Last Modified December 14, 2009

22. Seismic Risks for Embankments

Key Concepts
There are very few instances where an earthquake has damaged an embankment dam enough to result in the uncontrolled release of reservoir water. Many embankment dams are exposed to earthquake shaking each year, but either the damage caused by the earthquake was not extensive enough, or in the rare cases where damage was extensive, the reservoir was far below the damage and uncontrolled releases did not happen. The failure probability estimation procedures described below are built upon standard analysis techniques used to predict responses of soil to dynamic loading and upon observations from case histories of embankments that have been exposed to earthquakes. Dynamic loading from an earthquake changes the stress states within an embankment, causing permanent damage if the stress changes cause shear or tensile strength to be exceeded. Loose, saturated, cohesionless soils, when subject to earthquake shaking and initial shearing, can contract as the soil particles are rearranged. Since the water within the pore spaces is virtually incompressible, this results in an increase in porewater pressure. If the pore pressure increase is enough to reduce the effective stress to zero, the soil is said to have liquefied and the soil experiences a significant reduction in shear strength. Extensive shear strength reduction beneath an embankment slope can trigger a flow slide which, in turn, can result in a very rapid dam failure. Many cycles of low-amplitude loading can also induce a fatigue-like shear strength loss in dense, saturated, materials, a phenomenon some call 'cyclic mobility'. Whether or not the soil of an embankment or its foundation liquefies completely, pore pressure increases can still result in a decrease in shear resistance. If enough reduction in shear resistance occurs, over a sufficient extent, large deformations can result. A translational failure can occur if the entire foundation beneath an embankment liquefies and the reservoir pushes the embankment downstream far enough to create a gap in the vicinity of an abutment. Overtopping erosion failure can occur if crest deformations exceed the freeboard at the time of the deformations. If the deformations do not result in an immediate release of the reservoir, the embankment can be cracked or disrupted to the point where seepage erosion can occur through the damaged remnant. This failure mechanism can occur with or without liquefaction. There are many ways in which cracking can occur due to seismic shaking, such as differential settlement upon shaking, general disruption of the embankment crest, offset of a foundation fault, or separation at spillway walls. See the section on Internal Erosion and Piping Risks for Embankments for other conditions that may make a particular dam more susceptible to transverse cracking and subsequent seepage erosion.

22-1

Last Modified December 14, 2009 Compacted embankments are typically not considered to be subject to liquefaction upon shaking and initial shearing. Dense, cohesionless soils tend to dilate upon shearing, which increases the pore space between soil particles and reduces the pore pressures. Most Reclamation embankment dams are compacted, so the focus of liquefaction studies tends to be related to loose foundation soils. However, hydraulic fill embankments may be susceptible to liquefaction or pore pressure increases. Fine-grained soils, while not strictly 'liquefiable', may be susceptible to strength loss during an earthquake. Two aspects of a fine-grained soil's shear strength behavior can require investigation: the anticipated peak magnitude of earthquake-induced shear loading when compared to a soil's undrained shear strength determined from monontonic loading, and the potential for a reduction in the undrained shear strength due to the effects of many shearing cycles. If active faults or faults capable of co-seismic displacement cross an embankment dam foundation, the potential exists for foundation displacement that cracks or disrupts the dam core or water retaining element as well as transition zones or filters. The cracking can initiate concentrated seepage, and the translational movement can create locations where there would be unfiltered exit points for the seepage. Both conditions would increase the likelihood for failure from internal erosion or piping. Shearing of a conduit passing through an embankment dam as a result of fault displacement can result in transmission of high pressure water into the dam, leading to increased gradients and potential for internal erosion. Seiche waves can be generated by large fault offsets beneath the reservoir or by regional ground tilting that encompasses the entire reservoir. Sloshing can lead to multiple overtopping waves from these phenomena. Risk Analysis Methodology Appendix J (Reclamation, 2005) provides additional details on these topics.

Seismically-Induced Potential Failure Modes

The following are generic descriptions of how a dam might fail due to these potential failure modes. For a prototype dam, additional details would be needed in the descriptions, as described in the section on Potential Failure Mode Identification, Description, and Screening. Liquefaction and Overtopping Severe earthquake shaking would cause loose embankment or foundation materials to contract under cyclic loading, generating excess pore water pressures (liquefaction occurs). Pore water pressure increases would reduce the soils shear strength. Loss of shear strength over an extensive area would lead to slope instability and crest settlement. Crest deformation would exceed the freeboard existing at the time of the earthquake. The depth and velocity of water flowing over the crest would be sufficient to erode materials covering the downstream slope. Headcutting action would carve channels across the crest. The channels 22-2

Last Modified December 14, 2009 would widen and deepen. Subsequent human activities would not be sufficient to stop the erosion process. The embankment would breach and release the reservoir. Liquefaction and Transverse Cracking at the Crest Severe earthquake shaking would cause loose embankment or foundation materials to contract under cyclic loading, generating excess pore water pressures (liquefaction occurs). Pore water pressure increases would reduce the soils shear strength. Loss of Shear strength over an extensive area would lead to slope instability, deformations, and crest settlement. However, crest deformation would not exceed the freeboard existing at the time of the earthquake. Open and continuous transverse cracks would form across the crest and through all zones of the dam deep enough to intersect the reservoir. Water depth and velocity flowing through open cracks would be sufficient to erode the materials along the sides and across the bottom of the cracks. Material from upstream zones would not be effective in sealing the cracks (by being transported to a downstream zone or constriction point where a filter would begin to form). Headcutting action would carve channels across the crest. The channels would widen and deepen. Subsequent human activities would not be sufficient to stop the erosion process. The embankment would breach and release the reservoir. Liquefaction and Sliding Opening Gaps Severe earthquake shaking would cause loose embankment or foundation materials to contract under cyclic loading, generating excess pore water pressures (liquefaction occurs). Pore water pressure increases would reduce the soils shear strength. Loss of shear strength would occur in a layer that is continuous upstream to downstream. Reservoir loading would exceed the shearing resistance remaining in the layer and the entire embankment would slide downstream. Downstream deformation would open a gap at the crest deep enough to intersect the reservoir. Water depth and velocity flowing through the gap would be sufficient to erode the materials along the sides and across the bottom of the gap. Material from upstream zones would not be effective in sealing the gap (by being transported to a downstream zone or constriction point where a filter would begin to form). Headcutting action would carve channels across the crest. The channels would widen and deepen. Subsequent human activities would not be sufficient to stop the erosion process. The embankment would breach and release the reservoir. Deep Cracking Severe earthquake shaking would cause differential settlement over stiffness discontinuities, at near-vertical embankment/foundation contacts, or at contacts between embankment and concrete. Transverse cracks would form through the core and concentrate seepage flow through the cracks below the reservoir surface would occur. Seepage quantity and velocity would be sufficient to erode core material and transport it beyond the downstream shell material. Upstream zones would not be effective in sealing the cracks (by a mechanism whereby material from upstream zones would be transported to a downstream zone or constriction point where a filter would begin to form). 22-3

Last Modified December 14, 2009

Yes Yes Deformation > Freeboard? Yes No
0

Crack Leads to Failure? No

< 0.1g

Liquefaction? Yes No No Crack Leads to Failure?

