You are on page 1of 32

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/329078314

Seismic Response of Embankment Dams Based on Recorded Strong-Motion


Data in Japan

Article  in  Earthquake Spectra · November 2018


DOI: 10.1193/042918EQS107M

CITATIONS READS
0 308

2 authors, including:

Tadahiro Kishida
Khalifa University
74 PUBLICATIONS   524 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Engineering Implications of the 2014 South Napa Earthquake Strong Motions View project

GEER reconnaissance of central Italy earthquake sequence View project

All content following this page was uploaded by Tadahiro Kishida on 26 November 2018.

The user has requested enhancement of the downloaded file.


The Professional Journal of the Earthquake Engineering Research Institute

PREPRINT
This preprint is a PDF of a manuscript that has been accepted for
publication in Earthquake Spectra. It is the final version that was
uploaded and approved by the author(s). While the paper has been
through the usual rigorous peer review process for the Journal, it has
not been copyedited, nor have the figures and tables been modified for
final publication. Please also note that the paper may refer to online
Appendices that are not yet available.

We have posted this preliminary version of the manuscript online in


the interest of making the scientific findings available for distribution
and citation as quickly as possible following acceptance. However,
readers should be aware that the final, published version will look
different from this version and may also have some differences in
content.

The DOI for this manuscript and the correct format for citing the paper are
given at the top of the online (html) abstract.

Once the final, published version of this paper is posted online, it will
replace the preliminary version at the specified DOI.
1 Seismic Response of Embankment Dams based
2 on Recorded Strong Motion Data in Japan
3 DongSoon Parka) M.EERI, and Tadahiro Kishidab) M.EERI

4 It is important to investigate strong-motion time series recorded at dams to


5 understand their complex seismic responses. This paper develops a strong-motion
6 database recorded at existing embankment dams and analyzes correlations
7 between dam dynamic responses and ground-motion parameters. The JCOLD
8 database used here includes 190 recordings at crests and foundations of 60 dams
9 during 54 earthquakes from 1978 to 2012. Seismic amplifications and
10 fundamental periods from recorded time series were computed and examined by
11 correlating with shaking intensities and dam geometries. Peak ground acceleration
12 (PGA) at dam crest increases as PGA at foundation bedrock increases but their
13 ratio gradually decreases. Fundamental period broadly increases with dam height
14 and PGA at foundation bedrock. Nonlinear dam response becomes more apparent
15 as PGA at foundation bedrock becomes >0.2 g. Prediction models of these
16 correlations are proposed for estimating the seismic response of embankment
17 dams, which can inform the preliminary design stage.

18

19 INTRODUCTION

20 There are more than 59,000 large dams worldwide, of which 78% employ earthfill and
21 rockfill construction (ICOLD 2018). Therefore, the seismic safety of embankment dams is an
22 important issue. Several large earthquakes were recently recorded at embankment dams and
23 have been investigated by several researchers. For example, during the 2008 Wenchuan
24 earthquake (M7.9), the 156-meter-high Zipingpu concrete-faced rockfill dam (CFRD) was
25 damaged partially without any collapse or freeboard deficiency. The dam was designed with
26 peak ground acceleration (PGA) of 0.26 g at its foundation bedrock, whereas the recorded
27 data exceed 0.5 g. Moreover, PGAs were 1.65 g and 2.0 g for the stream and axis directions

a)
K-water Convergence Research Institute, 1689 beon-gil 125, Yuseong-daero, Yuseong-gu, Daejeon, Republic
of Korea 34045; Email: dspark210@gmail.com (D. P.)
b)
Khalifa University, Abu Dhabi, UAE P.O. Box 127788
28 at the crest, respectively (Kong et al. 2010). Similarly, during the 2011 Tohoku earthquake
29 (M9.0), Aratozawa rockfill dam experienced PGA of 1.04 g at foundation rock but 0.54 g at
30 the crest (JCOLD 2014), which may indicate the nonlinear response of the embankment.
31 However, these significantly strong excitations did not endanger the safety of the rockfill
32 dam. The USSD (2014) also reports that most modern embankment dams are capable of
33 withstanding significant seismic shaking with no detrimental consequences in the past events.
34 However, uncertainty remains regarding the seismic performance of embankment dams
35 (Wieland 2012).

36 Several factors affect the seismic responses of embankment dams. Gazetas and Dakoulas
37 (1992) show that pressure-dependent stiffness, inelastic behaviors, and bedrock geometries
38 are important factors. Kim et al. (2011) report that PGA is an important factor, whereas Yuan
39 et al. (2014) explain that frequency contents of input motions with reference to the
40 fundamental periods of the embankments or the period of reservoir impounding, and zoning
41 condition are important. Albano et al. (2015) and Park and Kim (2017) show that PGA,
42 duration, reservoir water level, and the resonance effect are also important. However, these
43 studies are based on analytical and physical modeling; hence, it is also necessary to employ
44 field observations in order to understand the seismic behaviors of existing dams subjected to
45 very high ground acceleration. Since only limited case histories are available for recent
46 earthquakes, the analyses based on these measured strong motion data are particularly
47 valuable.

48 Several studies show that the fundamental period of a dam is influenced by the dominant
49 period of shaking, with reference to its proximity to the fundamental period of a dam (Albano
50 et al. 2015; Gazetas and Dakoulas 1992; Kim et al. 2011; Park and Kim 2017; Uddin and
51 Gazetas 1995; Yuan et al. 2014). Moreover, variations in wave amplification and
52 fundamental period are correlated with PGA, Arias Intensity, and response spectrum due to
53 the nonlinear dam responses. However, for embankment dams, there are only limited
54 prediction studies of fundamental period based on recorded strong-motion data.

55 This paper empirically studies correlations between the characteristics of various seismic
56 parameters, and the seismic amplification features and fundamental period of embankment
57 dams, using the strong-motion database from the Japan Commission on Large Dams
58 (JCOLD, 2014). An overview of this database was previously reported by Sasaki (2015). The
59 database includes a large number of acceleration time series measured at dams from 1978 to

2
60 2012. These were recorded at various locations on the dams (e.g., foundations, ground, and
61 dam crests) with different azimuthal orientations (e.g., stream, dam axis, and vertical
62 directions).

