A Spatial Probabilistic Framework for Regional Seismic Simulation of Earth and Rockfill Dams

1 2

G. Chawla 1, J. D. Frost 2
School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, GA, USA School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, GA, USA retained water levels. Instead, a metamodeling-based methodology was adopted to efficiently simulate the seismic response of the dams. The methodology involves a series of steps where the entire dam population and the seismic hazard they may experience is first analyzed. Then, only a few representative dam models are analyzed and the results are used for damage prediction of all the dams. Fig. 1 shows a flowchart, which explains the step-by-step procedure adopted for computation of the response simulation of all the dams.

Abstract This paper presents a methodology for seismic response simulation of earth and rockfill dams in an earthquake vulnerable region. The methodology incorporates the effect of inherent uncertainties in factors such as dam construction and seismic hazard on the seismic response of the dam. Data for relevant dams was obtained from the National Inventory of Dams. The seismic response of dams was simulated using the finite difference program FLAC. An attempt to simulate the seismic response of all dams over a large region while incorporating parameter and seismic demand uncertainty would require millions of FLAC runs. This unrealistic task was avoided by metamodeling of FLAC models using the Design of Experiments (DOE) technique. Various model parameters that influence the seismic response of the dam were identified. A DOE technique was used to select a few combinations of these influential parameters. The Response Surface Methodology was used for metamodeling wherein the damage indices were approximated by a regression equation in which the model parameters were the predictor variables. Statistical distributions were assigned to the model variables and the resultant probabilistic response was computed by performing a Monte Carlo Simulation on the response surface equation. The entire analysis was integrated in a spatial framework where the probabilistic response of all dams in the region could be observed. Keywords Design of Experiments, Earth Dams, Dynamic Analysis, Metamodeling, Response Surface Methodology, Seismic Response Simulation

INTRODUCTION Earth and rockfill dams are a class of structures, which, if breached during an earthquake, can produce serious consequences ranging from direct loss of life to indirect economic losses. Hence, a detailed investigation of their seismic response is appropriate. This paper presents an integrated, scaleable framework for computation and presentation of the seismic response of earth and rockfill dams present in an earthquake vulnerable region such as the Mid America region.
METHODOLOGY

Fig.1. Flowchart of Analysis Methodology

INVENTORY OF EARTH AND ROCKFILL DAMS

This study focused on development of a methodology with which the seismic response of a large number of dams can be efficiently computed and viewed in a spatial framework. Clearly, this task cannot be accomplished if each dam were to be analyzed for different earthquake scenarios, different material properties and different

An inventory of earth and rockfill dams is available at the National Inventory of Dams (NID) website. Reference [1] describes the process of legal and technical aspects of dam inventory data collection. The dam inventory for the entire United States was obtained from the National

SDEE/ICEGE 2004

372

Slopes, Embankments, Dams and Waste Fills

Inventory of Dams website for this study. The database includes in excess of 65,000 dams that was then filtered for the required inventory. Data for only Earth, Rockfill and Earth-Rockfill dams was considered. Also, if essential data like dam type, latitude, longitude, height or storage, which is required for response simulation and presentation was not reported, then those inventory data was excluded from the analysis. These incomplete data sets accounted for less than 1% of the total inventory. It was noted that other data like dam age, design, and foundation type was not reported for some dams. e.g. dam age was not reported for 40% of the dams and dam design and foundation data is not reported for 95% of the dams. The data in the NID database gives no information regarding the embankment and foundation material. Hence, the dams were analyzed for a range of assumed material properties.
HAZARD DEFINITION

373

NID database, database integration with USGS seismic hazard maps was done using GIS tools. A series of maps were made and reviewed to give insight into the spatial distribution of dams with respect to seismic hazard. Fig 2 shows one such map with the NID dams overlaid on the 2% probability of exceedence USGS hazard maps.