0.1g to 0.3g

etc.

Earthquake Event

Earthquake Loading 0.3g to 0.5g > 0.5g etc. etc.

Figure 22-1. Typical Event Tree Model for Dam Failure due to Earthquake Loading

Event Tree
Figure 22-1 shows an event tree typically used when only one slope is potentially unstable (Reclamation dams frequently have their cutoff trenches offset upstream of the centerline which makes the upstream slope more stable than the downstream slope). The first node in Figure 22-1 splits the tree into several branches representing different earthquake loading conditions with selected ranges of peak horizontal acceleration (or other measure of earthquake shaking). The second node further separates situations where liquefaction is believed likely or unlikely for a given a peak horizontal acceleration range. If liquefaction will not take place, crest deformations usually do not lead to total freeboard loss, and the final node in this branch assesses whether or not water flowing through cracks, either at the dam crest or deep within the embankment, can lead to breach formation. If liquefaction does take place, a subsequent node treats conditions where embankment deformations might lead to freeboard loss and failure by overtopping. If in this case the dam is not overtopped, the possibility of failure initiating with flow through cracks is again assessed. Probabilities assigned to events or natural states are multiplied along each branchs pathway, leading to a joint probability for the particular combination of the events or states along that path. Each branch ending in a failure condition contributes to the total failure probability. Consequences are assigned to each failure branch and an annualized probability of life loss is calculated (see also section on Event Trees). Past situations where liquefaction has occurred resulted in significantly more extensively damaged embankments. Therefore, failure modes are analyzed in two categories: where liquefaction does and does not take place.

22-4

Last Modified December 14, 2009 The event tree is rather simple, but complex calculations are made outside the tree and then brought back in. The steps needed to evaluate the event tree are described in more detail below. When both upstream and downstream slope stability must be considered, the event tree becomes much more complex. Issues related to liquefaction probability estimation involving joint probabilities, independence, and correlation require consideration for liquefaction and shear strength loss upstream alone, downstream alone. These issues are discussed in more detail in Appendix J (Reclamation, 2005). Continuous Zone The first item to be addressed is the likelihood that a continuous layer or zone of potentially liquefiable material exists within the dam or foundation. This may be explicitly included as a node in the event tree. While simple in concept, estimating the likelihood for continuity requires significant insight. It is typically based on exploratory information and knowledge of the geologic and dam construction processes. For example, the extent of a potentially liquefiable foundation layer is formulated from what is known about the foundation. If the foundation is composed of lacustrine deposits, there would be reason to believe soil properties identified for a layer would in general be laterally continuous. The same may not be true for alluvial stream deposits. Borehole property data, such as Standard Penetration Tests (SPT), BeckerHammer Penetration Tests (BPT), Shear Wave Velocity Tests (SWV), and Cone Penetrometer Tests (CPT) can provide insights into the potential for a continuous layer. In this regard, the data should be reviewed looking for a continuous low strength layer and not as a population lumped together for statistical analysis. The extent of the loose layer can often be constrained to within some limits from this type of data. Then it becomes a matter of judging the likelihood that the identified layer is continuous enough to lead to a stability problem if it were to liquefy. Typically, continuity parallel to the dam axis of 1 to 2 times the dam height is needed to adversely affect stability without significant 3-dimensional effects contributing to stability. If the continuity transverse to the dam axis underlies most of the dam slope, it is probably of sufficient continuity to affect slope stability. Shorter transverse continuity can also affect slope stability depending on the geometry and strength. Slope stability analyses incorporating post liquefaction shear strengths can be useful in determining how far low-strength materials need to extend beneath a slope before stability becomes an issue. When there are few of the insitu tests normally used to evaluate liquefaction potential at a site, it has been common to first estimate the likelihood of continuity, and then estimate the likelihood that the zone thought to be continuous can liquefy. When there are many insitu tests, it is common to estimate a range of values for some material property (SPT blowcount or shear strength) related to liquefaction and thought to be 'representative' of a zone under the embankment 22-5

Last Modified December 14, 2009 slope that extends laterally two to three times the height of the embankment. Again, the 'representative' value should be judged based on a critical evaluation of the geology and insitu test data, taking care to look for weak zones which have continuity. It is best to avoid equating 'representative' with a statistical average (unless there truly is a statistically-significant sample size from a known badacting zone). A frequent mistake is to average all the available SPT blowcounts in a given geologic unit, regardless of whether the unit appears to have a recognizable low-blow zone of sufficient extent. Another mistake is to average all the available data in a unit when borehole spacing is much greater than two to three times the dam height. In this case, a single low-blowcount interval in a single borehole could be significant. Seismic Load Ranges Larger accelerations and longer durations are generally expected to occur less frequently. Earthquakes can occur randomly within a region of similar seismic activity, or can be associated with an identified seismogenic fault source. Perceived regional slip rates determine potential earthquake frequency on faults. Statistical models determine earthquake frequency where not associated with a fault. Seismic hazard is typically provided as a return period or an exceedance probability for peak horizontal acceleration or in spectral acceleration form at specified periods or period ranges. Reclamation also uses acceleration timehistory records thought likely to represent specified return period ranges. The selection and description of seismic load ranges is covered in the sections on Seismic Hazard Analysis and Event Trees. Liquefaction Estimating the likelihood of liquefaction for any given zone or layer depends on several factors and requires computations outside of the event tree. It is not the intent of this section to provide a detailed discussion of liquefaction analysis. Please refer to the embankment dam draft seismic design standard (Reclamation, 2001), Seed et al (2003), Bray and Sancio (2006), and Boulanger and Idriss (2004) for more information on this type of analysis. Several analyses need to be conducted before the risk analysis team activities take place. For example, the cyclic stress ratio will need to be calculated for each particular load level and at key locations beneath the dam. In addition, raw blowcount data will need to be normalized ahead of time. If CPT or shear wave velocity data is to be also used, that information must also be reduced ahead of time. The reader is referred to Seed et al (2003) for a discussion of these methods. Bray and Sancio (2006) report on how soils of differing Plasticity Index demonstrate liquefaction susceptibility. Boulanger and Idriss (2004) provide additional guidance on liquefaction and post-liquefaction behaivior of finegrained soils.

