Professional Documents
Culture Documents
10, 5 9 4 8 (1982)
SUMMARY
A statistical representation of seismic liquefaction is advanced based on the postulate that pore water pressure increases are
proportional to the dissipated seismic energy density. The representation, based on approximately fifty case histories,
relates the pore pressure increase to earthquake magnitude, distance to centre of energy release, initial effective overburden
stressand standard penetration value. The model may be used for analysisofseismic liquefaction risk. An example analysis
for the ‘South of Market Zone’ in San Francisco is carried out in relation to earthquakes on the San Andreas fault.
INTRODUCTION
Assessment of liquefaction potential for sandy soils has attracted considerable attention in recent times.
Significant amounts of damage have been directly or indirectly attributed to liquefaction phenomena in
several recent earthquakes (e.g. Niigata, 1964; Alaska, 1964; Tokachi-oki, 1968) as well as inferred in many
historical earthquakes. A number of techniques have been developed to estimate the likelihood that
liquefaction may or may not occur at a given site under specified conditions. These techniques generally
attempt to relate the soil liquefaction potential to parameters connected with, or a complete description of, the
expected ground motion.
Two general classes ofanalysis may be defined. In the first, rigorous testing and analysis of soil properties are
carried out and the resulting soil description is combined with extensive calculations of expected ground
responseqcf. Seed and Idriss’). This type ofanalysis is both costly and time consuming but would be applicable
in cases where considerable importance is attached to the consequences of liquefaction, such as with a nuclear
power project. The second class of analysis involves simplified procedures, often assessing soil properties only
through standard penetration values and incorporating the expected ground motion through peak ground
accelerations.2 This type of analysis would be applicable in cases where less serious consequences of
liquefaction are expected.
Both classes of analysis depend upon the basic assumption that development of excess pore pressure within
the soil mass is directly related to the shear stress generated by shaking. In the more complex class of analyses,
the investigator must infer from a time history of expected ground motion, an equivalent uniform stress level
and an equivalent number of cycles of stress reversal at this stress level. Empirical techniques have been
advanced to assist with this inference. In simplified analyses, the equivalent stress level is inferred from the
expected peak ground acceleration, also by empirical methods.
The development which we present here falls into the class of simplified analyses. We do not employ the
assumption that pore pressure increases depend upon shear stresses generated by the ground shaking. Rather,
we follow the assumption of Nemat-Nasser and Shokooh3 that the pore pressure is directly related to the
amount of seismic energy dissipated in the soil. Briefly, we do the following: The energy content of the seismic
waves arriving at the site is estimated from the total radiated energy of the earthquake and the distance to the
centre of energy release. The total radiated energy is related to the earthquake magnitude via the Gutenberg
and Richter relati~nship.~ The portion of arriving energy which is dissipated at the site is assumed to depend
upon the standard penetration value and the initial effective overburden stress. We determine the relationship
between standard penetration value and energy dissipation by examination of historical records of earthquake
0098-8847/82/0 10059-1 0$01.OO Received 8 April 1980
@ 1982 by John Wiley & Sons, Ltd Revised 23 December 1980
60 R. 0. DAVIS AND J. B. BERRILL
induced liquefaction. Once the dissipated energy has been estimated, the increase in pore pressure follows
directly. In this way we obtain a simple relationship giving the increase in pore pressure as a function of(i) the
earthquake magnitude, (ii) its distance from the site, (iii) the standard penetration value and (iv) the initial
effective overburden stress. The resulting equation is well suited for studies of seismic liquefaction risk. We
illustrate this with an example.
The development we present below is restricted to apply to fully saturated sands. At a certain point
[equation (lo)] we obtain a relationship between soil dissipation properties and standard penetration value
which is based almost entirely on historical data involving fine sands. Since the relationship is crucial to our
analysis, it should not be extended to other soil types.
A development closely related to ours has been advanced by Yegian and Whitman.’ They make no
assumption concerning the mechanism of liquefaction, however. Instead they perform a purely statistical
analysis of magnitude, hypocentral distance and standard penetration data for case histories in which
liquefaction may or may not have occurred.