Seismic hazard definition is required for seismic response computation and presentation. All the dams present in the Mid America region will not experience the same hazard levels. The inventory of dams cannot be fully understood until the seismic hazard, which can be experienced by the dam, is also presented in the database. To date no agency has built a database which relates the dam to the seismic zone in which it is present. The seismic hazard of the dams needs to be defined for three reasons:  To select the relevant dams to be analyzed based on the severity of hazard which could be experienced by the dam.  To define the model parameters for FLAC simulation.  To define the predictor variables for metamodeling analysis. Seismic hazard itself can be defined by deterministic scenario earthquakes (DSHA) or by probabilistic analysis (PSHA) which includes all the source effects in a probabilistic manner. For this study, the seismic hazard was defined using USGS maps, which are computed using PSHA [2]. USGS based seismic hazard definition yields some distinct advantages. First, the USGS method utilizes the PSHA method and incorporates all possible seismic zones, thereby eliminating the need for defining different earthquakes for every deterministic scenario. The USGS method also uses several weighted attenuation relationships to define the expected hazard at different locations for B/C boundary conditions. The hazard is computed in terms of Spectral Acceleration values for periods of 0.0, 0.2, 0.3 and 1.0 seconds and for 2%, 5% and 10% probability of exceedence levels and is presented in the form of Spectral Acceleration contours as shown in Fig. 2. As a result of their location, the dams will experience different levels of seismic hazard. To simulate this, the spatial variation in seismicity was coupled with the dam location. Since, dam locations were available from the

Fig. 2. Spatial Distribution of dams with respect to Seismic Hazard

A database integration of the USGS spectral acceleration values with the NID database was undertaken. After the integration, the seismic values were appended to the existing NID database. This resulted in an enhanced dam database, which also contains the seismic hazard defined in terms of spectral acceleration values for the three probability of exceedence levels. Table 1 shows the spatial distribution of dams present in the Mid America region. The PGA from USGS maps (2% P.E. in 50 years) was used as a measure of seismic hazard. Clearly there are a large number of dams in the low hazard range and a relatively small number of dams in the high hazard range but even this number of dams, which need to be evaluated justifies the proposed analysis approach. Dams which can experience a PGA of less than 0.1g with 2% P.E. were not included in the analysis.
TABLE 1: SPATIAL DISTRIBUTION OF DAMS PGA Range (g) 0.1 0.2 0.2 - 0.3 0.3 0.4 0.4 0.5 0.5 0.6 0.6 0.8 0.8 1.2 1.2 1.6 Average PGA (g) 0.15 0.25 0.35 0.45 0.55 0.70 1.00 1.40 Number of Dams 7606 2869 884 436 198 158 195 73 Maximum Height (ft.) 331 360 155 100 96 74 51 48 Maximum Storage (acre-ft) 3760700 3016000 1383800 16857 153000 82942 1110000 29500

Knowing the spectral acceleration values at the dam location, the earthquake ground motions that can be experienced at that location can be computed. Although other spectral acceleration values were available, only PGA values were used for defining seismic hazard for the subsequent analysis. This was because any attempt to define seismic hazard by more than one parameter would

