Professional Documents
Culture Documents
Earthq Eng & Eng Vib (2011) 10: 313-323 DOI: 10.1007/s11803-011-0068-y
Abstract: The hanging wall effect is an important factor that impacts the characteristics of strong ground motions in
near-fault areas. Based on a residual analysis of ground motion parameters characterizing the hanging wall effect and in
recognition of the nature of the effect, many models have been developed. In this study, after a comprehensive analysis of two
existing models, a new model is proposed and used to model the hanging wall effect in horizontal peak ground acceleration
(PGAH) and spectral acceleration (SAH) at a period of 0.1 s in the Wenchuan earthquake. Finally, comparisons between the
modeling results of the hanging wall effect in the Wenchuan earthquake and the results predicted by using Abrahamson and
Silva’s NGA model (AS NGA) indicate that the AS NGA model predicts a much higher hanging wall effect than the model
developed in this paper. Furthermore, the AS NGA model predicts a large hanging wall effect even at great distances, while
the proposed model more accurately captures the trend of the effect.
Keywords: hanging wall effect; Wenchuan earthquake; attenuation relationship; NGA model
site. The hanging wall effect is simply a geometrically PGAN and PGAP (peak ground acceleration for
related effect revealed by both empirical and numerical geometric mean horizontal components, fault normal
examinations. Such an effect can also be eliminated (FN) components, and fault parallel (FP) components),
by defining a so called root-mean-square distance, by and corresponding acceleration response spectra SAH,
which weighted averages are applied to sites located SAN, SAP for 5% damping are derived by the same
on a hanging wall and a footwall (Wang et al., 2008). attenuation relationship. Note that even though many
However, since the primary objective of this paper is to attenuation relationships are available, only the simplest
examine and model the hanging wall effect exhibited ones are needed for this study; thus the one used in
in the Wenchuan earthquake, the normally used closest Abrahamson and Somerville (1996) is quoted. Then,
distance is used to make the hanging wall effect emerge. the specific attenuation relationship for the Wenchuan
Recordings within a near-fault distance of 100 km were earthquake considering the effects of site, geometric
used in this study. Furthermore, the closest distance from spreading and near-source saturation is given as
each site to the ruptured fault was calculated using the
USGS fault plane solution (Chen and Gravin, 2008). y = b1 + b2 S + (b3 + b4 S ) ln(r + c) + (1)
Recordings off the ends of the ruptured fault are excluded
from our data set. The recording stations are listed in where y is the natural logarithm of PGAH, PGAN,
Table 1 as well as site conditions, rupture distance, PGAP, SAH, SAN or SAP, S is a parameter introduced to
PGA, SA, etc. Since detailed site conditions were not model the site conditions using 0 for rock sites and 1 for
accessible at the time of this article, a general idea of soil sites, r denotes the distance in km used to represent
the site conditions for each site can only be derived by the geometric spreading effect, ε is the regression error,
the modest information given in the recording data files. and c is the near-source saturation term and assumed
Therefore, site conditions are classified as simply soil or to be a constant as we focus on a single earthquake.
rock. Figure 1 shows the fault vertical projection on the By a nonlinear least square regression analysis, all
ground surface, epicenter and the position of the stations the coefficients in Eq. (1) are obtained and are listed
used in this study with respect to the fault projection as in Table 2 for PGAH, PGAN, PGAP and Table 3 for
well as their site conditions. spectral accelerations SAH, SAN, SAP. Corresponding
attenuation relationships are shown in Fig. 2 and Fig. 3.