DEVELOPMENT OF MODEL
Our ultimate aim is to predict the increase in pore pressure at a given site for a specified earthquake given the
earthquake magnitude, hypocentral distance, initial effective overburden stress and standard penetration
value. We can formulate the problem as follows: First, given the earthquake magnitude M ,the total radiated
energy E , is4
E , = 101,sM+1.8
(1)
where E , has dimensions of kJ. Only a fraction of this energy will be contained in seismic waves reaching the
site. Some energy will be dissipated by anelastic attenuation along ray paths, and further attenuation occurs
due to geometric spreading. The most simple energy attenuation model for spherically expanding waves is an
inverse relationship to distance squared. Thus, letting r denote the distance to the centre of energy release, the
energy content of the seismic waves arriving at the site is approximated by
E(r) = c1 E,/r2 (2)
where c1 is a constant. If r is large, we can effectively ignore geometric spreading at the site itself (i.e. the waves
are approximately plane) so that E(r)is an energy density, kJ/m3, provided r is measured in metres and c1 has
dimensions of m- ’.
Part of the arriving energy density E(r) will be dissipated in the soil at the site. Let the dissipated energy
density be denoted by AE(r). We assume that
AE(r) = A ( N ,6,J E(r) ( 3)
where A is a dissipation function which depends upon the standard penetration value N and the initial effective
overburden stress 6., Extensive laboratory studies by Hardin6 suggest that A should be proportional to 6 o f .
Thus we make the further assumption that
A = 1(N)a,* (4)
where 1 is only a function of N .
Finally, we assume that the increase in pore pressure Au is a linear function of the dissipated energy density.
AU = c2 AE (5)
where c2 is a constant. In fact, although Au is expected to be an increasing function of AE, it will be bounded
above, and as a complete liquefaction condition is approached, equation ( 5 ) will be less and less accurate.’
Nevertheless, this expression will approximate the pore pressure-dissipated energy density response up to
liquefaction, and, for our purposes, its simplicity outweighs its disadvantage in precision. Combining
equations (1HS)we now have
ENERGY DISSIPATION AND SEISMIC LIQUEFACTION IN SANDS 61
then liquefaction should occur. Conversely, if C ( N )is smaller than this value, liquefaction should not occur.
This suggests that we plot r2 O;l2 10-1'5M versus N for actual cases where earthquake-induced liquefaction
may or may not have occurred. Points corresponding to non-liquefaction should lie above C ( N )while points
corresponding to liquefaction should lie below.
We have followed this procedure with results shown in Figure 1. Rather than employ raw standard
penetration results, we have corrected all N values for overburden pressures using the formula of Peck et a1.'
where 6, has dimensions ofkPa. The data shown in Figure 1 are summarized in Table I. In all, 57 case histories
are represented, drawn from records of seventeen different earthquakes.
Some scattering or intermingling of liquefied and non-liquefied data points is evident in Figure 1.
Nevertheless, the general trend of non-liquefied points lying higher and to the right is quite clear. In fact, the
degree of scatter is surprisingly small considering the variety of sources of error inherent in the model. The
major sources of scatter are listed below:
(i) We have taken no account of the radiation pattern or directivity effects associated with the individual
earthquakes. Indeed, this would be possible in only a few of the most recent earthquakes for which source
models have been formulated. The actual pattern of radiated energy in many earthquakes will not be
geometrically uniform.
(ii) In many cases the exact location of the centre of energy release is unknown, and hypocentral or
epicentral distances have been used. For small earthquakes the effect will be minimal, but for large earthquakes
the hypocentre may lie tens or even hundreds of kilometres distant from the centre of energy release.
(iii) For older earthquakes, magnitudes are at best poorly determined. Also, seismograph saturation in very
large earthquakes may lead to underestimates of magnitudes.
(iv) Energy attenuation proportional to r - 2 may not be a good approximation for small values of r. In the
very near source region, the high frequency energy content will be large, and anelastic material attenuation
may be as significant or more significant than geometric attenuation.
(v) Almost all initial overburden stresses have been calculated from assumed soil densities. For soil above
the water table we have assumed a density of 1.70 x lo3kg/m3, for soil below the water table, 1.95 x lo3kg/m3.
In many cases the depth to the water table at the time of the earthquake is unknown, and water table elevations
from later measurements (sometimes decades later) have been used.
(vi) Standard penetration values have, in the majority of cases, resulted from measurements after the
earthquake and these may have been significantly altered from pre-earthquake values. In some cases, the
standard penetration values are obtained from borings near to, but not directly on, the sites in question.
