response showed
in figure 3 indicates that there is significant difference (at 95% probability level) between linear and
non-linear site response evaluation suggesting that the soil behaves nonlinearly at this station.
Figure 3 : Left : shear wave velocity profile of the station IWTH08. The triangles indicate the location of the
stations used to compute the borehole site response . Middle : Comparison of site response curves calculated
empirically (earthquake) and numerically (Haskell-Thopsom). The red curve displays the numerical outcrop site
response. The green curve displays the numerical borehole site response curve. The gray area represents the 95%
confidence limit of the empirical linear borehole site response. The black plain and dashed curves represent the
mean and 95% confidence limit of the H/V curve at the surface. Right : Comparison of the site response curve
obtained from weak motion (PGA at depth lower than 20 gals) and the one from strong motion (PGA at surface
greater than 100 gals). The black curve represents the mean and the gray area the 95% confidence limit.
Figure 5: Location of the epicentre of the 20 greatest earthquakes recorded at the site IWTH17 according to their
surface PGA.
In order to have more signals that increase the non linear behavior of the studied soil, 17 earthquake
records proposed by Iervolino and Cornell (2005) and Kayhan et al., (2011) are also used. The events
range in magnitude between 5.2 and 7.6 and the recordings are at site-to-source distances from 15 to
50km and dense-to-firm soil conditions (i.e. 360m/s < Vs 30m < 800m/s).
RESULTS
We compare the deterministic PGAs and the mean PGAs calculated with different coefficient of
variation associated to the random generation of soil profiles. The shear wave velocity of each layer is
first considered as random (figure 1) whereas the soil nonlinear properties are fixed. Figure 6 shows
PGA values at the surface ground motion as a function of the PGA input motion (PHA). When PHAs
are small, the ground motion is amplified in the soil column and computed PGA is predominantly
greater than PHA. When the input increases, PGA increases more slowly due to soil nonlinearity (i.e.
the soil damping increases affecting the higher frequencies). Above a PHA of 0.25g, PGA values at
surface are lower than those at depth; nonlinear soil behavior deamplifies the input motion. This
saturation effect is characteristic of nonlinear effects (e.g. Bonilla et al., 2011).
Moreover, deterministic PGAs at surface are most of the time greater than mean PGAs whatever the
coefficient of variation (figure 6). Additionally, higher coefficients of variation lead to lower PGAs at
surface as shown in figure 7. Figure 8 displays the PGA coefficients of variation at surface as a
function of PHA. When PHA increases, the coefficient of variation for PGA also increases and
becomes higher than the Vs profiles coefficients of variation.
Figure 6: Deterministic PGA (“det”) and mean PGA calculated with different coefficients of variation of the
shear wave velocity (10%, 20%, 30%) for different input motion (“T1_nh” from Iervolino et al., 2005; “PS1a2”
from Rayhan et al., 2010; “acc” recorded at station IWTH17)
Figure 7: Ratio of mean PGA and deterministic PGA as a function of PHA for the different coefficients of
variation of the shear wave velocity profiles and the different input motions.
Figure 8: PGA coefficients of variation as a function of PHA for the different coefficients of variation of the
shear wave velocity profiles and the different input motions.
Therefore, taking into account the shear waves velocity variability with constant nonlinear properties
leads to a decrease of the mean PGA at surface while increasing the coefficient of variation. Rathje et
al. (2010) obtained similar results with an equivalent-linear soil constitutive model.
As suggested by Idriss (2011), the presence of nonlinearity may be observed in real data by looking at
the PGA as a function of PGV/Vs30 (PGV: Peak Ground Velocity) that roughly approximates the
shear strain. We show in figure 9 the comparison between observed data at station IWTH08 and
computed values with real accelerograms. When PGV/Vs30 increases, PGA increases more slowly
and a saturation effect appears, showing the presence of nonlinearity.
We now consider the variability of shear wave velocity profile together with the soil nonlinear
properties (figures 1 and 2). Figures 10 and 11 shows the mean PGA related coefficient of variation as
a function of the PHA. Previous results are added for comparison. With the variability of both shear
waves velocity profiles and soil nonlinear properties, mean PGA decrease is more pronounced than
with the shear waves velocity variability only, whereas the coefficient of variation is higher.
Figure 10: Ratio of mean PGA and deterministic PGA as a function of PHA for 20% and 30% coefficient of
variation of the shear wave velocity profiles and the nonlinear properties for the different input motions.
Figure 11: PGA coefficients of variation as a function of PHA for 20% and 30% coefficient of variation of the
shear wave velocity profiles and the nonlinear properties for the different input motions.