Table 2 Values of coefficients of the Wenchuan-specific attenuation relationship for different components of PGA
Component b1 b2 b3 b4 c ε
PGAH 7.9988 -0.46955 -0.8552 0.25216 1.8543 0.40592
PGAN 8.6887 -1.0401 -1.0309 0.39635 2.2206 0.39189
PGAP 7.3232 0.084031 -0.68251 0.11166 1.4655 0.45574
Table 3 Values of coefficients of the Wenchuan-specific attenuation relationship for different components of SA at several periods
Component Period (s) b1 b2 b3 b4 c ε
SAH 0.1 8.8802 -0.94880 -0.92116 0.43436 0.82994 0.45529
0.5 8.7459 -0.77373 -0.948000 0.29963 3.01370 0.63148
0.6 8.0680 -0.22258 -0.80906 0.12463 3.04810 0.60202
2.0 4.4587 1.66290 -0.16787 -0.46188 0.91246 0.63394
SAN 0.1 9.5339 -1.56710 -1.07990 0.58915 1.34520 0.45361
0.5 9.2772 -0.76037 -1.06640 0.29392 5.13010 0.66642
0.6 8.0570 0.044951 -0.79881 0.061037 2.97330 0.63420
2.0 5.1994 1.20190 -0.36784 -0.32000 1.33450 0.63510
SAP 0.1 8.2533 -0.35547 -0.76789 0.28483 0.40300 0.49233
0.5 8.2715 -0.81422 -0.84117 0.31222 1.05420 0.63746
0.6 8.0802 -0.49123 -0.81954 0.18844 3.13710 0.61145
2.0 3.7413 2.10670 0.027129 -0.60003 0.51033 0.67757
103
Peak acc. of different components (cm/s2)
102
PGA H
Recorded PGAH
PGAN
Recorded PGAN
PGAP
Recorded PGAP
101 0
10 101 102
Rupture distance (km)
Fig. 2 Comparison of the Wenchuan-specific attenuation relationship for PGAH, PGAN and PGAP at soil site
with observations in previous analysis of the hanging hanging wall sites at a distance range of 8 km to 30 km
wall effect on near-fault ground motions. In Fig. 4, is 0.29±0.15 (Abrahamson and Somerville, 1996) and
the footwall residuals are placed at a negative rupture in the Chi-Chi earthquake, at a distance range of 3 km
distance for purposes of comparison with hanging wall to 30 km, the hanging wall bias is 0.43±0.058 and the
residuals. Note that in the 5 km to 40 km distance range, footwall bias is 0.03±0.006 (Shabestari and Yamazaki,
the hanging wall has the most significant effect, and the 2003). Such a large variety among the results from
residuals have a mean bias of 0.3147±0.1408. However, different earthquakes may be a result of scarce near-
the scarcity of data in the short rupture distance area fault data and mechanism of faults. The synthetic results
and especially the lack of available data renders the (Wang et al., 2008 and Wang, 2010) suggested that as
analysis of the Wenchuan-specific hanging wall effect the dip of the fault increases from 0º to 90°, the hanging
inconclusive with regard to exactly how intensively wall effect immediately becomes increasingly obvious
the hanging wall effect impacts the near fault ground and then gradually weakens after a peak at around 20º
motions, even though it can be inferred from the current dipping until it completely disappeared at 90º dipping.
results that an obvious hanging wall effect did exist in Furthermore, the hanging wall effect is larger for reverse
this devastating earthquake. From an overview of the and thrust earthquakes than for strike-slip earthquakes
previous studies of hanging wall effect, it was found (Somerville and Graves, 2003).
that in the Northridge earthquake, the mean bias of the
No.3 Peng Lei et al.: A new attenuation model of near-fault ground motions with consideration of the hanging wall effect in the Wenchuan earthquake 317
104
T=0.1 s T=0.5 s
103 103
104 104
T=0.6 s T=2.0 s
103 103
102 SAH
102 SAH
Recorded SAH Recorded SAH
SAN SAN
Recorded SAN Recorded SAN
SAP SAP
Recorded SAP Recorded SAP
101 101
100 101 102 100 101 102
Rupture distance (km) Rupture distance (km)
Fig. 3 Comparison of the Wenchuan-specific attenuation relationship for SAH, SAN and SAP at soil site
1.5
terms of piecewise function as shown below in Eqs.