22-6

Last Modified December 14, 2009 The basic probabilistic liquefaction models (Liao et al, 1988, Youd et al, 2002, and Seed et al, 2003), have all been based on statistical regressions using corrected SPT (N1)60 blowcount, fines content or percent passing the #200 sieve (FC), and cyclic stress ratio (CSR) as the basic input parameters. These various relationships can produce decidedly different liquefaction probability estimates. Reclamation has typically considered the Seed et al relationship to be the most reliable, since it is the most recent and the authors went to efforts to ensure only data of high quality was used in developing the model. However, some weight has been given to the other two relationships in certain cases. The model of Youd et al included only SPT samples known to be within geologic units that were responsible for lateral spreading as a result of liquefaction. It generally shows more spread than the other two relationships, producing lower liquefaction probabilities at higher blow counts and higher liquefaction probabilities at lower blow counts. The model by Liao et al is the most general, being based on many different manifestations of liquefaction behavior. Their regression associated the observed liquefaction behavior with the lowest blowcount recorded in each individual borehole drilled in the vicinity of that behavior. They did not try to identify a representative value for specific geologic units, amalgamating data from several boreholes. The risk analysis team typically develops a distribution of (N1)60 to represent the potentially liquefiable layer or zone of interest. Since there typically is not enough blow count information to develop reliable statistical distributions for a given layer or zone, the simplest way is to develop a cumulative probability distribution based on the degree of belief of the team. The seismic toolbox (Reclamation, 2005) provides the appropriate questions to develop such a distribution based on a review of the available information, as follows: The representative (N1)60 value is not likely to be less than _?_ (10th percentile) and is not likely to be more than _?_ (90th percentile). It cannot be less than _?_ (0 percentile) nor more than _?_ (100th percentile). It is equally likely to be more or less than _?_ (50th percentile). Using these data pairs, a cumulative probability distribution can be defined in @Risk. It may be appropriate to examine more than one distribution, depending on the available information. A similar approach can be used to develop a probability distribution for fines content (percent passing the No. 200 sieve). A spreadsheet is then set up to calculate the probability of liquefaction using the probabilistic liquefaction models selected to perform the analysis. A cell is set up with the (N1)60 distribution and another with the fines content distribution. At each load for which a cyclic stress ratio has been calculated or estimated, the probability of liquefaction is calculated by referencing the cells containing the distributions. These values are than converted to load range values (e.g. conservatively by averaging the load range partition values, or by weighting in some other fashion (see section on Event Trees) and returned to the event tree. The equations for doing this are summarized below:

22-7

Last Modified December 14, 2009

g = (N

Seed et al, 2003: 1, 60 * (1 + 0.004 * FC ) 13 .32 * ln(CSR ) 29 .53 * ln( M W ) 3.7 * ln( ' ) + 0.05 * FC + 44 .97 )
PL = NORMDIST ( g ,2.7, TRUE )

where MW is the average moment magnitude of the earthquakes contributing to the hazard (determined by de-aggregation of the hazard), is the effective overburden stress (in lb/ft2), and the probability of liquefaction, PL, is determined from the standard cumulative normal distribution using the Excel function NORMDIST. The fines content, FC, is capped at 35 percent. Youd et al, 2002: QLY = 7.633 + 2.256 * M W 0.258 * N 1,60,CS + 0.395 * ln(0.65 * CSR / K )
N 1,60,CS = EXP (1.76 (190 / FC 2 ) + (0.99 + ( FC 1.5 / 1000) * N 1, 60

K = 0.6 + 0.0007101 * (16 ' / 1000) 2.4 PL = 1 (1 + 2.718 QLY )

where the correction for N1,60,CS applies only to FC>5 percent, and is capped at 35 percent. Liao et al, 1988 (silty sand, K defined above): QLL = 6.4831 + 2.6854 * ln (0.65 * CSR /( K * K m ) ) 0.1819 * N 1,60
2.56 K m = 173 * M W PL = 1 (1 + 2.718 QLL )

(Note: more than one of these liquefaction models can be used and weighted according to the likelihood (summing to 1.0) that each is thought to represent the case being evaluated.)
Deformation Exceeds Freeboard An estimate of the undrained residual shear strength of the liquefied soil materials is needed to estimate deformations. This is typically done using the relationship developed by Seed and Harder (1990), and shown in Figure 22-2. Although the curve is based on a limited number of case history back analyses, it is typically assumed that soils will tend to have strengths within the limits of the curves drawn by Seed and Harder. For a given (N1)60 clean sand equivalent, a triangular distribution is frequently assigned with its peak corresponding to approximately the midpoint between the two curves, and the upper and low limits corresponding to the upper and lower curve, respectively. Olson and Stark (2002, 2003) may also be considered (Figure 22-3). The relationship of Olson and Stark accounts for the initial effective vertical stress in estimating the strength. Linear extrapolation of these curves beyond a corrected blowcount of 15 is generally considered to be conservative, since the relationships should be concave upward.