CONCLUSIONS
This study shows that when adding variability to the soil profiles (shear waves velocity and soil
nonlinear properties) the PGA generally increases its variability while decreasing its mean value as
Rathje et al. (2010) previously noticed. This suggests that introduction of variability is not necessarily
conservative. Therefore, when soil variability is introduced, the choice of the coefficient of variation
must be carefully chosen and potentially assessed with data (i.e. Moss (2008) gives values of
coefficients of variation associated with different methods of Vs30) helping the determination of
realistic coefficients of variation to be used in earthquake response analyses.
AKCNOWLEDGEMENT
This research was partly funded by the 2012 GIS-RAP project entitled « Influence de la variabilité du milieu et
de la sollicitation sismique sur l’évaluation des effets de site ».
REFERENCES
Baise, L.G., 2011. Complex Site Response - Does One-Dimensional Site Response Work? in 4th IASPEI/IAEE
International Symposium: Effects of Surface Geology on Seismic Motion, August 23-26, 2011, University
of Santa Barbara, 2011.
Bonilla L. F., Archuleta J. R. and Lavallee D., 2005. Hysteretic and dilatant behaviour of cohesionless soils and
their effects on nonlinear site response : field data observation and modelling. Bull. Seism. Soc. Am., 95, 6,
2373-2395.
Bonilla, L., C. Gelis, and J. Regnier, The challenge of nonlinear site response: field data observations and
numerical simulations, in 4th IASPEI/IAEE International Symposium: Effects of Surface Geology on
Seismic Motion, August 23-26, 2011, University of Santa Barbara, 2011.
Griffiths D. V. and Fenton G. A. (2001) Bearing capacity of spatially random soil: the undrained clay Prandtl
problem revisited. Géotechnique 51:4, 351-359.
Helton J. C., Johnson J. D., Sallaberry C. J. and Storlie C. B. (2006) Survey of sampling-based methods for
uncertainty and sensitivity analysis. Reliability Engineering and System Safety 91:9-10, 1175-1209.
Iai S., Matsunaga Y. and Kameoka T., 1990. Strain space plasticity model for cyclic mobility, Report of the Port
and Harbour Research Institute, Vol.29, pp. 27 –56.
Idriss, E. 2011. Use of Vs30 to Represent Local Site Conditions, in 4th IASPEI/IAEE International Symposium:
Effects of Surface Geology on Seismic Motion, August 23-26, 2011, University of Santa Barbara, 2011.
Iervolino I. and Cornell C.A., 2005. Record Selection for Nonlinear Seismic Analysis of Structures, Earthquake
Spectra, Volume 21, No. 3, pages 685–713
Ishihara K., 1996. Soil Behaviour in Earthquake Geotechnics. Oxford engineering science series 46.Oxford
science publications.
Kayhan A.H., Korkmaz K.A., Irfanoglu A. (2011). Selecting and scaling real ground motion records using
harmony search algorithm. Soil Dynamics and Earthquake Engineering, 31, 941–953
Konder, R.L., and J.S. Zelasko (1963). A hyperbolic stress-strain formulation for sands, in Proc. of 2nd Pan
American Conference on Soil Mechanics and Foundation Engineering, Brazil, 289-324.
Koutsourelakis S., Prévost J. H. and Deodatis G. (2002) Risk Assessment of an interacting structure-soil system
due to liquefaction. Earthquake Engineering and Structural Dynamics 31:4, 851–879.
Liu P.C. and Archuleta J. R., 2006. Efficient modelling of Q for 3D numerical simulation of wave propagation.
Bull. Seism. Soc. Am. Vol. 96, No. 4A, pp. 1352-1358.
Moss R.E.S. (2008). Quantifying Measurement Uncertainty of Thirty-Meter Shear-Wave Velocity. Bulletin of
the Seismological Society of America, Vol. 98, No. 3, pp. 1399–1411, doi: 10.1785/0120070101
Popescu R., Prévost J. H., Deodatis G. and Chakrabortty P. (2006) Dynamics of nonlinear porous media with
applications to soil liquefaction. Soil Dynamics and Earthquake Engineering 26:6-7, 648-665.
Rathje E. M., Kottke A. R. and Trent W. L. (2010) Influence of Input Motion and Site Property Variabilities on
Seismic Site Response Analysis. Journal of Geotechnical and Geoenvironmental Engineering 136:4, 607-
619.
Régnier, J. Cadet, H., Bonilla, F., Bertand, E., Semblat, J.F. (2012) Quantifying the effects of soil non-linear
behavior on site response: Statistical analysis on KiK-net data, To be submitted.to BSSA
Towhata I. and K. Ishihara, 1985. Modeling soil behavior under principal axes rotation, paper presented at the
Fifth International Conference on Numerical Methods in Geomechanics, Nagoya, Japan. Pp 523–530.
Xu C., He H. S., Hu Y., Chang Y., Li X. and Bu R. (2005). Latin hypercube sampling and geostatistical
modeling of spatial uncertainty in a spatially explicit forest landscape model simulation. Ecological
Modelling 185:2-4,255-269.