63 These acceleration time series were processed using the methodology developed in
64 previous studies (Chiou et al. 2008; Ancheta et al. 2013; Kishida et al. 2016). The ground-
65 motion parameters (e.g., PGA, Arias Intensity) are investigated to determine their influences
66 on the response characteristics of embankment dams, such as fundamental periods and wave
67 amplification factor. Correlations between these variables are studied using multiple linear
68 regression analysis with mixed-effect models.

69

70 STRONG-MOTION DATABASE

71 The database includes acceleration time series for 60 embankment dams in Japan over 34
72 years (1978–2012). There are 190 three-component time series from 54 earthquakes, which
73 were recorded at both the foundation and crest of the dams, giving more than 1,100 time
74 series. The dams examined in this study are classified as rockfill (R; n=118, 62%) or earthfill
75 (E; n=72, 38%) type, and the database is summarized in Table 1 and Fig. 1. The heights and
76 lenghts of the dams range from 19 to 176 m and 90 to 1,600 m, respectively (Table 1).
77 Epicentral distance (Re) and magnitude (M) range from 2 to 300 km and 2.0 to 9.0,
78 respectively (Fig. 1).

79 Table 1. Summary: dimensions of analyzed rockfill and earthfill dams

H (m) L (m) L/H


Min. 19.0 90 0.51
Med. 48.9 310 5.56
Max. 176.0 1,597 42.03
STD 32.5 270 7.73
80
81
82 Figure 1. Scatter plots of (a) magnitude and epicentral distance, and (b) magnitude and focal depth in
83 the database.

84

3
85 OVERVIEW OF ANALYZED SEISMIC PARAMETERS

86 VARIATION IN GROUND-MOTION PARAMETERS

87 Table 2 shows the data distribution in the stream direction. PGA ranges from 0.012 g to
88 1.044 g at foundation rock (i.e. ap,f), and the majority of the data are <0.1 g. For example, the
89 75% quartile value of PGA is 0.082 g at foundation rocks. Since minimum and maximum
90 PGA range widely from 0.01 to 1.0 g, the proposed equations in the following sections will
91 be applied within this range limitation. The parameters listed in Table 2 are defined in the
92 glossary section.

93

94 Table 2. Statistical distribution of strong-motion data in the stream direction


Parame Unit Max. 75% Median 25% Min. Mean STD STD
-ter quartile quartile Error
Mean
ap.f g 1.044 0.082 0.044 0.029 0.012 0.079 0.114 0.008
ap.c g 1.495 0.218 0.119 0.076 0.022 0.179 0.180 0.013
ap.c/ap.f 13.7 3.8 2.7 1.7 0.5 3.0 2.0 0.1
AIf m/s 5.167 0.103 0.014 0.005 0.0 0.183 0.566 0.042
AIc m/s 16.404 0.967 0.145 0.039 0.003 1.327 2.943 0.219
AIc/AIf 129.2 16.9 10.2 3.7 0.2 15.1 18.5 1.4
Tn.fft sec 0.794 0.501 0.398 0.324 0.054 0.412 0.152 0.014
Tn/H sec/m 0.028 0.011 0.007 0.005 0.0005 0.008 0.004 0.0003
Tn.h/Tn.v 11.0 2.4 1.9 1.5 0.0 2.3 1.6 0.1
95

96 CALCULATION OF FUNDAMENTAL PERIODS

97 One of the important factors influencing the dynamic response of a dam is the frequency
98 content of the excitation relative to the natural frequency of the dam (Uddin and Gazetas
99 1995). The fundamental period of a dam is defined as the period that leads to the maximum
100 spectral amplification from the dam foundation to its crest. Various researchers have
101 employed many different approaches to obtain the fundamental periods of dams (Abdel-
102 Ghaffar and Scott 1979; Ahlberg et al. 1972; Bray and Travasarou 2007; Chugh 1985;
103 Gazetas 1987; Ha 2011; Kim and Ha 2012; Makdisi and Seed 1979; Mejia et al. 1982; Oner
104 1984; Papadimitriou et al. 2014; Uddin and Gazetas 1995).

105 In this study, fundamental periods (Tn) were calculated using Fourier amplitude spectra
106 (FAS). FAS at foundation rock and crest were smoothed separately using the function
107 proposed by Konno and Ohmachi (1998).

4
108 𝑊 = [sin(log10 (𝑓⁄𝑓𝑐 )𝑏 )⁄log10 (𝑓 ⁄𝑓𝑐 )𝑏 ]4 (1)

109 Here, W is a proposed logarithmic window function. b and fc are coefficients of band
110 width and the central frequency of the smoothing window. f represents the frequency of FAS.
111 This function was originally used to find peak responses from ratios between horizontal and
112 vertical FAS (H/V spectra). The smoothing parameter of b varies according to the specific
113 purpose of the study. Boore (2008) used b=20 to smooth FAS. Woolery et al. (2009)
114 similarly used b=20 to evaluate linear site response from H/V spectra. Ktenidou et al. (2016)
115 used b=40 to study the directional site effects from FAS. Laurendeau et al. (2017) and Perron
116 et al. (2017) uses b=30 to study hard-rock reference motions from downhole recordings and
117 robust kappa () measurements in low-to-moderate seismicity areas, respectively. Goulet et
118 al. (2018) used b of approximately 190 to preserve the original FAS characteristics after
119 smoothing, to predict PSA by modeling FAS through random vibration theory.

120 In the present study, the objective of smoothing FAS is to determine the fundamental
121 periods of dams for different azimuthal angles from downhole array records including soil
122 nonlinear behaviors. Based on this objective, a visual review was performed of the window
123 size regarding the reduction of data variability and the sensitivity of the peak responses.
124 Through this process, a value of b=60 was selected for calculating the ratio of FAS in order
125 to determine the fundamental periods.