62 R. 0.DAVIS AND J. B. BERRILL
Figure 1. Liquefaction case histories. Open circles denote cases where liquefaction occurred, closed triangles where it did not
00 r Lique-
Case Earthquake Year Site M (kPa) (km) r2S:'210-1'5M N N faction? Refs
Table I (cont.)
C,, r Lique-
Case Earthquake Year Site M (kPa) (km) rZbi'Z N N faction? Refs
(vii) Some of the data refer to sites which were not level ground or which were subjected to external loads or
both.
Of these sources oferror, the first is probably the most important overall. It will affect all the data, both for
large and small earthquakes and for large and small values of r. Its effectshould be entirely random. Errors in r
should also be random, but would be more important in those cases involving smaller earthquakes closer to
the affected sites. The fourth source of error will be especially important at sites very near the earthquake and
may tend to bias the data in a conservative manner. Errors in ii0 should be roughly random and probably
small, at least in comparison with directivity and attenuation pattern effects. This is also the case with errors in
standard penetration values. The last source oferror, external loads or sloping ground, should bias the data in
a conservative manner.
Two data points are particularly bothersome in that they appear to contradict the general trend of the
remaining data. These are cases 41 and 57, the 1957 San Francisco earthquake which induced liquefaction at
Lake Merced and the 1971 San Fernando earthquake which caused liquefaction at the Jensen Filtration Plant.
In the case of the Jensen Plant, it appears clear that the major portion of the earthquake energy was directed
towards the site from the rupture." Also, the site was unique among all the data, having a very large initial
overburden stress. Both the Jensen Plant and Lake Merced sites were located quite close to the respective
earthquakes, suggesting that the assumed r - 2 attenuation may have been in error. Both sites were not level
64 R. 0. DAVIS AND J. B. BERRILL
ground, and the 1957 San Francisco earthquake is by far the smallest magnitude earthquake in the data set.
Because of these reasons, we have elected to omit these two case histories from further consideration.
In order to partition the data of Figure 1 we have used a simple discriminant analysis.I9*2o A similar analysis
was used by Christian and Swiger” to partition ground acceleration versus relative density data. The analysis
is based on all the data in Table I with the exception of cases 41 and 57. The resulting expression for C ( N )is
given by
C ( N ) = 450N-2 (10)
The corresponding line partitioning the liquefied and non-liquefied data is shown on Figure 1. Note that C ( N )
is dependent upon dimensions. For the SI units we employ, equation (10) has dimensions of kN3” m- I .
Using equation (10) in equation (6) we may now write
AU = 450 101.5M
r2 N 2 46,
where r is in metres and 6, and Au are in kPa. This expression is the main result of our development. It follows
from the assumed dependence of pore pressure on dissipated energy, simple assumptions regarding seismic
energy release and attenuation and statistical analysis of historical data.
It is possible through the discriminant analysis to calculate the probability that any new data will be
misclassified by equation (10). That is, based on the scatter in the existing data, we may determine the
probability that a new case classified by equation (10)as non-liquefied may in fact have exhibited liquefaction.
This probability is 165 per cent. Considering the sources of scatter described above, this number intuitively
seems reasonable.
It is also interesting to note that in a study of Japanese earthquakes Kuribayashi and T a t s u ~ k a have
’~
suggested a relationship between earthquake magnitude and the maximum distance to any liquefied site.
Letting R denote the distance in metres between the earthquake and the most distant site at which liquefaction
may occur, they suggest that
IOg,, R = 077M -06 (121
for M equal to or greater than 6.0. We note that for given soil parameters N and a,, equation (11) will predict
the maximum possible distance to a site of liquefaction provided we set Au equal to 6, (indicating complete
liquefaction). Rearranging terms, we have
Although a comparison between the constant 0.6 in equation (12) and the second term on the right-hand side of
equation (13) is inappropriate (since Kuribayashi and Tatsuoka do not consider soil properties),it is clear from
the coefficients of M in these two equations that the two theories share a similar energy attenuation
relationship.