Soil site
Rock site
(2) and (3) that were used for the Northridge and Chi-
1.0 Chi earthquakes, respectively. These two models have
a similar form, and both have four boundary distances
0.5 x1, x2, x3, x4 to be determined. Thus, the AS1996 model
PGAH residual
0
requires supplementation of data from other earthquakes
due to limited data from Northridge earthquake. Then,
-0.5 data from the Northridge earthquake was used alone
to regress the coefficient c of the model after boundary
-1.0 distances have been fixed. Thus, an inter-event error (i.e.,
-1.5
Footwall Hanging wall error originated from variability from one earthquake
-100 -80 -60 -40 -20 0 20 40 60 80 100 to another) was introduced (Abrahamson and Youngs,
Rupture distance (km)
1992). The model, therefore, to some extent deviates
Fig. 4 Residuals of PGAH at recording sites during Wenchuan from the accurate modeling of the hanging wall effect in
earthquake with those at hanging wall placed at one earthquake. In addition, it was found that the process
positive rupture distance and those at footwall placed of obtaining the four boundary distances is quite intricate
at negative rupture distance and it was difficult to determine how the authors could
obtain the four boundary distances all being an exact
integer, if the least square method was used to regress
4.2 Empirical model of the hanging wall effect them. Of course, the final results may be rounded, but
for PGAH such a process would also introduce subjectivity.
As noted above, the hanging wall effect exhibits a The KS2003 model differs from the AS1996 only
trend, with residuals of hanging wall sites having more in that the constant value which was used to describe
extensive ground motion as shown in the residuals of the the lasting part of the most obvious hanging wall effect
PGAH in Fig. 4. It is possible and necessary to model between distances of x2, x3 in the AS1996 model is
the trend specifically for the earthquake of concern to changed into a function of Pr d (r ) Hanging wall − Pr d (r ) All .
thoroughly analyze its hanging wall effect. Abrahamson This function is designed to move the general mean
and Somerville (1996) (AS1996) and Shabestari and attenuation value based on data from all the sites away
Yamazaki (2003) (KS2003) both proposed models in from the mean attenuation value based on the data from
318 EARTHQUAKE ENGINEERING AND ENGINEERING VIBRATION Vol.10
only the hanging wall sites. Since there was a rich near the hanging wall and the footwall (Abrahamson and
fault data set from the Chi-Chi earthquake, the authors Somerville, 1996). Therefore, x1 should be determined
used only the Chi-Chi data. Nevertheless, the KS2003 by the depth of the upper edge of a dipping fault. Then,
model also seems to have a problem determining the to determine the rest of the six unknowns i.e., the x2 and
four boundary distances, and the process of finding these c1 to c5, six conditions can be established, including: (1)
distances is still intricate. In addition, the KS2003 model four continuity conditions of R1 and R2 and their first
used the mean attenuation value only in the range of x2, x3 derivatives with respective to r at the joint point x2, and
to obtain Pr d (r ) Hanging wall − Pr d (r ) All , and the coefficients (2) two suite regression analyses using iterations of the
c1, c2 equal to the values of Pr d ( x2 ) Hanging wall − Pr d ( x2 ) All boundary distance x2 to achieve a least total summation
and Pr d ( x3 ) Hanging wall − Pr d ( x3 ) All , respectively. of the squared residuals, one on the recorded data over
Consequently, the influence of data falling in the the region of x1 to x2 and one on the recorded data over
distance range of less than x2 and larger than x3 was the entire region.
ignored, resulting in a bias to be generated.