374

11th ICSDEE / 3rd ICEGE Proceedings

The best FLAC prototype was selected based on the following guidelines:  The FLAC prototype should model the actual dams as closely as possible. Minor design details could be omitted.  The FLAC model should be compatible with other assumptions e.g. hazard definition.  The FLAC model parameters should be readily quantifiable and should be used as metamodeling parameters in the Design of Experiments step.  The FLAC model should not be defined by a large number of variables because this would result in too many factors for Design of Experiments (DOE) and too many FLAC runs. Furthermore some of the variables would likely be proven to be insignificant using the response surface regression equation.  The degree of complexity and details in the FLAC models serve no significant advantage and should not be encouraged because such dam details are unavailable.  The FLAC model should simulate the seismic response of the soil as closely as possible. Model Specification The following FLAC model was finalized based on the above criteria.  Physical and Design parameters: The height of the dam (Hd) and the height of retained water (Hw) were identified as important variables. The height of the dam was obtained from the NID database. The height of retained water was represented by a ratio of height of water to height of dam (0 < Hw/Hd <1). The effect of height of retained water was studied by taking different values for this ratio. Other design parameters were not so important and were either defined as a constant or as a multiple of height, e.g. embankment slopes were taken to be 2.5H: 1V, core slopes as 1H: 1V, crest width as 10% of height of the dam, etc.  Embankment and Foundation material parameters: Since material parameters were unavailable in most cases, the dams were analyzed for some assumed values of material parameters. Three sets of soil parameters were selected for analysis purpose. These parameters represent poor, fair and good type of soil. The poor type of soil can also be construed as unengineered soil and the good type of soil can be interpreted as engineered soil. The parameter values are compatible with each other where some of the parameter values are derived from (N1)60 using established correlations. Table 2 lists the various soil parameters for each type of soil. These soil parameters are assigned to the embankment material. For the initial set of analysis, the earth embankment is assumed to be constructed on rock. This simplifies the analysis, requires fewer FLAC runs and is compatible with the hazard definition. However, such an assumption doesnt allow for

require more FLAC runs. This would in turn demand a different design of experiments setup for the correlated spectral acceleration values. Most importantly, the response surface regression equation would require the removal of one of the highly correlated parameters used to define the hazard. Hence, the seismic hazard was defined using the PGA value. The time histories for various PGA values were generated using the SMSIM program [3]. This program uses stochastic methods and random vibration theory to generate synthetic ground motions. The program requires magnitude and distance as input and computes peak values (PGD, PGV, PGA) and a desired number of time histories. Time histories were generated by selection of a feasible combination of magnitude and distance, which resulted in the desired level of PGA. These time histories were used in FLAC dynamic analysis of earth dams. It is emphasized that these time histories represented a particular level of hazard in the form of PGA. Since the entire response spectrum is not utilized in generation of time histories, this hazard definition has an epistemic component of uncertainty which can be removed by better hazard definition i.e. definition of hazard by entire response spectrum. This however was not done in the present study because of reasons discussed above. Even if the entire response spectrum is defined, the time histories cannot be uniquely defined. This leads to an aleatory uncertainty component in the seismic response. This aleatory uncertainty cannot be removed. The effect of these uncertainties in time history on the dam inventory seismic response requires further study.
FLAC MODELING

This study utilized FLAC (Fast Lagrangian Analysis of Continua) for seismic response simulation of dams [4]. FLAC is an explicit finite difference program for geomechanics computations. The dynamic option of FLAC offers a non-linear and effective stress based approach for seismic response simulation. FLAC needs a minimum set of input parameters for complete definition of the dam prototype and for calculation of its seismic response. These input parameters can be classified as  Physical and design parameters (e.g. dam height, embankment slopes, foundation type and depth, etc.)  Embankment and foundation material parameters (e.g. strength and permeability of soil, etc.)  Material constitutive law (Cyclic stress-strain behavior of the soil)  Seismic hazard (Acceleration time histories of expected ground motions) Modeling Options Several FLAC models, ranging in complexity were developed. Every model had its own assumptions and no specific model could be classified as the correct model.

Slopes, Embankments, Dams and Waste Fills

simulation of phenomena such as liquefaction of sand foundations. Other models with sand and clay foundations are currently being developed.
TABLE 2. SOIL PARAMETERS FOR FLAC MODELING SAND CLAY

375

 Stabilization of the system by model cycling.  Application of retained water level and development of phreatic surface by model cycling.  Definition and application of acceleration time histories.  Dynamic analysis of the dam.

