Atatürk dam on the Euphrates river is the key structure of a multipurpose project for hydro power generation and water supply. The 170m high rockfill dam was completed in 1990. For the safety monitoring of the dam, a detailed finite element model of a representative section was created, assuming elasto-plastic behaviour of the main dam materials. Two basic load cases were considered, i.e. the gravity and water loads. The inelastic deformations due to these two load cases were superimposed linearly using two time-dependent coefficients to represent the creep-type deformations. These coefficients were calibrated on the basis of the available geodetic measurements, also taking into account the results of the laboratory and field tests. By using this semi-empirical model, the deformations of the accessible dam surface could be predicted with high accuracy (correlation coefficients of over 95%). The calibrated model provides an insight into the processes within the dam body, which lead to the observed deformations at the dam surface. With the help of the calibrated model, the effect of the raising of the reservoir level on the dam deformations can be predicted, which would not be possible by other methods such as statistical models. The deformations predicted on the basis of this model are useful for the safety monitoring of the dam.

**Atatürk dam**

The ongoing deformations of Atatürk dam are being regularly and systematically monitored. The dam is instrumented with different types of sensors, which provide information about the internal state of the dam. The most reliable data of the dam deformations are, however, from geodetic measurements carried out twice a year at several sections of the dam. In this paper, a semi-empirical method for the interpretation of the geodetic deformation measurements of one particular dam section is presented and discussed. The main objectives are (i) to analyse the deformations that have occurred within the dam, and (ii) to predict the future displacements for monitoring the dam behaviour in the coming years. These objectives have been attained with the help of a two-dimensional finite-element model of the dam-foundation system.

The prediction of the future displacements of individual observation points could also be accomplished by means of a statistical or neural network model. However, such a non-physical model would not provide any physical insight into the deformation processes occurring in the dam. In contrast, a finite element model has the advantage that the mechanism of the deformation of the whole dam can be simulated based on measurements made at observation points located on its exterior surface.

It may be argued that it is not possible to know what has occurred inside a dam by measuring deformations at reference points located only on the accessible surface. The answer is simple: all geophysical methods that are used to describe the global structure and properties of the earth, and geophysical exploration methods are based on measurements on the earth’s surface. These methods have been most powerful in conjunction with models of the underground and are universally accepted. In the case of a dam, we are in a much better position because the interior structure and the basic properties of the different zones are known. Thus, it is possible not only to set up a model with geometrically correct material zones but also to establish the best estimates of the material properties by fitting the calculated deformations with the measurements made at the dam surface, while taking into account the values determined by laboratory and field tests. For Atatürk dam, a relatively simple finite element model could be created and calibrated to achieve an excellent fit with the measured deformations.

The concept described above differs from a ‘blind’ prediction of dam behaviour made on the basis of the material properties from laboratory tests, although this cannot be avoided in analyses performed at the design stage. The quality of such a prediction can vary widely, especially in the case of embankment dams. For instance, in the icold Benchmark Workshops, predictions made by different analysts of a benchmark problem tend to show a large scatter, even when they use identical basic geometry data, material properties and applied actions. The inverse analysis discussed in this paper is a rational way to overcome the problem of a pure ‘design’ model of a dam and its foundation. The material properties obtained from an inverse analysis must, however, be checked and compared with laboratory data, which exist for most large dams. Processes that cannot be explained by data available from standard laboratory tests may have to be investigated further with the objective of understanding the underlying physical processes.

*Dam features *

The Atatürk dam is a zoned rockfill dam with a central core and is located on the Euphrates river (figure 1). The main features of the dam are as follows: Dam height – 170m; Crest length – 1670m; Crest level – 549m; Maximum base width – approx. 900m; Dam volume – 84 x 10^{6}m^{3; }Reservoir volume – 48km^{3; }Installed capacity – 2400MW; Annual energy generation – 8100GWh.

The construction of the cofferdam lasted from 1985 to 1987. The fill work for the main dam began in 1987 and was completed in 1990. The reservoir level reached 535m asl in March 1994 and has varied between 526m and 537m asl since then. The maximum and minimum operation reservoir levels are 542m and 526m asl respectively.

**Observations on deformations during and after construction **

The main observations on the dam deformations during and after the construction are as follows:

•It is estimated that the central portion of the dam crest has settled by around 7m since the end of the construction in 1990. Since the start of the detailed geodetic monitoring of the dam crest in autumn 1992, crest settlements of up to 4.3m have been measured. The settlements are, however, very small near the downstream edge of the crest, roughly beyond the interface between the core and the downstream filter.