4.3 Comparison of the proposed model with the
⎧ 0 for 0 ≤ r ≤ x1 AS1996 and KS2003 models
⎪
⎪ c ⎡cos( π(r − x1 ) + π) + 1⎤ for x1 < r < x2 Figure 5 shows a comparison of PGAH residuals
⎪ 2 ⎢⎣ x2 − x1 ⎥
⎦ obtained by using the Northridge earthquake data fitted
⎪ by the proposed model and by the AS1996. Note that
( )
HW r = ⎨ c for x2 ≤ r ≤ x3
⎪c the two models yield almost the same results. However,
⎪ ⎡cos( π(r − x3 ) ) + 1⎤ for x3 < r < x4 as stated previously, the AS1996 model introduced an
⎪ 2 ⎢⎣ x4 − x3 ⎥
⎦ inter-event error by mixing data from other earthquakes
⎪ in determining the boundary distances, and its definition
⎩ 0 for r ≥ x4
of the distance limits the effect of the disappearance;
AS1996 model (2) thus, the results depart from the actual hanging wall
effect. However, due to the scarcity of data, and also
the dominant influence of intra-event errors compared to
⎧ c1 ⎡ π(r − x1 ) ⎤ inter-event errors, this type of departure was not obvious
⎪ 2 ⎢cos( x − x + π) + 1⎥ for x1 < r < x2
⎪ ⎣ 2 1 ⎦ and discernable. Thus, the proposed model is designed
⎪ to avoid these aspects of bias, and the resulting curve
HWeffect = ⎨Pr d(r )Hanging wall − Pr d(r )All for x2 ≤ r ≤ x3
⎪c is balanced among the data. A similar comparison is
⎪ 2 ⎡cos( π(r − x3 ) + 1) ⎤ for x3 < r < x4 shown in Fig. 6 for the Chi-Chi earthquake data fitted
⎪⎩ 2 ⎢⎣ x4 − x3 ⎥
⎦ by the proposed model and by KS2003. Note that these
two models predict a significant difference, even though
KS2003 model (3) the KS2003 model involves no inter-event error terms.
The major difference between the two models is in the
where HW (r ) and HWeffect denote residuals of PGAs on KS2003’s omission of the influence of the data that falls
hanging walls used in AS1996 and KS2003, respectively. in the distance range of less than x2 and larger than x3.
In this study, to improve the models of AS1996 and The Chi-Chi earthquake provides much more data than
KS2003, it is suggested that: (1) the hanging wall effect the Northridge earthquake, and the data distributes more
should be reduced gradually instead of being restricted evenly over the distance range, thus the departure from
in a definite distance limit, and disappear immediately
once the rupture distance limit is over, such as the
distance limit x4 in AS1996 and KS2003; and (2) all data 1.5
on hanging wall sites should be used in the regression Our model
Model in AS1996
analysis. Based on the analysis results, the model is 1.0 Northridge recording data
revised as:
PGAH residual
where the residual R may refer to both the PGA and SA -0.5
on hanging wall sites; the boundary distances x1 and x2
and coefficient c1, c2, c3, c4 and c5 are all to be determined -1.0
Footwall Hanging wall
by residual distribution of hanging wall sites. In -40 -30 -20 -10 0 10 20 30 40
Rupture distance (km)
particular, because the hanging wall effect originated
from the general proximity of the hanging wall sites to Fig. 5 Comparison of the predictions by using the proposed
a fault plane, sites directly over the ground surface trace model and the AS1996 when applied to the Northridge
earthquake
of the fault rupture or the vertical projection of the upper
edge of a buried fault should be the division between
No.3 Peng Lei et al.: A new attenuation model of near-fault ground motions with consideration of the hanging wall effect in the Wenchuan earthquake 319
the actual hanging wall effect due to the omission of Table 4 Boundary distances and coefficients of model for
some sections of data would become significant. The hanging wall effect of the PGAH in Wenchuan
modeling curve of KS2003 in Fig. 6 turns out to enclose earthquake
almost all the data in the distance range of less than Parameter Value
40 km, where the most significant hanging wall effect x1 (km) 0
occurs, while the proposed model runs through all data x2 (km) 39.29
in a more balanced way. The results indicate that the c1 -0.0006943
KS2003 model predicts a much higher hanging wall c2 0.0314
effect for the Chi-Chi earthquake. On the other hand, c3 0
due to the abundance of near-fault data from the Chi- c4 -0.1433
Chi earthquake, the proposed model can ensure that c5 3.8093
the hanging wall effect weakens smoothly at distances
of around 50 km rather than being forced to disappear
immediately, as expected when designing the model. distribution is shown in Fig. 7, where the hanging
Further, this type of slowly disappearing effect also can wall effect term used in the AS NGA model with some
be captured in some NGA models, such as the model modifications to the original form used in Abrahamson
developed by Campbell and Bozorgnia (2007). and Silva (1997) is also plotted. The previous model
lead to jumps in the hanging wall effect scaling for some
0.6 cases and it was not clear how to apply the model for
steeply dipping faults (Abrahamson and Silva, 2008).