Soil Quality Poor Fair Good Poor Fair Good Soil Index -1 0 1 -1 0 1 (N1)60 5 20 35 5 20 35 Unit Weight (Kg/m3) 1950 2050 2130 1500 1700 1900 Friction Angle 28 37 42 15 20 25 Cohesion (kPa) 0 0 0 10 55 100 Permeability (m/sec) 10-3 10-4 10-5 10-6 10-7 10-8 Bulk Modulus (MPa) 10 55 100 10 55 100 Shear Modulus (MPa) 6 25 33 2 11 21

 Material Constitutive Law: Ideally, a comprehensive model for soil should account for all the physical effects that occur during static and dynamic loading, such as volume changes, energy dissipation and modulus degradation. In this study, the Mohr Coulomb model was used to model clay and sand behavior. Additionally, the Byrne model was used to simulate liquefaction in sands. The constitutive model for cyclic stress-strain behavior for sands as suggested by Byrne [5] is (vol /) = C1 exp (-C2 (vol /)) (1) where C1 and C2 are interrelated constants and are also related to other soil parameters as C1C2 =0.4, C1 =7600 Dr2.5 and C1 =8.7 (N1)601.25 (2) The correlation of the dynamic soil parameters with other simple soil parameters like (N1)60 makes the task of definition of type of soil fairly unambiguous.  Seismic Hazard: The dynamic excitation is applied to the dam in the form of acceleration time histories. The time histories represent the expected PGA at the site for B/C boundary conditions. The acceleration histories are applied in the horizontal direction to the base of the dam foundation, thereby representing horizontally polarized and vertically propagating shear waves.  Damage Index: The settlement of the crest of the dam was employed as the damage index, which quantifies the extent of damage to the dam from a seismic activity. The quantification of damage using other indices requires detailed study and is in process. FLAC Response Simulation Program The FLAC response simulation program is a series of FLAC command and FISH functions. Fig 3 shows a FLAC model run, which simulates the following process in a step-by-step manner:  Placement of the foundation material.  Construction of the embankment in a single stage.

Fig. 3. FLAC Model Run Which models to run? The FLAC program is an efficient tool for computation of the seismic response of a dam. However, even this program is computationally expensive and time consuming. Solving a FLAC simulation requires more than one hour on a Pentium-IV computer. Even the mundane task of converting the SMSIM computed time histories to the format required by FLAC takes more than half an hour. There are thousands of dams present in the inventory database, which should be analyzed. Also, the dams need to be analyzed for parametric uncertainty. This could mean millions of non-linear time history computational runs. This impractical approach is avoided by adopting the metamodeling technique, which is achieved through a Design of Experiments (DOE) procedure. The design of experiments is aimed at selection of the best combinations of FLAC prototype variables that should be analyzed and used as input for FLAC program runs. .
DESIGN OF EXPERIMENTS

Metamodeling is modeling of models and is achieved through a Design of Experiments (DOE) procedure. A Design of Experiment (DOE) is a structured and organized method of designing sets of experiments which will generate enough useful data required for deriving relationships between factors (Xi) affecting the experiment and the outputs of that experiment (Yj) without costing too much or taking too long [6]. In implementing DOE, the entire population of dams was first analyzed and their physical and material parameters were treated as variables. These variables are the ones that may affect the seismic response of the dam.