126 Fig. 2 shows the FAS in the stream direction at the crest and foundation for
127 Surikamigawa dam during the 2011 Tohoku Earthquake. FASs are smoothed by Konno–
128 Ohmachi window (b=60) at the sampling frequencies that distribute 100 points per log10
129 cycle (Goulet et al. 2018). The amplification factors are also plotted from foundation to crest.
130 The figure also shows the minimum and maximum usable frequencies of amplification
131 factors obtained from the signal-to-noise ratio during processing of downhole recordings
132 (Kishida et al. 2016). Between these usable frequencies, the peak amplification and the
133 corresponding predominant period of Tn are selected. This process does not include any
134 subjective judgement, hence the obtained results will be consistent if the same data
135 processing method is applied.

136

137 Figure 2. Example FASs and amplification factors obtained at Surikamigawa dam during 2011
138 Tohoku Earthquake.
139

5
140 AMPLIFICATION OF GROUND MOTION PARAMETERS

141 PEAK ACCELERATIONS BETWEEN FOUNDATION AND CREST

142 Fig. 3 plots peak accelerations at the dam foundation and crest. Regression models
143 between these variables are also presented in Table 3.

144 The most common functional form for the relationships suggested in the literature might
145 be f(x) = 𝑎 𝑥 𝑏 , where the constants a and b are determined by statistical regression of a data
146 set (e.g., Park 2018). By taking the natural logarithm, this power function can be converted to
147 a simple linear line format, which is easy to understand. The main reasons for using the
148 functional form are that: (1) it gives a more intuitive linear plot; (2) it is effective in
149 accounting for fixed effects and random effects of a certain term; and (3) it is in line with
150 other regression models of site amplification factors (e.g., Seyhan and Stewart 2014).

151 The equation is given in the form of a bilinear fitting because of the decrease of
152 amplification ratio at higher g-level (e.g., >0.2 g) possibly due to nonlinearity. As seen in
153 Table 3, for ap.f ≤0.2 g, the slope of fit, b is close to 1.0, which implicates the amplification is
154 nearly elastic at small shaking. In contrast, for ap.f >0.2 g, the value of b is much lower, which
155 shows inelastic response. From Fig. 3, the proposed model provides a reasonable overall fits
156 to the data. It also shows that the direction of motion or the dam type does not make any
157 significant difference to the shape of fitting curves. Note should be taken that the majority of
158 data are within the smaller PGA regime (95% for PGA <0.2 g) so that larger PGA gives some
159 epistemic uncertainties. Residual data were further reviewed, but we did not observe
160 noticeable biases. Standard deviations () show that the horizontal direction data contain
161 more uncertainty than the vertical direction data (Table 3).

162

163 Table 3. Fitting of peak acceleration between dam crest and foundation for embankment dams
Direction Data Fitting of ln(ap.c) = a + b ln(ap.f)
ap.f ≤0.2 g ap.f >0.2 g
a b  R2 a b 
Dam axis E+R 0.398 0.848 0.512 0.547 -0.74 0.141 0.788
Stream E+R 0.147 0.731 0.557 0.410 -0.62 0.254 0.399
Vertical E+R 0.468 0.885 0.454 0.713 -0.42 0.333 0.279
164
165
166 Figure 3. Regression fitting of peak acceleration between dam crest and foundation: (a) dam axis
167 direction, (b) stream direction, and (c) vertical direction.

6
168 The relationship of ap.f and the ratio of ap.c to ap.f (ap.c/ap.f) is plotted in Fig. 4 with the
169 regression fit, and are basically the same as those shown in Fig. 3 and Table 3. With some
170 exceptions, peak accelerations are amplified. However, amplifications become smaller at
171 higher PGA, as observed in Fig. 3. The ap.c/ap.f mainly ranges from 0.4 to 8, with ranges (and
172 median values) of 0.2–9.3 (2.3); 0.5–13.7 (2.7); and 0.6–11.1 (2.3) in the dam axis, stream,
173 and vertical directions, respectively. The reduction of ap.c/ap.f with ap.f for ap.f >0.2 g is
174 associated with energy dissipation by increased damping, and with nonlinearity of soil
175 stiffness due to increased shaking intensity (Albano et al. 2015; Kim et al. 2012; Yuan et al.
176 2014). At a PGA of approximately 0.4 g, the amplification ratio approaches 1.0. The review
177 showed no clear difference in the trend, either between rockfill and earthfill dams, or between
178 the stream- and dam-axis directions.

179

180 Figure 4. Regression fitting of PGA at foundation and peak acceleration amplification of motion for
181 earthfill and rockfill dams: (a) dam axis direction, (b) stream direction, and (c) vertical direction.

182 We reviewed the correlation between dam dimensions of L/H and ap.c/ap.f (Fig. 5). Along
183 with the large scatter in the dataset, the overall trend is that ap.c/ap.f decreases as L/H
184 increases, which implies that a dam with relatively longer dam axis and lower height will
185 tend to display less amplification of a motion. These observations may relate to the two-
186 dimensional effects of dam responses, but further investigation of this issue is required to
187 reach a definite conclusion.

188
Figure 5. Plot of amplification ratio and L/H for: (a) dam axis direction, (b) stream direction.

189

190 ARIAS INTENSITIES BETWEEN DAM FOUNDATION AND CREST

191 Arias Intensity (AI) has the advantage of reflecting the combined effects of the amplitude
192 and duration of a motion, which is important for estimating structural damage (Kim and Ha
193 2012). Regression analyses were performed on AI ratio (AIc/AIf) between foundation and
194 crest for the dam axis and stream directions, respectively. Table 4 shows the bilinear
195 regression models between these variables.

196 Fig. 6 shows a scatter plot of AIc/AIf against ap.f with the regression lines presented in
197 Table 4. AIc/AIf is as large as 100 when ap.f is small; however, it tends to decrease as ap.f

7
198 increases, and it is more apparent as ap.f approaches 1.0 g, potentially due to nonlinear
199 behavior of embankment dams. There is little difference between the two horizontal
200 components in Fig. 6 (as seen in the data distribution) and in Table 4 (as seen in the similar
201 values of a, b, and ). We also reviewed rockfill and earthfill dams, but found no noticeable
202 differences.