Of all the sources oferror outlined above, the most bothersome is probably the complex attenuation pattern
expected in regions very near the earthquake. In order to consider this source of error, we may re-plot the data
of Table I eliminatingall cases for which r is less than or equal to 20 km. The resulting graph is shown in Figure
2, together with the proposed relationship for C(N)given by equation (10).The separation of liquefied and non-
liquefied data in this plot is remarkably clear. Only two cases, 18 and 23, are misclassified, and both by only
small margins.
focus may be easily incorporated in the risk analysis by methods outlined in Cornell’s paper.” We employ the
simplest configuration here only to illustrate the analysis. Let x denote the distance along the fault measured
from its intersection with the normal from the site. Then if an earthquake occurs at point x, the probability of
liquefaction at the site is
The right-hand side of this expression follows from equation (6) with Au > ~ 7 Note~ . that any fraction of 6,
could be used, for example we could calculate the probability that Au equals or exceeds a6, for any a between 0
and 1.0. The hypocentral distance r is given by
r2 = x 2 + d 2
I
where d is the normal distance from site to fault, and P [ A B] means the probability of A given B.
The annual number of earthquakes n for which M exceeds agiven magnitude m may be estimated from the
Gutenberg and Richter r e l a t i o n ~ h i p ~ ~
log,, n = log,, T- bm
66 R. 0.DAVIS AND J. B. BERRILL
where Trepresents the total annual number ofearthquakes with positive magnitude and b is a constant related
to local seismicity. From this equation we have
n
P[M>rn] = - =
T
Using this result in equation (14) gives
The cumulative probability of liquefaction results if we integrate over the length of the fault,
where a uniform probability of earthquake occurrence is assumed along the length of the fault. Cornell has
shown22 that this integral may be approximated by
where
y=$b-l
provided the fault is very long compared with d. Finally, assuming the occurrence of earthquakes along the
fault is a Poisson process, the annual probability of liquefaction PLat the site is given by
P, = l - e x p ( - 7 P [ A ~ > 6 ~ ] ) (17)
Using equation (16) in equation (17) gives
where
v = T/L
represents the number of positive magnitude earthquakes per year per unit length of fault.
Given the fault geometry and seismicity together with the standard penetration value and soil profile for the
site, we may use equation (18) to estimate the annual probability of liquefaction. We can outline a typical
calculation by considering the San Andreas fault as it affects the ‘South of Market’ site described by Youd and
Hoose.” For this site d is approximately 14km. At a soil depth of 4.5 m the value of Oo is about 50 kPa and the
corrected standard penetration value is 8.7. This corresponds to a value of C(N) of about 5.95. Seismicity
parameters may be taken from the article by Ryall et to obtain v = 0.0564 earthquakes per year per metre
length of fault, and b = 0.90. Using these data in equation (18) we find that
P, = 0.00812
CONCLUSIONS
The method of liquefaction analysis presented here is appealing due to its simplicity and to its reliance on
fundamental earthquake parameters magnitude and distance to centre of energy release rather than peak
ground acceleration. The entire analysis hinges on the assumption that pore pressure depends only upon
dissipated energy. This assumption seems well supported by the analysis of Nemat-Nasser and Shokooh3 and
further support may be inferred from the evident separation of liquefied and non-liquefied data in Figures 1
and 2.
Some qualifications on the use of the model are required. It is apparent that for small values of r our assumed
energy attenuation relationship may be in error. The model should be used with caution for sites very near
potential earthquake sources. Also, we note the scarcity of data in Table I for N values in excess of 20. To
extrapolate the technique to sites with corrected standard penetrations greater than about 20 is clearly
unwarranted. Finally we point out that nearly all of the soils in the case histories we have used are at relatively
shallow depths. With the exception of case 57, the initial overburden values are smaller than 135 kPa. Thus
caution in applying the technique to deep sand strata is advised.
ACKNOWLEDGEMENTS
Thanks are gwen to K. J. McLay and G. J. Cameron for help in collecting and processing liquefaction case
history data.
Added note
In reviewing the original manuscript for this paper, T. Leslie Youd has suggested that distance to centre of
energy release r may not be the most appropriate measure of distance for defining geometric attenuation in
equation (2). Instead, he suggested the minimum distance between the site and the actual rupture surface be
used. This dimension has considerable appeal, especially in the case of a long, narrow rupture such as San
Francisco 1906. Unfortunately, its use in the development presented above is not possible, since in only a very
few of the case histories listed in Table I has the rupture surface been clearly identified. As pointed out above,
the assumption of a spherically radiating point source is probably the greatest potential source for error in our
development. Nevertheless, until more data become available, the distance to centre of energy release appears
to offer the best measure for definition of geometric energy attenuation.