0.4
In fact, the previous model used in the ground motion
0.2 attenuation relationship of Abrahamson and Silva (1997)
PGA resultant residual
of the peak values of the two as-recorded orthogonal recognized that the rupture directivity has a significant
horizontal components. This definition of ground motion effect in long periods. Therefore, the residuals of SAH
depends on the orientation of the sensor as installed in at several periods from short to long are examined in
the field. This means that the ground motion parameter an attempt to identify the influence of the directivity
could differ for the same ground motion when the effect. Because of the large reverse slip in some areas
sensor has a different orientation. This dependence on of the fault plane (Chen and Gavin, 2008), it is assumed
the sensor’s orientation is most significant for strongly that the difference between the spectral acceleration
correlated motions that often occur at periods of 1 sec residuals of the hanging wall and the footwall caused
or longer (Campbell and Bozorgnia, 2007). The new by the hanging wall effect would be influenced by the
measure of ground motion, GMRotI50, removes the rupture directivity effect at long periods.
dependence on the sensor orientation, as it is computed
based on a set of geometric means of the as-recorded 5.1 Residual distribution of SAH
orthogonal horizontal motions rotated through all
possible non-redundant rotation angles. And, a single- The residuals of SAH at four different periods
are plotted in Fig. 8 with the hanging wall sites at a
positive rupture distance and the footwall sites at a
1.5
negative rupture distance. Note that at the short period
1.0 of 0.1s, residuals within 60 km on the hanging wall
side are inclined to be mostly positive and those on the
0.5 footwall side are inclined to be mostly negative. Thus,
PGAH residual
1.0 1.0
0.5 0.5
SAH residual
SAH residual
0 0
-0.5 -0.5
-1.0 -1.0
Footwall Hanging wall Footwall Hanging wall
-1.5 -1.5
-100 -80 -60 -40 -20 0 20 40 60 80 100 -100 -80 -60 -40 -20 0 20 40 60 80 100
Rupture distance (km) Rupture distance (km)
1.0 1.0
0.5 0.5
SAH residual
SAH residual
0 0
-0.5 -0.5
-1.0 -1.0
Footwall Hanging wall Footwall Hanging wall
-1.5 -1.5
-100 -80 -60 -40 -20 0 20 40 60 80 100 -100 -80 -60 -40 -20 0 20 40 60 80 100
Rupture distance (km) Rupture distance (km)
Fig. 8 Residuals of averaged SAH of 5% damping at periods 0.1 s, 0.5 s, 0.6 s and 2.0 s in the Wenchuan earthquake
would exhibit in a broader area away from the fault trace effect disappears, the hanging wall effect on SAH is
on the footwall. Thus footwall stations will experience still very pronounced. Presumably, the scarcity of data
much more influence from rupture directivity under and the resulting aleatory distribution pattern of data
the conditions of the rupture directivity effect in the especially at large distances in this earthquake might be
Wenchuan earthquake, and thereby explaining the the explanation for this phenomenon. But as mentioned
reverse trend of residual distribution. previously, it is possible that the AS NGA model’s