203 Table 4. Fitting of Arias Intensity ratio between dam crest and foundation depending on ap.f for
204 embankment dams
Direction Data ln(AIc/AIf) = a + b ln(ap.f)
ap.f ≤0.2 g ap.f >0.2 g
a b  a b 
Dam axis E+R 1.101 -0.272 1.041 -0.140 -1.043 0.579
Stream E+R 1.038 -0.384 1.067 -0.060 -1.066 0.432
Vertical E+R 1.088 -0.198 0.844 -0.300 -1.061 0.474
205
206
207 Figure 6. Plot of Arias Intensity amplification ratio depending on PGA: (a) dam axis direction, (b)
208 stream direction, and (c) vertical direction.

209

210 FUNDAMENTAL PERIOD OF EMBANKMENT DAMS

211 EFFECT OF DAM HEIGHT AND LENGTH ON FUNDAMENTAL PERIOD

212 It is expected that the responses along the dam stream direction and those along the dam
213 axis direction would be affected by the dam height and crest length (or the valley width),
214 respectively. Quite possibly, the crest length over dam height ratio could play a role in
215 fundamental period. The correlations between H and Tn, and L and Tn in both directions are
216 evaluated in this section.

217 One traditional approaches to predicting the Tn of embankment dams is to use the dam
218 height (H). Linear regression models are typically proposed between Tn versus H. Fig. 7
219 shows the scatter plots between Tn and H obtained in this study; and the regression results
220 (the parameters for which are presented in Table 5). Most of the fundamental periods range
221 from 0.1 to 1.0 s. The data distributions and fitting shapes of the dam axis and stream
222 directions show reasonable agreement. A residual plot revealed no apparent bias in the data.
223 The results show that the predicted Tn in the horizontal direction is longer than that in the
224 vertical direction.

225

8
226 Table 5. Fitting of dam height and fundamental period of embankment dams
Direction Dam type Fitting of H – Tn in the form of ln(Tn) = a + b ln(H)
a b  R2
Dam axis E+R -2.629 0.377 0.464 0.140
Stream E+R -2.685 0.430 0.375 0.254
Vertical E+R -2.793 0.283 0.456 0.100
227

228 Figure 7. Fitting of dam height versus fundamental period of embankment dams.

229

230 Fig. 8 compares H and Tn obtained in this study with the prediction equations in Table 6
231 for stream direction. The figure shows that the proposed model reasonably fits the observed
232 data compared with those proposed in previous studies. Specifically, previous models
233 generally over-predict Tn for dams taller than 110 m, partly due to the complex effect of dam
234 geometry (including valley effect), material characteristics represented by shear stiffness
235 (particularly stiffness inhomogeneity due to dependence of soil stiffness on confining
236 pressure), and nonlinear inelastic response to strong ground excitation (Gazetas 1987;
237 Gazetas and Dakoulas 1992; Park and Kim 2017). Tn might be correlated simply with H and
238 shear wave velocity of material (Vs) for a triangular-shaped sliding mass that largely has a
239 two-dimensional (2D) response. However, because of the complicated nature of nonlinear
240 correlation of stiffness inhomogeneity (Vs versus confining pressure) combined with
241 nonlinear inelastic soil behavior (i.e., soil nonlinearity tends to reduce near-crest peak
242 accelerations), the correlation between Tn and H might be stronger or weaker. Further
243 investigation will be beneficial to understanding the observations presented in Fig. 8.

244

245 Table 6. Comparison of H – Tn fitting equations of this study and in the literature
Source Type Direction Fit equation Remark
This study E+R Stream ln Tn = -2.685 + 0.430 ln H
Papadimitriou et al. (2014) E Stream Tn = 0.024 H0.75 Elastic Tn
Sasaki (2015) E Stream Tn = 1.101/100*H  0.094
R Stream Tn = 0.542/100*H  0.148
246
247
248 Figure 8. Comparison of H – Tn fitting plots of this study and in the literature.

249

9
250 It is useful to explore any potential effect of crest length over dam height on the
251 fundamental period. Because Tn is correlated with H, Tn/H is a meaningful term in order to
252 exclude individual dam height differences from any further correlations. Fig. 9 shows Tn/H
253 versus L/H obtained from the database. The regression models in Table 7 are plotted in Fig.
254 9, showing that Tn/H increases as L/H increases. There seems to be a stronger correlation
255 between L/H and Tn/H for rockfill dams than for earthfill dams. However, these observations
256 are from a limited dataset in which most of L/H values for earthfill dams are below 20. When
257 the data are combined for earthfill and rockfill dams, the resulting model exhibits reasonable
258 correlation between L/H with Tn/H on average.

259

260 Table 7. Nonlinear fitting of L/H versus Tn/H of embankment dams


Direction Dam type ln(Tn/H) = a + b ln(L/H)
a b  R2
Dam axis E+R -6.022 0.481 0.482 0.251
Stream E+R -5.508 0.292 0.433 0.167
Vertical E+R -6.580 0.515 0.496 0.309
261
Figure 9. Regression fitting of L/H versus Tn/H of embankment dams: (a) dam axis direction, (b)
stream direction. Note that the shaded region is the 95% confidence interval for the mean value of
Tn/H, and the dotted line is the 95% prediction interval for a single value of Tn/H.

262

263 The fundamental period of embankment dams is regressed with dam height and length.
264 Table 8 shows a multiple linear regression model of Tn with variables H and L, where the
265 correlation between ln(H) and ln(L) is <0.1 in the database.  is the standard deviation of the
266 random error. The results show that Tn increases as H and L increase for all directions. The
267 stream direction shows that H has the largest effect on Tn, whereas L has the smallest effect.
268  is slightly smaller in the stream direction than in other directions.