REFERENCES
1. H. B. Seed and I. M. Idriss, ‘Analysis of soil liquefaction: Niigata earthquake’, J . Soil Mech. Found. Diu., ASCE 94, 1053-1122
( 1968).
2. H. B. Seed and I. M. Idriss, ‘A simplified procedure for evaluating soil liquefaction potential’, Report No. EERC 70-9, University of
California, Berkeley (1970).
3. S. Nemat-Nasser and A. Shokooh, ‘A unified approach to densification and liquefaction of cohesionless sand in cyclic shearing’,
Canadian Geotech. J . 16, 6 5 M 7 8 (1979).
4. B. Gutenberg and C. F. Richter, ‘Magnitude and energy of earthquakes’, Ann. Geofis. 9, 1-15 (1956).
5. M. K. Yegian and R. V.Whitman, ‘Risk analysis for ground failure by liquefaction’,J. Geotech. Eng. Din, ASCE 104,921-938 (1978).
6. B. 0. Hardin, ‘The nature of damping in sands’, J. Soil Mech. Found. Div., ASCE 91, SMl, 63-97 (1965).
7. J. H. Schmertmann, ‘Energy dynamics of SPT, J . Geotech. Eng. Din, ASCE 105, 909-926 (1979).
8. R. B. Peck, W. E. Hanson and T. H. Thornburn, Foundation Engineering, Wiley, New York, 1974.
9. H. Kishida, ‘Characteristics of liquefied sands during Mino-Owari, Tohnankai, and Fukui earthquakes’, Soils Found. 9,75-92 (1969).
10. K. Ishihara, ‘Liquefaction of subsurface soils during earthquakes’, Technocrat 7, 8 1-98 (1974).
11. E. Kuribayashi, T. Iwasaki and F. Tatsuoka, ‘A history of soil liquefaction in Japan’, 6th World Conf: Earthq. Engng, New Delhi, pp.
2448-2454 (1977).
12. T. L. Youd and S. N. Hoose, ‘Liquefaction during 1906 San Francisco earthquake’, J. Geotech. Eng. Diu., ASCE 102,425-440
(1976).
13. E. Kuribayashi and F. Tatsuoka, ‘Brief review of liquefaction during earthquakes in Japan’, Soils Found. IS,81-92 (1975).
14. Y. Ohsaki, ‘Effects of sand compaction on liquefaction during the Tokachioki earthquake’, Soils Found. 10, 112-128 (1970).
15. H. Kishida, ‘Characteristics ofliquefaction oflevel sandy ground during the Tokachi-oki earthquake’, SoilsFound. 10,10~111(1970).
16. T. Yamonouchi, S. Taneda and T. Kimura, ‘Damage features in the 1968 Ebino earthquakes from the viewpoint of soil engineering’,
Soils Found. 10, 129-144 (1970).
17. S. J. Dixon and J. W. Burke, ‘Liquefaction case history’, J . Soil Mech. Found. Diu. ASCE 99, 921-938 (1973).
68 R. 0. DAVIS AND J. B. BERRILL
18. J. B. Berrill, ‘A study of high frequency strong ground motion from the San Fernando earthquake’, PhD Thesis,California Institute of
Technology, 1975.
19. R. D. Bock, Multivariate Statistical Methods in Behavioral Research, McGraw-Hill, New York, 1975.
20. T. W. Anderson, An Introduction to Multioariate Statistical Analysis, Wiley, New York, 1958.
21. J. T. Christian and W. F. Swiger,‘Statistics of liquefaction and SPT results’, J . Geotech. Eng. Div., ASCE 101,1135-1 150( 1975).
22. C. A. Cornell, ‘Engineering seismic risk analysis’, Bull. Seism. Soc. Am. 58, 1583-1606 (1968).
23. B. Gutenberg and C. F. Richter, Seismicity ofthe Earth, Princeton University Press, 1954.
24. A. Ryall, D. B. Slemmons and L. D. Gedney, ‘Seismicity, tectonism, and surface faulting in the western United States during historic
time’, Bull. Seism. SOC.Am. 56, 1105-1135 (1966).