22-8

Last Modified December 14, 2009 Using only the Seed and Harder relationship would typically result in a slightly more conservative estimate. To obtain the clean sand equivalent blow counts for undrained residual shear strength assessment of silty soils, blow counts are added to the (N1)60 value according to interpolations form Table 22-1.
Table 22-1. Blowcount correction to obtain clean sand equivalent for strength Fines Content (%) Blow Counts Added to (N1)60

5 25 50

1 2 4

Figure 22-2. Undrained Residual Shear Strength as a Function of Corrected Blowcount (adapted from Seed and Harder, 1990)

22-9

Last Modified December 14, 2009

Figure 22-3. Undrained Residual Shear Strength Determination (after Olson and Stark, 2002)

Reclamation often uses the computer program FLAC to analyze seismicallyinduced deformation. FLAC is a two-dimensional explicit finite-difference program. This program can be used to simulate the behavior of structures built of soil, rock or other materials that may undergo plastic flow when their yield limits are reached. Materials are represented by zones and regions that may be shaped by the user to conform to the physical structure being modeled. Each zone is assumed to behave according to a prescribed linear or nonlinear stress/strain law in response to applied forces or boundary constraints. The represented material can yield and flow, and the grid can deform and move with the material being represented. However, caution and experience is needed when using such sophisticated nonlinear computer programs to ensure the results are reasonable. The models should be thoroughly tested, validated, and verified to ensure reasonable performance. The results of this testing should be documented so that those reviewing the results of the analyses will have as much confidence as possible in the results. Modeling using FLAC can therefore be quite timeconsuming and expensive. Even so, model uncertainty can be included in the probability estimates rather than strictly relying on the output numbers (e.g. to account for three-dimensional effects if two-dimensional models were used). Deformation can result from slope instability under gravity loading alone. If an earthquake can trigger liquefaction, pore water pressure increases reduce shear strength and the slope might become unstable. After liquefaction triggering, a slope can continue to deform even though the earthquake shaking has ceased if the static factor of safety is less than 1. Should liquefaction initiate early in the earthquake, continued shaking provides inertial forces that add to deformation. Modeling experience using FLAC has shown that when the static factor of safety is less than 1, the dynamic deformation portion is typically a small fraction of the

22-10

Last Modified December 14, 2009 total deformation. Intuitively, the dynamic component will be more significant for earthquake acceleration records of long duration, particularly when the earthquake provides strong accelerations at the dams natural frequency. Resource constraints usually dictate that FLAC results are generated for a limited number of loadings and assumed initial conditions. For the example in figures 22-4a and 22-4b below, a foundation layer beneath an embankment slope was assigned residual shear strength values of 50, 100, and 200 psf. Gravity loading alone produced the deformation values labeled Static. A relatively strong earthquake was responsible for the additional deformation labeled Dynamic. Connecting the six model point-estimates with lines, as in figure 22-4a, is reasonable. One could easily analyze the model with additional parameter assumptions to fill in the spaces between previous runs. Likewise, extrapolating the lines to the right, as in figure 22-4b, is appropriate and we would expect verification with additional analysis for higher shear strength values. Extrapolation to the left in figure 22-4b is much more problematic. FLAC becomes unstable with strengths this low. Also, it becomes questionable as to whether a plastic constitutive model as provided by FLAC is appropriate.
14
Crest Deformation (feet)

9
Crest Deformation (feet)

8 7 6 5 4 3 2 1 0 0 50 100

12 10 8 6 4 2 0 0 50 100 150 200 250 300

Static Dynamic (>1.0g pha)

Static Dynamic (>1.0g pha)

150

200

250

Residual Shear Strength (psf)

Residual Shear Strength (psf)

Figure 22-4a.

Figure 22-4b.

Limit equilibrium modeling can be used to discern an approximate value for maximum crest deformation when the embankment slope has a factor of safety less than 1. Presumably, when an upstream slope is unstable there is a downstream remnant where one would expect insignificant deformation. Essentially, this remnant of relatively undisturbed embankment material would provide the highest remaining barrier to uncontrolled reservoir release. The peak of the undisturbed remnant would determine the maximum crest deformation estimate used to gage overtopping likelihood. Figure 22-5a shows a series of circular and wedge-shaped failure surfaces analyzed using a limit equilibrium method. Figure 22-5b shows the same cross section modeled using FLAC. The deformation arrows are absent in 22-5b on the downstream slope at a point where the figure 22-5a shows a failure surface that has a Safety Factor of 1.12. The FLAC analysis shows highly deformed material remaining above the elevation of the peak of the undeformed section. An upper bound crest deformation estimate would result if one assumed that all the highly deformed material would slide into 22-11

Last Modified December 14, 2009 the reservoir. If a FLAC or an alternative deformation analysis does not exist, potential failure surfaces that produce a Safety Factor of about 1.1 in a limit equilibrium analysis might be interpreted to provide a crude estimate for the location of the highest remnant piece of undeformed dam. The likelihood of attaining a safety factor along such a surface less than this can be estimated using reliability analysis (see section on Reliability Analysis).

Figure 22-5a.

Figure 22-5b.

Often a cutoff trench located beneath the dam crest determines the location of the remnant undisturbed block, particularly for analysis of downstream slope stability. If potentially liquefiable materials were removed during original construction to create the cutoff trench, limit equilibrium analysis of postearthquake stability typically reveals Safety Factors well above 1.0 as the analyzed failure surface geometries begin to intersect the cutoff trench. The bottoms of wedge-shaped failure surfaces are limited by the point of intersection between the modeled liquefiable layer and the cutoff trench. Back angles extend up from these points at an approximate angle of (45+/2) from the horizontal for critical failure surfaces, where is the friction angle of the embankment material. A rough estimate for maximum crest displacement can be calculated from the point where this extension intersects the upstream slope.

22-12

Last Modified December 14, 2009 The loss of freeboard needs to be compared to the reservoir elevation at the time of the earthquake. To do this, a curve is prepared similar to that described in the section on Reservoir Level Exceedance Curves for reservoir exceedance probability. A general distribution function can be assigned to the curve using the RiskGeneral function in @Risk. The probability of failure due to overtopping is then typically estimated as the probability of the reservoir exceeding the elevation of the deformed dam crest. If an evaluation using one of the simplified methods described above results in an estimated annual failure probability or annualized loss of life that exceeds Reclamations public protection guidelines, more refined studies are probably justified. This requires detailed FLAC analyses to estimate the loss of freeboard due to various seismic loads. Typically, enough FLAC analyses are run to develop curves (high, median, and low) for freeboard loss as a function of undrained residual shear strength of the liquefied layers or zones. Team judgment incorporating model uncertainty is also included in the development of the curves. These curves are then used in a fashion similar to that described in the preceding paragraph to estimate the probability of overtopping erosion failure.
Alternate Approach