269

270 Table 8. Regression model of Tn as a function of H and L for embankment dams


Direction Dam type ln(Tn) = a + b ln(H) + c ln(L)
a b c  R2
Dam axis E+R -3.855 0.325 0.246 0.458 0.182
Stream E+R -2.738 0.429 0.010 0.377 0.254
Vertical E+R -3.787 0.250 0.194 0.449 0.135
271
272

10
273 PGA AND FUNDAMENTAL PERIOD

274 It is beneficial to correlate ap.f with Tn for embankment dams because nonlinearity
275 (represented by Tn) depends on the magnitude of strong shaking (i.e., ap.f). The obtained Tn
276 values are plotted with ap.f in Figs. 10(a–c) for different directions. The resulting regression
277 models are presented in Table 9(a). The majority of the data are <0.2 g, in which data scatter
278 is significant.  ranges from 0.39 to 0.49 depending on the shaking direction. Figs. 10(a–c)
279 also show that the fundamental period of the dam increases with peak acceleration. The slope
280 of b is steeper for the stream direction than the dam axis direction. The plot for the vertical
281 direction shows the shallowest slope, which may indicate a less nonlinear effect than in the
282 horizontal direction. Dam type has no obvious influence on the fundamental period.

283 A correlation of ap.f versus Tn/H was investigated in a similar fashion, as seen in Table
284 9(b). The parameters display a broadly proportional relationship to each other. The slope of b
285 is greater in the stream direction than the dam axis direction, which may indicate that
286 nonlinearity is more apparent for the stream direction as PGA increases. Conversely, the
287 vertical direction shows much less nonlinearity than the horizontal direction. The data for the
288 vertical direction show smaller Tn/H values than those for other directions.

289

290 Table 9. Fitting of peak ground acceleration at dam foundation and fundamental period of
291 embankment
(a) Fitting of ap.f – Tn in the form of ln Tn = a + b ln ap.f
Direction Dam type a b  R2
Dam axis E+R -0.700 0.135 0.488 0.055
Stream E+R -0.303 0.223 0.390 0.194
Vertical E+R -1.238 0.127 0.467 0.058

(b) Fitting of ap.f – Tn/H in the form of ln Tn/H = a + b ln ap.f


Direction Dam type a b  R2
Dam axis E+R -4.672 0.169 0.536 0.070
Stream E+R -4.431 0.178 0.449 0.103
Vertical E+R -5.473 0.045 0.595 0.005
292
293
Figure 10. Regression models of fundamental period of dam with peak acceleration at foundation.

294

295 Figs. 11(a) and 11(b) show the distribution of Tn.h/Tn.v for the dam axis and stream
296 directions, respectively. Most values are >1.0, indicating that Tn in the horizontal directions is

11
297 typically longer than that in the vertical direction, regardless of dam type. The median value
298 of Tn.h/Tn.v is about 1.8 (1.7 for dam axis direction and 1.9 for stream direction). In the
299 seismic design of embankment dams, this is a useful finding when it is necessary to
300 reasonably assume the vertical component of motion from the known horizontal component
301 of motion. It is also useful for developing design spectra requiring specific consideration of
302 multiple spectral accelerations (e.g., Kishida 2017).

303

Figure 11. Distribution plot of Tn.h/Tn.v for embankment dams (red circle: earthfilled; blue triangle:
rockfilled).
304

305 MULTIPLE LINEAR REGRESSION

306 Because Tn is expressed as a function of dam geometry (e.g., H and L) and input motion
307 (e.g., ap.f), it is meaningful to propose a regression model combining these effects. Table 10
308 shows the multiple liner regression model of Tn depending on the direction of shaking.
309 Among the three directions, the stream dataset shows the smallest standard deviation. The
310 regression results show that the fundamental period of a dam increases with increasing dam
311 height and ap.f.

312 It is important to note that we analyzed 60 dams during 54 earthquakes. This means that
313 some of the dams are being analyzed for more than one earthquake or vice-versa, which
314 implies that some of the uncertainties seen in the proposed relationships are due to motion-to-
315 motion variability in the response of dams, whereas other uncertainty is also due to the
316 differences in response from dam-to-dam variability.

317 The obtained regression model in Table 10 was further analyzed by adding a random
318 effect between each dam, as shown in Table 11. This statistical approach evaluates the dam-
319 to-dam variability by subdividing the residuals into two types: between-dam residuals () and
320 within-dam residuals (). This approach has been applied to various models, including
321 predictions of ground-motion (e.g., Abrahamson et al. 2014) and shear-wave velocity (e.g.,
322 Brandenberg et al. 2010). i denotes the residual depending on ith dam site, where it follows
323 normal distributions with the standard deviation of . ij denotes the residuals of the jth
324 measurements from the ith dam site with the standard deviation of . The obtained parameters
325 shown in Table 11 are similar to those in Table 10, which implies that the dataset is relatively
326 stable for the analyzed mixed-effect model. Total variance increases by 5–30% from Table

12
327 10 to Table 11 by assigning the dam site as a random effect. The variances of 2 in Table 11
328 are smaller by approximately 40–70% for all three directions compared with those in the
329 fixed model. This highlights the importance of site-specific studies to understand the seismic
330 behavior of embankment dams.

331

332 Table 10. Regression model of Tn as a function of dam height and ap.f for embankment dams
Direction Dam type ln(Tn) = a + b ln(H) + c ln(ap.f)
a b c 
Dam axis E+R -2.272 0.396 0.148 0.451
Stream E+R -1.971 0.404 0.205 0.333
Vertical E+R -2.332 0.258 0.106 0.448
333
334
335 Table 11. Mixed-effect regression models of Tn as a function of dam height and ap.f for embankment
336 dams
Direction Dam type ln(Tn)ij = a + b ln(H) + c ln(ap.f) + i + ij
a b c  
Dam axis E+R -2.215 0.397 0.173 0.419 0.299
Stream E+R -2.321 0.438 0.145 0.335 0.176
Vertical E+R -2.886 0.415 0.149 0.292 0.351
337
338

339 DISCUSSION

340 It is useful to develop prediction models of embankment dams by using the recoded
341 ground-motion database of downhole arrays. However, several limitations were encountered
342 in developing the models. The first limitation concerns large uncertainties in the nonlinear
343 response of embankment dams (i.e., Figs. 3 and 8), due to the limited amount of data at
344 higher PGA. The second limitation concerns large uncertainties in the responses of specific
345 dams (i.e., Table 11), because of the lack of information regarding each dam's construction
346 history and material properties. Therefore, further uncertainties in dam responses are
347 expected, as the dams’ differing geometries and construction histories are not included in the
348 analyzed database. Moreover, the analyzed database was obtained in Japan; therefore, the
349 regression models developed in this study cannot cover the regional variability associated
350 with regional geological features such as foundation rock types and ground-motion
351 characteristics of high-frequency contents. Thus, it is useful to discuss the applicability and

13
352 limitations of the developed models if seeking to apply them to geographical areas beyond
353 those covered in the present database.