376

11th ICSDEE / 3rd ICEGE Proceedings

TABLE 4. DOE TABLE Pattern ++--+--+---++ --++----++-++ 0A000 --+-+ A0000 0a000 0000A +-+-+++-+ +---+ -+-+000a0 a0000 00a00 -++-000A0 ++++-+--+ +-+++ 00000 0000a 00A00 -++++ Hd Hw/Hd (m) 108 0.9 108 0.1 4 0.1 4 0.1 4 0.1 108 0.9 56 0.9 4 0.1 108 0.5 56 0.1 56 0.5 108 0.1 108 0.9 108 0.1 4 0.9 56 0.5 4 0.5 56 0.5 4 0.9 56 0.5 108 0.9 4 0.9 108 0.1 56 0.5 56 0.5 56 0.5 4 0.9 S -1 -1 -1 1 -1 -1 0 1 0 0 0 1 1 -1 -1 0 0 -1 1 0 1 -1 1 0 0 1 1 C -1 1 1 1 -1 1 0 -1 0 0 0 -1 -1 -1 1 -1 0 0 -1 1 1 -1 1 0 0 0 1 PGA (g) 0.150 0.150 1.400 0.150 0.150 1.400 0.775 1.400 0.775 0.775 1.400 0.150 1.400 1.400 0.150 0.775 0.775 0.775 0.150 0.775 0.150 1.400 1.400 0.775 0.150 0.775 1.400 H (m) 8.0000 5.5000 0.0700 0.0005 0.0020 14.5000 1.3500 0.0500 10.0000 0.7000 1.8000 5.0000 13.5000 12.0000 0.0040 1.2000 0.0300 1.3000 0.0030 0.9500 7.0000 0.2000 11.0000 1.0000 0.5000 0.7000 0.0500 Log (H) 0.9031 0.7404 -1.1549 -3.3010 -2.6990 1.1614 0.1303 -1.3010 1.0000 -0.1549 0.2553 0.6990 1.1303 1.0792 -2.3979 0.0792 -1.5229 0.1139 -2.5229 -0.0223 0.8451 -0.6990 1.0414 0.0000 -0.3010 -0.1549 -1.3010 H/Hd 0.0741 0.0509 0.0175 0.0001 0.0005 0.1343 0.0241 0.0125 0.0926 0.0125 0.0321 0.0463 0.1250 0.1111 0.0010 0.0214 0.0075 0.0232 0.0008 0.0170 0.0648 0.0500 0.1019 0.0179 0.0089 0.0125 0.0125

Only the key variables were emphasized and other less important variables were taken as functions of the important ones. The five selected variables for the modeling and metamodeling were height of the dam (Hd), ratio of retained water to height of dam (Hw/Hd), the sand quality index (S), the clay quality index (C) and the seismic hazard (PGA). These five are the key factors, which define a FLAC model. A three level design that can also be used to predict the second order (quadratic) effects was selected for DOE. The minimum, average and the maximum values of the five factors are presented in Table 3
TABLE 3: THREE LEVELS FOR THE FIVE FACTOR DOE Factor Hd (in meters) Hw/Hd S C PGA (in g) Minimum 4 0.1 -1 -1 0.15 Average 56 0.5 0 0 0.775 Maximum 108 0.9 +1 +1 1.4

A five-factor design was selected for the five parameters that define the model. Several options for a three-level five-factor design exist [6]. Fig. 4 shows some of the commonly used three level three factor designs. The red points correspond to the input data sets used for performing the experiment and later creating the metamodel regression equation. Each design has its own assumptions, merits and demerits. A five-factor Central Composite Design was selected for designing the FLAC experiments. The effect of type of design will be investigated in future studies.

METAMODELING

Fig. 4. Commonly used DOE Techniques A five factor Central Composite Design requires 27 experimental runs. These runs are sufficient to explore the effect of each factor on the damage index. Since, the time histories are not unique for each PGA level, two different time histories from SMSIM were used for each model and the average damage index was calculated from them. This resulted in 54 FLAC simulations using which the response surface equation was developed. Table 4 shows the DOE table of the 27 experimental runs (FLAC models) developed using SAS JMP program [7]. The table also lists the average settlement (H) that the models experienced during FLAC simulation.