The methods described above rely heavily upon formulaic means to obtain estimates of liquefaction probability from SPT data and to choose representative strength values. There are some instances when it may be preferable to use more of a degree-of-belief approach. Such cases include: Sites where no SPT blow count information is available Sites where many different approaches have been used to assess liquefaction triggering, including shear wave velocity testing (which cant be meaningfully converted to SPT blow counts) or in-situ density testing Sites that have previously experienced a significant earthquake loading, and the measured SPT blow counts may not correlate with the past observed signs of liquefaction or no liquefaction in terms of predicted performance Sites where qualitative arguments (such as questionable continuity, questionable degree to which coarse materials can experience strength loss, potential for drainage of excess pore pressures, etc.) may be used to estimate a low (or in some cases a high) probability of liquefaction Foundations that contain fine-grained soils that may be susceptible to strength loss but not classical liquefaction Foundations that contain very coarse-grained soils, which have few case histories to draw upon to judge the value of empirical formulas for liquefaction probability and residual strength An alternate approach to estimating SPT blow count distributions is to use a degree-of-belief estimating approach to the likelihood of liquefaction or strength loss. The event tree is similar to that already described, and may consist of branches following the earthquake load range probabilities that include: The probability of sufficiently widespread susceptible soils (continuity)

22-13

Last Modified December 14, 2009 The probability that soils will lose strength under a specific load increment The probability that deformations will exceed available freeboard Each of these branches, as well as possible variations, will be briefly discussed.
Continuity of susceptible materials

In cases where continuity is a critical issue, it may be helpful to have this initial branch that estimates the probability (and uncertainty) of whether sufficiently continuous susceptible soils exist beneath a dam to lead to instability or excessive cracking. Factors including geologic deposition, presence of cutoff trenches or natural barriers, effects of embankment loading, and results of available testing or explorations will influence this probability distribution.
Probability of strength loss

In this case, the probability that liquefaction or strength loss occurs is estimated for each of the earthquake loading increments being considered. Risk analysts make experience-based judgments given the available information on the site, which may include performance during past loadings; laboratory test data; triggering analyses using shear wave velocities or penetration tests; and finegrained soil susceptibility based on vane shear tests or CPT values in clayey soils. When SPT blow counts are available for sandy soils, probabilistic curves are consulted (but not rigidly tied to the final estimates of probability).
Deformations exceeding freeboard

This probability is typically estimated by developing curves of expected deformation. For example the team may estimate the range of absolute minimum crest loss, reasonable minimum probable crest loss, best estimate or median crest loss, reasonable maximum probable crest loss, and absolute maximum crest loss. These values then form a probability distribution of crest loss for the strength loss value assumed to result from liquefaction or cyclic failure. If the reservoir remains relatively constant, the deformation curves represent the likelihood of losing freeboard when compared to the operating conditions. When the reservoir fluctuates considerably, the operations cycles are reviewed to get a feel for an annual mean reservoir level that can be used in the estimates. In some cases, these three branches are combined by considering the probability of a given strength scenario, and the resulting deformations given each strength scenario. Specifically, the first two probabilities (probability of continuity and probability of strength loss) are instead phrased as the probability that a given strength will result from a given increment of earthquake loading. This is particularly useful when a team has developed deformation models for several different strength scenarios. The strengths assigned in these scenarios are meant to model a likely range of values and include reasonable upper and lower bounds. For example, if Newmark and/or FLAC analysis have been performed for three different strength assumptions, the risk team estimates the likelihood of each of the three strength assumptions, with the sum of the three probabilities adding to 1.0. Expected deformation curves for each of the three strength scenarios can 22-14

Last Modified December 14, 2009 than be developed as described above. This approach is useful in allowing teams to reflect the (sometimes considerable) uncertainty in estimating the strength loss (and corresponding deformations) that will result from earthquake shaking.

Internal Erosion through Cracks

If the embankment and foundation do not liquefy, and/or if the freeboard is not effectively lost through seismic deformations, the dam will not fail due to overtopping (or rapid erosion of the severely damaged dam crest), but there is still the potential for a slower internal erosion through cracks in the embankment, typically in the crest and upper portions of the dam. Fell et al (2008) include considerations for internal erosion through seismically-induced cracks, based in part on observed damage to embankment dams following large earthquakes. The primary goal is to determine how deep the embankment is likely to crack, and how open the cracks are likely to be relative to the possible reservoir elevation. Once this is determined, the likelihood of internal erosion is assessed a similar fashion as for static loading. The first step in the procedure is to determine the damage class from figures 5.7 and 5.8 in Fell et al (2008). This may require asking the seismologists to deaggregate the seismic hazard to determine the magnitudes of the earthquakes that contribute most to the hazard at various peak horizontal ground accelerations. If liquefaction occurs, assume Damage Class 3 or 4, depending on how severe the liquefaction is judged to be. A Damage Class is determined for each earthquake load partition, and the earthquake load partitions are selected to coincide with the Damage Class contours, if possible. For this reason, it is usually desirable to develop a separate event tree to evaluate internal erosion through cracks (as opposed to tacking it on to the end of the liquefaction tree at the non-failure nodes). If a separate tree is developed, care must be taken in combining these risks with liquefaction overtopping risks (and other seismic risks), as discussed in the section on Combining and Portraying Risks (common cause adjustment) so as to not assign a combined conditional failure probability that is too high for a given load range. Given the Damage Class, determine the likely settlement as a percentage of dam height from Table 5.36 in Fell et al (2008). Cracking begins at the new elevation of the crest after seismically-induced settlement and extends downward from there.

22-15

Last Modified December 14, 2009

Fell et al (2008) Figure 5.7 - Incidence of transverse cracking versus seismic intensity and damage class contours for earthfill dams (Pells and Fell, 2003)

Fell et al (2008) Figure 5.8 - Incidence of transverse cracking versus seismic intensity and damage class contours for earthfill and rockfill dams (Pells and Fell, 2003)

22-16

Last Modified December 14, 2009

Fell et al (2008) Table 5.36 Damage classification system (Pells and Fell, 2003)
Damage Class Number Description Maximum Longitudinal Crack Width (1) mm Maximum Relative Crest Settlement (2) %

0 1 2 3 4

No or Slight Minor Moderate Major Severe

< 10 mm 10 - 30 30 - 80 80 - 150 150 - 500

<0.03 0.03 - 0.2 0.2 - 0.5 0.5 - 1.5 1.5 - 5

(1) Maximum crack width is taken as the maximum width, in millimetres, of

any longitudinal cracking that occurs.

(2) Maximum relative crest settlement is expressed as a percentage of the

structural dam height.