354 Dam height in the present database ranges from 24 m to 176 m (median 51.7 m), and
355 construction year ranges from 1914 to 2013 (median 1985). One point of note is that the
356 dominant dam construction period spans the 1970s to 1990s (25% quartile and 75% quartile
357 completion years are 1976 and 1995 respectively). Historically, since the 1960s, modern dam
358 types have been constructed using appropriate compaction methods and modern construction
359 equipment (Wiltshire 2002). Thus, we believe that most of the embankment dams in the
360 database were constructed using modern equipment and compaction methods compatible
361 with international best practice. Thus, it can be inferred that the proposed model can be
362 extended to other dams of similar ages and heights. However, it is clear that the collection of
363 additional strong-motion data is necessary in order to update the proposed models in the
364 future, especially for areas outside Japan. Although significant effort has been made to
365 analyze the given database, the study clearly shows that site-specific conditions strongly
366 influence the dynamic response of dams. Thus, the proposed model should only be used for
367 the initial screening stage or for reference checking of numerical modeling.

368 Throughout the paper, a total of 8 correlations are presented – (1) ap.c – ap.f (Table 3), (2)
369 AIc/AIf – ap.f (Table 4), (3) Tn – H (Table 5), (4) Tn/H – L/H (Table 7), (5) Tn – H and L
370 (Table 8), (6) Tn – ap.f (Table 9), (7) Tn – H and ap.f (fixed model in Table 10), and (8) Tn – H
371 and ap.f (mixed model in Table 11). All these correlations can be condensed into two primary
372 correlations – (1) how much seismic amplification would be (Table 3 and 4), and (2) how
373 fundamental period can be correlated with dam geometry and/or seismic intensity (the other
374 Tables). All different correlations are academically and practically meaningful in that the
375 relationships can likely be useful as a first estimate of seismic parameters for dams where
376 other information is not available. It should be noted that there is a considerable amount of
377 scattering in data.

378 It will be more valuable to present a couple representative correlations with less
379 uncertainty to the profession practically. In that sense, fitting models of ap.c – ap.f (Table 3)
380 and Tn – H, ap.f (mixed model in Table 11) are preferred to first use in industry. Table 3
381 shows a straightforward correlation of peak acceleration between dam foundation and crest
382 with less uncertainty. Table 11 shows the desirable empirical correlation of fundamental
383 period of dams as a function of dam height and PGA with dam-to-dam variability.

14
384 In spite of large uncertainty, the promising point is that the proposed models are based on
385 a long time series of 34 years (1978 to 2012). To the best of our knowledge, this is the first
386 study developed for the practical engineering community based on such a comprehensive
387 collection of strong-motion data. Hence, this study using the database helps extend existing
388 knowledge. More thorough analyses are planned for future studies using the same database,
389 in order to assess its reliability and applicability.

390

391 CONCLUSION

392 This study analyses a strong-motion database of embankment dams published by JCOLD
393 (2014). The database includes 190 pairs of three-component time series from dam crests and
394 foundations obtained from 54 earthquakes and 60 embankment dams recorded from 1978 to
395 2012. The paper presents multiple linear regression models that characterize the seismic
396 responses of dam by investigating the correlations with ground motion parameters such as
397 ap.f.

398 The results show that the ap.c/ap.f decreases as ap.f increases. This trend becomes more
399 apparent for ap.f values >0.2 g. ap.c/ap.f becomes <1.0 when ap.f is >0.4 g. In contrast, the ratio
400 of AIc/AIf shows that there is little de-amplification of AI even though ap.f is as large as 1.0 g.
401 It is also observed that Tn increases as H, L, and ap.f increase. The proposed model fits the
402 database better than those of previous studies. The median value of Tn,h/Tn,v is about 1.8.
403 Tn/H increases as L/H increases. Multiple linear regression models show that Tn is reasonably
404 predicted by H, L, and ap.f for embankment dams. Mixed-effect models show that the
405 prediction variance is reduced by 40–70% considering dam-specific variation. This
406 observation indicates that it is useful to investigate specific-dam responses in greater detail in
407 order to reduce these uncertainties.

408 From a practical perspective, the proposed prediction equations will be useful for
409 estimating or evaluating the seismic responses of dams at the preliminary design stage. More
410 specifically, seismic hazard analysis will predict the PGA of the dam foundation (i.e., ap.f)
411 and not the dam crest (i.e., ap.c). Thus, in practice it would be helpful to know how ap.c is
412 related to ap.f (Armstrong 2017). Furthermore, it is helpful to know the fundamental period of
413 the embankment in selecting the appropriate spectral period for seismic hazard analysis (e.g.,
414 Newmark-type prediction equations; Bray and Travasarou 2007). This study also presents the

15
415 prediction model of Tn of embankment dams, including the influence of nonlinear responses.
416 These models are useful in checking whether the dynamic analysis of a given numerical
417 model is performed appropriately with nonlinear effects regarding variations in Tn and ap.c/ap.f.
418 This study also proposed models of Tn as a function of input motion and the H and L
419 geometries. These models will be referenced when the influences of both H and L on Tn are
420 of interest for which the univariate regression models proposed in previous studies may not
421 appropriate by introducing some biases due to the model limitations.

422 As a final remark, this study shows that the seismic response of embankment dams
423 depends on several factors including dam geometry (e.g., dam height and crest length) and
424 the characteristics of input motion. However, the uncertainties between embankment dams
425 are very large from mixed-effect models that are not characterized by their geometry. This
426 indicates that other uncertainties such as the material properties of embankment dams will
427 play a significant role, as previously discussed by the authors (Park and Kishida 2018a;
428 2018b).