Metamodeling is undertaken to avoid the large number of computational runs, which would otherwise being required. It also adds scalability to the analysis methodology [8]. There are different metamodeling techniques like Response Surface Methodology (RSM), and Artificial Neural Networks. This study has used RSM as the metamodeling technique for the seismic response analysis of dams. A Response Surface regression equation approximates the damage index as polynomial functions of model variables. The coefficients of the equation are calculated by a least square regression on the data obtained from the designed FLAC experiments. The equation is then used to approximate the damage indices of other combinations of model variables. The quality of a response surface depends on whether the degree of the approximating polynomial can be fixed at 1 or 2 since low-degree models contain fewer terms than higher-degree models and thus require fewer experiments to be performed. The lower order terms and interaction effects are more important than higher order

Slopes, Embankments, Dams and Waste Fills

terms. The predicted response function considering 2degree polynomial model is: Y =0 +1X1 +2X2 +11X12 +22X22 +12 X1 X2 +... (3) The response surface equation was developed for logarithm of the settlement rather than the settlement value itself. This is because the regression for log of settlement gave a higher R2 value (=0.997) and never predicted negative settlement values when compared to the regression on the settlement values. It also gave a constant variance, which was observed in the residuals versus fitted values plot thereby reinforcing the validity of the linear regression model. The settlement values were computed later by raising the predicted log settlement value to power of 10. The response surface equation as computed using SAS JMP program was: Log (H) = -3.37 + 0.0494Hd + 1.02(Hw/Hd) 0.195S - 0.118C + 1.67PGA -0.00317 Hd*r + 0.001570Hd*S + 0.00177(Hw/Hd)*S + 0.00109Hd*C 0.00808(Hw/Hd)*C - 0.0303S*C - 0.01007(Hw/Hd)*PGA - 0.190(Hw/Hd)*PGA + 0.000937S*PGA + 0.00310C*PGA - 0.000114Hd2 - 0.375(Hw/Hd)2 - 0.0683S2 - 0.0194C2 - 0.181PGA2 . (4) The response surface was then used to predict the settlement for all dams in Mid America. The five input factors of the metamodel were used as input. Height was obtained from NID database and PGA from the enhanced database obtained from integration with USGS seismic hazard maps. The ratio of height of water to dam is a variable but was set to average value = 0.5 (half full) for presentation purposes. Since quality of construction is not known, average values of sand/clay quality index (=0) was used. A parametric analysis can also be done by performing a Monte Carlo Simulation on the response surface equation, e.g. for the William Shaddan Dam as shown in Fig. 6, H = 12.9m, PGA=1.4g. Using uncertain values for Hw/Hd ~ N (0.5, 0.12), S ~ N (0, 0.22) and C ~ N (0, 0.22), the uncertain settlement values as predicted by Monte Carlo simulation are: Average of settlement = 0.2 m Standard deviation of settlement = 0.025 m Coefficient of variation = 12.5 % The response surface equation serves a distinct advantage here. The Monte Carlo simulation is done on the response surface equation i.e. the random numbers generated for Hw/Hd, S and C during the Monte Carlo simulation compute the corresponding damage index from the response surface equation rather than asking for a FLAC model run for the random inputs.
INTEGRATION IN SPATIAL FRAMEWORK

377

contours for relative settlement values (H/Hd) can also be viewed. Fig.6 shows how the data for a dam can be viewed and queried. The William Shaddan Dam located near the NMSZ is selected as an example. The integrated database shows all the relevant values including the settlement of the dam (= 0.2 meters for a height of 43ft = 12.9m) for the average conditions of S = 0, C = 0 and Hw/Hd = 0.5. RESULTS The results of FLAC simulation program were summarized in Table 4. The results from the metamodeling exercise were summarized in the response surface, i.e. equation (4). The settlement for any dam can be computed from the response surface equation. The analysis was applied for prediction of settlement for the dams in Mid America, the results of which are summarized in Fig. 5.
DISCUSSION