The probability of transverse cracking and the likely crack width at the crest can be estimated from Table 22-2, which shows the range of values suggested in Fell et al (2008). Given the crack opening at the crest, the opening at various depths below the crest can be estimated from Table 5.25 from Fell et al (2008).
Table 22-2. Probability of Transverse Cracking and Maximum Likely Crack Width at Dam Crest Damage Class Probability of Maximum Likely Crack Transverse Cracking Width at Crest mm (in) 0 0.001-0.01 5-20

1 2 3 4

0.01-0.05 0.05-0.1 0.2-0.25 0.5-0.6

20-50 (2) 50 (2)-75 (3) 100 (4)-125 (5) 150 (5)-175 (7)

22-17

Last Modified December 14, 2009

Fell et al (2008) Table 5.25 - Likely crack width at the depth shown versus maximum crack width at the dam crest ( Depths in feet and metres)
Maximum crack width at dam crest 5 feet Inches Millimetres 1.5 metres 1 2 20 40(1.6) 60(2.4) 210(8) 1 5 20 35(1.4) 180(7) 1 5 15 140(5.4) 2 7 110(4.4) 3 90(3.5) 1 60(2.4) Likely crack width at depth shown, (Crack width in millimetres (inches)) 10 feet 3 metres 15 feet 4.5 metres 20 feet 6 metres
(1)

25 feet 7.5 metres

30 feet 10 metres

0.5 1 2 3 4 10

10 25 50 75 100 250

The likelihood of erosion in a crack is a function of the erodibility of the soil and the average hydraulic gradient (and resulting traction shear stress). As reported in Fell et al (2008), highly erodible soils such as silts, silty sands, or dispersive clays may be likely to erode at a crack width of to inch under a hydraulic gradient as low as 0.1, and at widths as small as 1 or 2 mm under hydraulic gradients of 0.5 or more. Clays may not be likely to erode until cracks reach 1 or 2 inches in width and gradients approach 0.5 or more. The likelihood of erosion initiation is the product of the likelihood of a transverse crack forming, the likelihood of the crack extending beneath the reservoir (which is determined in the same manner as likelihood for loss of freeboard, as discussed above), and the likelihood that there is sufficient gradient across the crack to initiate erosion given that the cracking extends beneath the reservoir. Typically, a gradient is estimated that would result in erosion, and the likelihood of the reservoir reaching an elevation to create that gradient is estimated from the reservoir exceedance curve. The rest of the event tree is evaluated using applicable nodes from the event tree described in the section on Internal Erosion and Piping Risks for Embankments. An important consideration while evaluating the rest of the tree is the likelihood that the upstream and downstream filters, transition zones, and/or shells are capable of sustaining a crack to the same depth and width as the core. If this likelihood is high, the chances of a filtered exit (continuation) and upstream crack-filling (progression) may be low.

Foundation or Reservoir Fault Displacement

Where an active fault, or fault capable of coseismic displacement exists in the foundation of a dam, offset along the fault can cause cracking of the embankment and/or conduits passing through the dam. Since each dam and geometry is 22-18

Last Modified December 14, 2009 unique, a site specific event tree needs to be developed to evaluate this on a caseby-case basis. The loading in this case involves fault offsets of various magnitude ranges, and their associated probability. Input from Quaternary geologists specializing in fault and seismic source characterization is typically needed to develop this input. An event tree is developed to describe the specific potential failure mode being evaluated. Nodes on the tree include the potential impacts of fault offset in leading to dam failure. For example, what is the likelihood that each range of displacement is sufficient to develop an open crack through the core, disrupt or offset a filter zone, and/or crack through a conduit. Another possibility concerns an active fault passing through the reservoir of an embankment dam. Fault offset within the reservoir could create a seiche wave capable of overtopping and eroding the dam. Again, it is necessary to develop an event tree, establish return periods for various levels of fault offset, assess the potential for an overtopping wave to develop, and evaluate the likelihood of short duration overtopping to lead to an erosional breach. An initial estimate of wave height equal to the vertical fault offset is probably reasonably conservative in most cases. The reader is referred to Wilson (1972) and Hammack (1973) for additional discussion on modeling seiche waves. However, overtopping failure of a dam due to seiche waves is a relatively improbable failure mode which is only considered when seismotectonic specialists indicate a high likelihood for development of a seiche wave.

Accounting for Uncertainty

Uncertainty is accounted for in the distributions that are input in making the calculations. Spreadsheet cells containing input values are described in terms of a distribution rather than a single value. Then, a Monte-Carlo analysis is performed, typically with 10,000 iterations, to develop a distribution of annual failure probability and annualized loss of life. Setting up spreadsheets to perform seismic risk evaluation of embankment dams requires several calculations to be made outside of the event tree; for example, probability of liquefaction, crest deformation (settlement), and likelihood of deformation exceeding freeboard all can involve calculations rather than just assigning probabilities to the event tree nodes.

Relevant Case Histories

Relatively few dams have actually failed as a result of liquefaction, internal erosion through seismically-induced cracks, or other seismically-related failure modes. However, a few case histories provide relevant insights.
Lower San Fernando Dam: 1971 The upstream slope of Lower San Fernando Dam failed during the 1971 San Fernando Earthquake (Seed et al, 1975). Intact blocks of embankment material moved tens of feet on liquefied hydraulic fill shell material. There was evidence to suggest the slope failure took place after the shaking had stopped. Fortunately,

22-19

Last Modified December 14, 2009 a remnant of the dam remained above the reservoir water level at the time, and the dam did not breach.
Sheffield Dam: 1925 Sheffield Dam failed during the Santa Barbara earthquake of 1925. Although there were no witnesses to the breach, it was believed that the sandy foundation soils which extended under the entire dam liquefied and that a 300-foot long section of the dam slid downstream, perhaps as much as 100 feet (Seed et al, 1969). The dam was located quite close to the city of Santa Barbara, and a wall of water rushed through town, carrying trees, automobiles, and houses with it. A muddy, debris-strewn aftermath was left behind. Flood waters up to two feet deep were experienced in the lower part of town before they gradually drained away into the sea. No fatalities were reported. Austrian Dam: 1989 Austrian Dam was severely cracked and damaged by the 1989 Loma Prieta Earthquake (Forster and MacDonald, 1997), with peak ground accelerations estimated at 0.5 to 0.6g from the nearby Magnitude 7 event. Longitudinal cracks 14 feet deep (based on trenching) formed just below the dam crest on the upstream and downstream slopes. Transverse cracks formed at both abutments, 1 to 9 inches wide, and the embankment separated from the concrete spillway wall, opening a gap of about 10 inches. Fortunately, the reservoir was low at the time of the earthquake, and no subsequent internal erosion ensued. San Fernando Powerplant Tailrace Dam: 1994 A small embankment dam forming the tailrace for the San Fernando powerplant was shaken by large ground motions during the 1994 Northridge earthquake. The earthquake occurred early in the day, and the tailrace dam was intact when powerplant personnel left for the day. The next morning he dam had failed (Davis, 1997). The tailrace concrete lining had buckled in several locations. It was suspected that a layer of loose sand beneath the dam, identified by CPT data, liquefied, and piped through the gaps in the concrete lining undetected, slowly throughout the day. Cracking in Dams Exposed to the Loma Prieta Earthquake: 1989 Harder, 1991, lists the damage that occurred to 35 dams exposed to the Loma Prieta Earthquake. The completion date, maximum dam height, distance to the epicenter, and estimated peak ground accelerations are included along with the damage descriptions. The Loma Prieta Earthquake was a magnitude 7.0 earthquake with approximately 7 to 10 seconds of strong shaking. Dams exposed to less than 0.2g did not experience damage. Dams exposed to peak ground accelerations between 0.2g and 0.35g either experienced no damage or developed longitudinal cracks. Transverse cracking was only noted in dams exposed to greater than 0.35g, though 7 of 19 dams exposed to this level of shaking experienced no damage, 7 of 19 dams experienced either minor or longitudinal cracking and only 5 of 19 dams experienced transverse cracking.