429

430 ACKNOWLEDGMENTS

431 The acceleration time histories used for the analyses in this study were obtained from the
432 “Acceleration records on dams and foundations No. 3,” published by the Japan Commission
433 on Large Dams in 2014. The authors recognize their pioneering work and thank the JCOLD,
434 especially Dr. Takashi Sasaki and Dr. Norihisa Matsumoto, for their contribution and for
435 permitting the use of the data. Anonymous reviewers provided valuable review comments
436 that are greatly appreciated. The authors are grateful to Dr. Richard Armstrong at California
437 State University, Sacramento for our helpful discussions and his sincere review of the
438 manuscript.

439 GLOSSARY

440 Each parameter used in this paper is defined below.


441 M : moment magnitude of earthquake
442 Re : epicentral distance
443 E : earthfill-type dam
444 R : rockfill-type dam
445 H : height of dam

16
446 L : crest length of dam
447 PGA : peak ground acceleration (g)
448 ap.f : peak acceleration at the dam foundation (in g) (= PGA in this study)
449 ap.c : peak acceleration at the dam crest (in g)
450 ap.r : the ratio of peak acceleration at the dam crest and the foundation
451 AIf : Arias Intensity of acceleration time history at the dam foundation (in m/s)
452 AIc : Arias Intensity of acceleration time history at the dam crest (in m/s)
453 AIr : ratio of Arias Intensity at the dam crest and the foundation
454 Tn.fft : FFT-based fundamental period of the dam embankment (in sec)
455 Tn.h : fundamental period in the horizontal direction
456 Tn.v : fundamentoal period in the vertical direction
457 Min : minimum value of given data
458 Max : maximum value of given data
459 Med : median value of given data
460 Mean : sum of the non-missing values divided by the number of values
461 STD () : standard deviation of given data, indicating the spread of a distribution around the
462 mean. Square root of the sample variance
463 STD Err Mean : value dividing the sample standard deviation by the square root of number of
464 values
465 R2 : estimated proportion of variation in the response that can be attributed to the model rather
466 than to random error. The model with R-square value closest to 1 achieves a better fit
467

468 REFERENCES

469 Abrahamson, N. A., Silva, W. J. and Kamai, R. (2014) Summary of the ASK14 Ground
470 Motion Relation for Active Crustal Regions. Earthquake Spectra: August 2014, Vol.
471 30, No. 3, pp. 1025-1055.
472 Abdel-Ghaffar, A. M., and Scott, R. F., 1979. Analysis of earth dam response to earthquakes,
473 Journal of the Geotechnical Engineering Division 105, 1379–1404.
474 Ahlberg, J. E., Fowler, J., and Heller, L. W., 1972. Earthquake Resistance of Earth and
475 Rock-fill Dams: Analysis of Response of Rifle Gap Dam to Project RULISON
476 Underground Nuclear Detonation, Report 2.
477 Albano, M., Modoni, G., Croce, P., and Russo, G., 2015. Assessment of the seismic
478 performance of a bituminous faced rockfill dam, Soil Dynamics and Earthquake
479 Engineering 75, 183–198.
480 Ancheta, T. D., Darragh, R. B., Stewart, J. P., Seyhan, E., Silva, W. J., Chiou, B. S.,
481 Wooddell, K. E., Graves, R. W., Kottke, A. R., Boore, D. M., Kishida, T., and
482 Donahue, J. L., 2013. NGA-West 2 database, Earthquake Spectra 30, 989–1005.
483 Armstrong, R. J., 2017. Personal Communication.
484 Boore, D. M., 2008. TSPP—A Collection of FORTRAN Programs for Processing and
485 Manipulating Time Series, U.S. Geological Survey Open-File Report 2008-1111,
486 Version 4.4, 04 September 2013.

17
487 Brandenberg, S. J., Bellana, N., and Shantz, T., 2010. Shear wave velocity as function of
488 standard penetration test resistance and vertical effective stress at California bridge
489 sites, Soil Dynamics and Earthquake Engineering 30, 1026–1035.
490 Bray, J. D., and Travasarou, T., 2007. Simplified procedure for estimating earthquake-
491 induced deviatoric slope displacements, Journal of Geotechnical and
492 Geoenvironmental Engineering 133, 381–392.
493 Chiou, B., Darragh, R., Gregor, N., and Silva, W., 2008. NGA project strong-motion
494 database, Earthquake Spectra 24, 23–44.
495 Chugh, A. K., 1985. Estimation of the fundamental period of large earthfill and rockfill dams,
496 Soils and Foundations 25, 146–150.
497 Gazetas, G., 1987. Seismic response of earth dams: some recent developments, Soil
498 Dynamics and Earthquake Engineering, 6, 2–47.
499 Gazetas, G., and Dakoulas, P., 1992. Seismic analysis and design of rockfill dams: state-of-
500 the-art, Soil Dynamics and Earthquake Engineering, 11, 27–61.
501 Goulet C. A., Kottke A., Boore D. M., Bozorgnia Y., Hollenback J. Formerly, Kishida T.,
502 Der Kiureghian A., Ktenidou O. J., Kuehn N. M., Rathje E. M., Thompson E. M., and
503 Wang X. Y., 2018. Effective amplitude spectrum (EAS) as a metric for ground
504 motion modeling using Fourier amplitudes, in 2018 SSA-LACSC Conference, 14–17
505 May, 2018, Miami, Florida.
506 Ha, I.-S., 2011. Evaluation for fundamental periods of domestic rockfill dams with micro-
507 earthquake records, Journal of Korean Geo-environmental Society 12, 53–60.
508 ICOLD, 2018. World Register of Dams, International Commission on Large Dams,
509 http://www.icold-cigb.net/GB/world_register/general_synthesis.asp.
510 JCOLD, 2014. Acceleration Records on Dams and Foundations No. 3, Japan Commission on
511 Large Dams, Tokyo, Japan.
512 Ktenidou, O.-J., Chávez-García, F.-J., Raptakis, D., and Pitilakis, K. D., 2016. Directional
513 dependence of site effects observed near a basin edge at Aegion, Greece, Bulletin of
514 Earthquake Engineering 14, 623–645. https://doi.org/10.1007/s10518-015-9843-x
515 Kim, D., Kim, M., Kim, S., and Choo, Y., 2012. Seismic behaviors of earth-core, concrete-
516 faced-rock-fill, and composite dams, in Eurofuge 2012, 23–24 April, 2012, Delft, The
517 Netherlands.
518 Kim, M.-K., Lee, S.-H., Choo, Y. W., and Kim, D.-S., 2011. Seismic behaviors of earth-core
519 and concrete-faced rock-fill dams by dynamic centrifuge tests, Soil Dynamics and
520 Earthquake Engineering 31, 1579–1593.
521 Kim, N.-R., and Ha, I.-S., 2012. Evaluation of fundamental period of rockfill dam using
522 blasting vibration test, Journal of the Korean Society of Civil Engineers 32, 185–192.
523 Kishida, T., Ktenidou, O.-J., Darragh, R. B., and Silva, W. J., 2016. Semi-Automated
524 Procedure for Windowing Time Series and Computing Fourier Amplitude Spectra for
525 the NGA-West2 Database, Pacific Earthquake Engineering Research Center, PEER
526 Report 2016/02.
527 Kishida, T., 2017. Conditional mean spectra given a vector of spectral accelerations,
528 Earthquake Spectra, 33, 469–479.
529 Kong, X.-J., Zhou, Y., Xu, B., and Zou, D.-G., 2010. Analysis on seismic failure mechanism
530 of zipingpu dam and several reflections of aseismic design for high rock-fill dam,
531 Earth and Space, 3177–3189.
532 Konno, K., and Ohmachi, T., 1998. Ground-motion characteristics estimated from spectral
533 ratio between horizontal and vertical components of microtremor, Bulletin of the
534 Seismological Society of America 88, 228–241.