This paper has presented an integrated scaleable method for seismic response prediction of earth dams. The methodology relies on the metamodeling-based approach, with which the damage for all the dams can be predicted from simulation results of just a few models. The methodology follows a systematic step-by-step procedure presented in Fig.1. Each step has its own assumption and predictions, which adds to the analysis and hence the overall quality of the analysis and the prediction is dependent on the poorest step in the methodology. Hence, every effort was made that no individual step is poorly executed when compared to other steps. The methodology also calls for compatible assumptions in all steps, which was also ensured. At present, the methodology includes a simple and relatively accurate method for computation of seismic response. The methodology will be further enhanced by improvements in all steps. In future, the current FLAC models will be replaced with more complicated models that can also simulate settlement and liquefaction of foundation soils. Other methods for Design of Experiments like Box-Behnken design will be investigated. The spatial integration and presentation of the results will be improved by developing an interactive interface, which will ask the user for dam data and can be used to compute the damage index using the metamodel equation, thereby relieving the user from the nuances of computations done at database level. CONCLUSIONS This paper presented a methodology for seismic response simulation of earth and rockfill dams in an earthquake vulnerable region. The methodology was applied for seismic response prediction of earth dams present in Mid America for a 2% P.E. earthquake. Dam inventory was obtained from the National Inventory of

All the original NID data, USGS data, predictor variables and response variables are integrated in one single database, which is used for final GIS presentation. Fig. 5 shows the settlement prediction values (H) contours for the dams present in Mid America. Similarly,

378

11th ICSDEE / 3rd ICEGE Proceedings

Dams, which was then analyzed for frequent typologies. The hazard for the dams was defined using USGS probabilistic hazard maps. The seismic response of dams was simulated using FLAC. A Design of Experiments based technique was adopted to reduce the number of simulation runs required to model all the dams. Various model parameters that influence the seismic response of the dam were used as model and metamodel parameters. The Response Surface Methodology was used for metamodeling the models wherein the damage indices were approximated as a function of the model parameters. Uncertainty in seismic response due to parameter uncertainty was evaluated by performing a Monte Carlo Simulation on the response surface equation. The entire analysis was integrated in a spatial framework where the probabilistic response of all dams in the region was observed. ACKNOWLEDGMENTS This research was conducted at Georgia Institute of Technology and was supported through the Mid-America Earthquake Center by the National Science Foundation under grant ECC- 9701785. Discussions with B. Goodno and J. Craig during the development of the approach presented herein are greatly appreciated. REFERENCES

[1] USACE, U. S. Army Corps of Engineers, National Inventory of Dams, Methodology, State and Federal Agencies Manual, Version 2.0, US Army Corps of Engineers, Washington, DC, 2001 [2] Frankel A.D., Mueller C.S., Barnhard T., Perkins D.M., Leyendecker E.V., Dickman N., Hanson S. and Hopper M., National Seismic Hazard Maps: Documentation - Open-File Report 96-532, U.S. Geological Survey, 1996 [3] Boore D.M., SMSIM - Fortran programs for simulating ground motions from earthquakes: version 2.0, A revision of OFR 9680-A. Open-File Report 00-509. US Geological Survey, CA, 2000 [4] Itasca, FLAC - Fast Lagrangian Analysis of Continua: Optional Features: Dynamic Option, Version 4.0, Itasca Consulting Group, MN, 2000 [5] Byrne, P. M., A cyclic shear-volume coupling and pore pressure model for sand, Proceedings, 2nd international conference on Recent Advances in Geotechnical Earthquake and Soil Dynamics, St. Louis, Missouri, March 2001, Vol.1, pp. 47-56 [6] Neter J., Kutner M. H., Nachtschiem C. J., Wasserman W., Applied Linear Statistical Models, McGraw-Hill/Irwin; 4th edition 1996, pp. 1045-1310 [7] SAS Institute, JMP Design of Experiments Guide, Version 5, SAS Institute Inc., Cary, NC, 2002. [8] Friedman L. W., The Simulation Metamodel, Kluwer Academic Publishers, 1995.

Slopes, Embankments, Dams and Waste Fills

379

Fig 5. Settlement Prediction values presented in GIS framework

Fig. 6. Settlement of a dam as queried from GIS framework