22-20

Last Modified December 14, 2009

Comprehensive Facility Review (CFR) Considerations

There is seldom time during a CFR to set up detailed spreadsheets incorporating Monte Carlo simulations to perform the calculations described in this section. Therefore, simplified event trees are generally developed, and probabilities estimated directly for each branch of the event tree using judgment and subjective probability estimates (see section on Subjective Probability and Expert Elicitation) based on whatever information is available for a particular dam. Table 22-3 below is one example typically used in CFRs.
Table 22-3

A second simplified spreadsheet analysis example would be structured to treat uncertainty to a limited extent. This spreadsheet considers likely low, best, and high estimates for the key variables, giving weightings of 20% to the high and low estimates, and 60% to the best estimates. The user-supplied information is the dam's hydraulic height; 16th, 50th, and 84th percentile values for historical freeboard; representative SPT blowcount as a weighted percentile in each of five blowcount ranges; 16th, 50th, and 84th percentile values of exceedance probabilities for three ranges of earthquake loading; and an embankment erodiblity factor. Assumptions built into the spreadsheet are shown below. The liquefaction probability as a function of blowcount for the three load ranges were assigned high, median, and low values roughly based upon Liao, et.al., with cyclic shear stress taken as half the peak horizontal acceleration. The assumptions for crest elevation change as a fraction of dam height are multiplied by the user-supplied dam height, and the user-supplied freeboard is subtracted to determine if the dam remnant is overtopped. If freeboard remains, the probability of cracking failure is assigned according to the amount of freeboard remaining, adjusted with a multiplier to account for the magnitude of crest loss (the probability is reduced to 1.0 if the multiplier increases the failure probability above 1.0). The erodibilty factor is also multiplied by the assigned cracking failure probabilities, and can be 0.5, 1.0, or 2.0, depending upon the user-supplied assessment of low, medium or high degree of erodibility. All combinations of load range, blowcount range, liquefaction probability, crest loss, and historical freeboard are multiplied across the spreadsheet, according to the 0.2 and 0.6 weighting factors described above, and the end columns are summed to obtain the best estimate for annual failure 22-21

Last Modified December 14, 2009 probability. Separate calculations are made multiplying-across and summingdown only the rows containing the high values, and then all the rows containing the low values in order to present a range for the annual probability of failure estimate. Figure 22-6 below shows some of the output from the spreadsheet.

Figure 22-6 Simplified Spreadsheet Output

Issue Evaluation Uncertainty Analysis

The process frequently used for Issue Evaluation risk analysis of embankment dams subjected to seismic loading is depicted in figure 22-7 below. Note that the curve representing failure likelihood as a function of post-earthquake residual freeboard may include some probability of failure with positive freeboard remaining. This represents the likelihood of rapid erosion through a severely damaged crest, rather than slower seepage erosion through earthquake-induced cracks, which was covered earlier in this section.

22-22

Last Modified December 14, 2009

Figure 22-7 Summary of Probability Estimation Process

Available information is assessed to generate a belief about liquefaction probability (Step 1 in figure 22-7 above). An understanding of site geology and judgment regarding depositional environment, layering and material properties

22-23

Last Modified December 14, 2009 leads to selection of a representative distribution for standard penetration test (SPT) blow count. The probability density function used to model SPT blow count is more an amalgam of many pieces of information than it is a statistical representation of a large sample of blow count data. There may be extensive SPT data for some sites, but more often there is limited SPT data or no SPT data at all. Becker penetration tests (BPT), cone penetrometer tests (CPT), density or shear wave velocity measurements and general geologic information all influence judgment used to create the representative blow count distribution. An industrystandard functional relationship between cyclic shear stress ratio (CSR) and SPT blow count forms the basis of the liquefaction probability estimate (Step 1 in Figure 22-7). A residual shear strength is assigned to each material assumed to be capable of liquefaction (Step 2 in Figure 22-7 above). The functional relationship between residual shear strength and SPT blow count used in traditional deterministic analysis as a proxy for undisturbed sampling and shear strength testing is used here. The uncertainty in residual shear strength for a given blow count is modeled using a random variable that characterizes the following beliefs: The residual shear strength is not likely to be more than the upper bound, not likely to be less than the lower bound, cannot be less than zero, and has a best estimate that is about a third to half way up from the lower bound. A functional relationship between residual shear strength and crest deformation is constructed from the results of dynamic deformation analysis (Step 3 in Figure 22-7). Residual shear strength is varied in a parametric study of seismicallyinduced deformations to obtain estimates for potential crest deformations. Upstream and downstream slope instabilities are both considered, as is the possibility that the entire embankment could be pushed downstream in a translational shear failure. The initial reservoir elevation could be anywhere in its operating range when an earthquake occurs. Historical reservoir operation data indicates the frequency at which the reservoir has been at a given elevation and is used to create the reservoir operating curve, which is used to predict how much freeboard is likely to exist when an earthquake does occur. This difference between the dam crest elevation and the reservoirs initial elevation is treated as a random variable. Given probability distributions for the pre-earthquake freeboard and the likelihood for crest deformation, the likelihood for post-earthquake freeboard is calculated as a joint probability of the two (Step 4 in Figure 22-7). Given post-earthquake freeboard, the likelihood for continuing breach development by overtopping and down-cutting erosion is assessed using another intuitive functional relationship (Step 5 in Figure 22-7). The reasoning used to form the shape of this relationship is as follows: If crest deformation is greater than initial freeboard, overtopping and erosion is virtually certain. Failure from this point would most likely take place rapidly, depending on the degree of overtopping and the erodibility of the embankment materials. If crest deformation is only slightly less than initial freeboard, there may be interconnected cracks in 22-24