18
535 Laurendeau, A., Bard, P. Y., Hollender, F., Perron, V., Foundotos, L., Ktenidou, O.-J., and
536 Hernandez, B., 2017. Derivation of consistent hard rock (1000 < VS < 3000 m/s)
537 GMPEs from surface and down-hole recordings: analysis of KiK-net data, Bulletin of
538 Earthquake Engineering. https://doi.org/10.1007/s10518-017-0142-6
539 Makdisi, F. I., and Seed, H. B., 1979. Simplified procedure for evaluating embankment
540 response, Journal of the Geotechnical Engineering Division 105, 1427–1434.
541 Mejia, L. H., Seed, H. B., and Lysmer, J., 1982. Dynamic analysis of earth dams in three
542 dimensions, Journal of the Geotechnical Engineering Division 108, 1586–1604.
543 Oner, M., 1984. Estimation of the fundamental period of large earthfill and rockfill dams,
544 Soils and Foundations 24, 1–10.
545 Papadimitriou, A. G., Bouckovalas, G. D., and Andrianopoulos, K. I., 2014. Methodology for
546 estimating seismic coefficients for performance-based design of earthdams and tall
547 embankments, Soil Dynamics and Earthquake Engineering 56, 57–73.
548 Park, D. S., 2018. Fundamental period of embankment dams based on strong motion records,
549 in 2018 USSD Annual Conference and Exhibition, April 29, 2018.
550 Park, D. S., and Kim, N.-R., 2017. Safety evaluation of cored rockfill dams under high
551 seismicity using dynamic centrifuge modeling, Soil Dynamics and Earthquake
552 Engineering 97, 345–363.
553 Park, D. S., and Kishida, T., 2018a. Shear wave velocity profiles of fill dams, Soil Dynamics
554 and Earthquake Engineering 104, 250–258.
555 Park, D. S., and Kishida, T., 2018b. Shear modulus reduction and damping ratio curves for
556 earth core materials of dams, Canadian Geotechnical Journal,
557 https://doi.org/10.1139/cgj-2017-0529.
558 Perron, V., Hollender, F., Bard, P.‐Y., Gélis, C., Guyonnet‐Benaize, C., Hernandez, B., and
559 Ktenidou, O.‐J., 2017. Robustness of kappa (κ) measurement in low‐to‐moderate
560 seismicity areas: insight from a site‐specific study in Provence, France, Bulletin of the
561 Seismological Society of America 107, 2272–2292.
562 Sasaki, T., 2015. Tentative report — analysis on acceleration data of dams collected by
563 JCOLD, in ICOLD 25th Congress / 83rd Annual Meeting, Technical Committee
564 Meeting, 13–20 June, 2015, Stavanger, Norway.
565 Seyhan, E., and Stewart, J. P., 2014. Semi-empirical nonlinear site amplification from NGA-
566 West2 data and simulations, Earthquake Spectra, 30, 1241–1256.
567 Uddin, N., and Gazetas, G., 1995. Dynamic response of concrete-faced rockfill dams to
568 strong seismic excitation, Journal of Geotechnical Engineering 121, 185–197.
569 USSD, 2014. Observed performance of dams during earthquakes, Volume III, United States
570 Society on Dams, Denver, CO, USA.
571 Wieland, M., 2012. Seismic design and performance criteria for large storage dams, in 15th
572 World Conference on Earthquake Engineering, 24–28 September, 2012, Lisboa,
573 Portugal.
574 Wiltshire, R. L., 2002. 100 years of embankment dam design and construction in the US, in
575 Bureau of Reclamation History Symposium, 18–19 June, 2002, Las Vegas, pp. 1–68.
576 Woolery, E., Street, R., and Hart, P., 2009. Evaluation of linear site-response methods for
577 estimating higher-frequency (>2Hz) ground motions in the Lower Wabash River
578 Valley of the Central United States, Seismological Research Letters 80, 525–538.
579 Yuan, L., Liu, X., Wang, X., Yang, Y., and Yang, Z., 2014. Seismic performance of earth-
580 core and concrete-faced rock-fill dams by large-scale shaking table tests, Soil
581 Dynamics and Earthquake Engineering 56, 1–12.
582

19
View publication stats

You might also like