estimation of the ending distance of no effect would
5.2 Empirical modeling of the hanging wall effect introduce uncertainty and cause such a large difference.
for SAH However, the AS NGA model’s prediction does include
all data points on the hanging wall within the distance
Because all first phase NGA models did not include range with the largest effect in Fig. 9. From this
any rupture directivity effect factors, the rupture phenomenon, it is possible that the hanging wall effect
directivity effect is mixed within the ground motion at of SAH in this earthquake remains in the predictable
long periods. When modeling the hanging wall effect for level for a distance range with the most significant effect
SAH and comparing it with the hanging wall function as well. And, the AS NGA model achieved an acceptable
for spectral acceleration in the AS NGA model, it would prediction of the hanging wall effect to some extent in
be best to model the hanging wall effect for SAH only this earthquake.
at 0.1s to avoid introducing the influence of rupture
directivity so no rupture directivity effect is included in
the comparison. The same model as previously used as for Table 5 Boundary distances and coefficients of the model for
PGAH is again used here. By using a similar procedure hanging wall effect of the SAH in Wenchuan earthquake
described in subsection 4.2 for PGAH, all coefficients
Parameter Value
and boundary distances of the model obtained for SAH
of 0.1s based on the 0.1s residuals shown in Fig. 8 are x1 (km) 0
listed in Table 5 and the SAH residual distribution for x2 (km) 24.17
the Wenchuan earthquake is shown in Fig. 9. Note c1 -0.0012313
that just like the previous analysis for PGAH, the AS c2 0.049363
NGA once again predicts much larger SAH residuals. c3 0
However, in this case, it predicts a very short distance of
c4 -0.021441
the hanging wall effect. In addition, in the distance range
c5 -0.22877
where the AS NGA model predicts that the hanging wall
322 EARTHQUAKE ENGINEERING AND ENGINEERING VIBRATION Vol.10
0.5 References
PGAH residual
20, 1999 Chi-Chi, Taiwan, Earthquake Estimated from Duration Effect of Rupture Directivity,” Seismological
Rayleigh-wave Phase Velocity,” Earth Planets Space, Research Letters, 68(1): 199–222.
53: 1171–1176. Spudich Paul and Chiou BSJ (2008), “Directivity in NGA
Li Yong, Zhou Rongjun et al. (2008), “Surface Rupture, Earthquake Ground Motions: Analysis Using Isochrone
Reverse-strike Slip in the Wenchuan Earthquake,” Theory,” Earthquake Spectral, 24(1): 279–298.
Journal of Chengdu University of Technology (Science Spudich P, Chiou BSJ, Graves R, Collins N and
& Technology Edition), 35(4): 404–413. (in Chinese) Somerville P (2006), “A Formulation of Directivity
Shabestari KT and Yamazaki Fumio (2003), “Near- for Earthquake Sources Using Isochrones Theory,”
fault Spatial Variation in Strong Ground Motion due U.S Geological Survey Open File Report 2004-1263,
to Rupture Directivity and Hanging Wall Effect from available at http://usgs.gov/of/2004/1268/.
the Chi-Chi, Taiwan Earthquake,” Earthquake Engng Wang Dong (2010), “The Hanging Wall/Footwall Effect
Struct. Dyn, 32: 2197–2219. of Near-Fault Ground Motions,” PhD Dissertation,
Somerville PG and Graves RW (2003), “Characterization Institute of Engineering Mechanics, China Earthquake
of Earthquake Strong Ground Motion,” Pure Appl. Administration, Harbin, China.
Geophys, 160(2003): 1811–1828. Wang Dong, Xie Lili and Hu Jinjun (2008), “Geometric
Somerville PG, Smith NF, Graves RW (1997), Effect Resulting from the Asymmetry of Dipping Fault:
“Modification of Empirical Strong Ground Motion Hanging Wall/ Footwall Effect,” Acta Seismologica
Attenuation Relations to Include the Amplitude and Sinica, 30(3): 271–278.