Last Modified December 14, 2009 the crest, open and deep enough to intersect the reservoir. Water flowing in these cracks would have to flow fast enough and with sufficient quantity to be capable of eroding and transporting embankment material. Upstream shell or crackstopping materials would have to fail to perform a self-healing, filter-forming operation and a functional downstream filter zone would have to be missing in the original design or displaced/disrupted by crest deformation. Whether breach formation would continue would depend on the depth and velocity of water in the open cracks. Also, there is a possibility that human interventions might be successful. The actual freeboard amount responsible for the various breachcontinuation likelihoods would also depend on the amount of crest deformation. Post-earthquake freeboard values transpiring from different amounts of preearthquake freeboard pose different potential for erosional failure. If crest deformation is much less than initial freeboard, the dam crest is not likely to be in jeopardy, but internal erosion and piping failure modes at other locations within the dam or foundation could become an issue, particularly for embankments already considered to be in marginally poor condition.

Exercise
Given an effective overburden stress of 6,440 lb/ft2, a maximum shear stress of 2,540 lb/ft2, and representative (N1)60 = 17, calculate the probability of liquefaction using the relationship of Liao et al if the average earthquake is MW = 6.0.

22-25

Last Modified December 14, 2009

References
Bureau of Reclamation (2005), Risk Analysis Methodology Appendix J, Estimating Risk from Seismic Loading of Embankment Dams and Foundations, Technical Service Center, Denver, Colorado. Davis, C.A. (1997), Response of the San Fernando Power Plant Tailrace to the 1994 Northridge Earthquake and Recommended Repairs, City of Los Angeles, Department of Water and Power, Water Supply Division, publication AX 215-48, DSR#68514. Fell, R., M. Foster, J. Cyganiewicz, G. Sills, N. Vroman, and R. Davidson (Final Draft 2008), Risk Analysis for Dam Safety, A Unified Method for Estimating Probabilities of Failure of Embankment Dams by Internal Erosion and Piping, URS Australia Pty Ltd, Sydney, New South Wales, Australia. Forster, I.R., and R.B. MacDonald (1997), Post-Earthquake Response Procedures for Embankment Dams Lessons from the Loma Prieta Earthquake, Australian National Committee on Large Dams Annual Conference. Hammack, J.L. (1973), A Note on Tsunamis: Their Generation and Propagation in an Ocean of Uniform Depth, Journal of Fluid Mechanics (Britain), Vol. 60, Part 4, pp. 769-799. Liao, S.S.C.., D. Veneziano, and R.V. Whitman (1988), Regression Models for Evaluating Liquefaction Probability, Journal of Geotechnical Engineering, American Society of Civil Engineers, Vol. 114, No. 4, pp. 389-409. Pells, S. and Fell, R. (2003). Damage and Cracking of Embankment Dams by Earthquake and the Implications for Internal Erosion and Piping. Proceedings 21st Internal Congress on Large Dams, Montreal. ICOLD, Paris Q83-R17, International Commission on Large Dams, Paris. Seed, R.B., K.O. Cetin, R.E.S. Moss, A.M. Kammerer, J. Wu, J.M. Pestana, M.F. Riemer, R.B. Sancio, J.D. Bray, R.E. Kayen, and A. Faris (2003), Recent Advances in Soil Liquefaction Engineering: A Unified and Consistent Framework, 26th Annual ASCE Los Angeles Geotechnical Spring Seminar, Long Beach, CA. Seed, R.B., and L.F. Harder (1990), SPT-Based Analysis of Cyclic Pore Pressure Generation and Undrained Residual Strength, H.B. Seed Memorial Symposium, Berkeley, CA, BiTech Publishing, Ltd., Vol. 2, pp. 351-376. Seed, H.B., K.L. Lee, I.M. Idriss, and F. I. Makdisi, (1975), The Slides in the San Fernando Dams During the Earthquake of February 9, 1971, Journal of the Geotechnical Engineering Division, American Society of Engineers, Vol. 101, No. GT7. 22-26

Last Modified December 14, 2009 Seed, H.B., K.K. Lee, and I.M. Idriss, Analysis of Sheffield Dam Failure, Journal of the Soil Mechanics and Foundations Division, American Society of Civil Engineers, Vol. 96, No. SM6. Wilson, B.W. (1972), Seiches, Advances in Hydroscience, Chow, U.T. (editor), Academic Press, New York, New York, Vol. 8. Youd, T.L., C.M. Hansen, and S.F. Bartlett (2002), Revised Multilinear Regression Equations for Prediction of Lateral Spread Displacement, Journal of Geotechnical and Geoenvironmental Engineering, American Society of Civil Engineers, Vo. 128, No. 12, pp. 1007-1017. Olson, S. M. & Stark, T. D, Liquefied strength ratio from liquefaction case histories, Canadian Geotechnical Journal, Volume 39, No. 3, pp 629647, March, 2002. Olson, S. M. & Stark, T. D., Yield strength ratio and liquefaction analysis of slopes and embankments, American Society of Civil Engineers, Journal of Geotechnical and Geoenvironmental Engineering, Volume 129, No. 8, pp 727737, August, 2003. Boulanger, R.W. and Idriss, I.M., "Evaluating the Potential for Liquefaction or Cyclic Failure of silts and Clays," Report No. UCD/CGM-04/01, Department of Civil and Environmental Engineering, University of California at Davis, December, 2004. Bray, J.D., and Sancio, R.B., "Assessment of the Liquefaction Susceptibility of Fine-Grained Soils," American Society of Civil Engineers, Journal of Geotechnical and Geoenvironmental Engineering, Volume 132, No. 9, pp 11651177, September, 2006. Harder, L.F. Jr., "Performance of Earth Dams During the Loma Prieta Earthquake," Paper LP05, Proceedings: Second International Conference on Recent Advances in Geotechnical Earthquake Engineering and soil Dynamics, Saint Louis, Missouri, March, 1991.

22-27

Last Modified December 14, 2009

This Page Left Blank Intentionally

22-28