•The total post-construction settlements of the downstream slope and the downstream edge of the dam crest have been relatively small (max. 1.5m). The largest settlement of the downstream slope has occurred at about three-quarters of the dam height. In particular, the post-construction settlements at the downstream edge of the dam crest are very small.

•The horizontal (radial) displacement increases steadily from the bottom to the top of the downstream slope; the maximum horizontal displacement since the end of construction is about 2.9m.

•The dam crest developed visible irregularities because of the large deformations.

•During the crest reinstatement carried out from 1997 to 1999, clear evidence of the downward movement of the core relative to the downstream filter was found.

•The highest deformations have occurred in the central part of the dam. The deformations become smaller towards the abutments and towards the downstream toe of the dam.

•The available information shows that the deformation of the foundation rock is not a significant factor in the deformation observed on the dam surface.

•Settlements and horizontal displacements are still taking place under more or less constant loading conditions. However, the deformation rates are slowing down with time.

•High pore pressures developed in the clay core during the construction of the dam, especially towards the end of the construction. The dissipation of the pore pressures with time has been extremely slow due to the very low permeability of the core material. In general, the piezometric heads in the clay core remain significantly higher than the reservoir level.

**Geodetic deformation measures**

The geodetic measurements of the dam deformations began in July 1990, just before the end of the construction. For this purpose, a number of benchmarks were established on the dam surface. Most of the benchmarks were located on seven cross-sections distributed evenly along the dam axis. The settlements and horizontal (radial) displacements of benchmarks located in various cross-sections situated away from the abutments are of similar magnitudes. This indicates that the dam behaves essentially as a two-dimensional structure. Hence, only the geodetic measurements made approximately at the central cross-section of the dam were used as the basis of the present calibrated model. A complete set of geodetic measurement data is available for five benchmarks located on this cross-section.

**Basic concept and methodology**

From the geodetic measurements, the settlements and the radial (horizontal) displacements of the dam surface are known accurately. The relative contributions of the gravity and water loads to the ongoing deformations are, however, not known a priori. The basic idea of the proposed model is to split the inelastic deformations that have occurred since reservoir impoundment into a term proportional to the static deformation due to the gravity load (self-weight) and a second term proportional to the static deformation due to the water load. This is not a new idea; for example, creep strains in concrete structures are usually assumed to be proportional to the elastic strains due to the applied loads. Moreover, the concept of separating elastic displacements of a structure into several so-called shape functions is an established concept forming the basis of the finite element method and the Galerkin method for solving boundary value problems. In the present model, the dam deformations due to the gravity load and the water load are used as shape functions as these have direct physical meaning. The new concept is that the shape functions are not obtained from a linear-elastic analysis but from an inelastic analysis and the measured deformations that have to be fitted are of inelastic nature.

The main steps of the procedure adopted for the analysis and calibration of the dam displacements are outlined below:

•A two-dimensional plane-strain finite-element model that included all relevant material zones in the dam and the foundation was prepared. In this model, relative sliding displacements governed by the Coulomb’s friction law could occur along the core-filter and filter-rockfill interfaces on both the upstream and downstream sides of the core.

•The dam displacements were calculated for the principal load cases: (i) gravity load and (ii) water load.

•The measured displacements, D, on the dam surface were fitted with the following linear combination:

D = a G + b W (1)

where: G = computed displacements due to gravity load (including buoyancy), W = computed displacements due to water load, a, b = time-dependent coefficients or multipliers

•The coefficients a and b were determined by the method of least squares. For simplicity, the displacements G and W were taken for a fixed reservoir level of 535m asl.

•The material properties of the various zones and the angles of sliding friction of the material interfaces were adjusted in the computational model taking into account the field and laboratory test results. The aim was to achieve the best fit of the geodetically measured displacements from July 1990 to June 1994 with the calculated dam displacements expressed in the form a G + b W, so that the coefficient b was equal to about 1.0. In other words, when the water load corresponding to a reservoir level of 535m asl was applied to the calibrated dam model, the computed displacements were about the same as the estimated contribution of the water load to the measured displacements until June 1994, i.e. during the first filling of the reservoir.

•The ongoing creep deformations were represented by coefficients a and b, expressed as gradually increasing functions of time. The increasing values of these coefficients could also have been expressed in terms of an equivalent reduction of the stiffness of the fill materials with time.

•The trends of the time-dependent coefficients a and b were fitted by regression analysis with exponential time functions that approached asymptotic values. This type of creep response corresponds to that of a Kelvin viscoelastic model.

•The expected dam displacements in the future were predicted based on exponential trends of the time-dependent coefficients a and b

This method has the advantage that the time-dependent inelastic deformations can be characterised by two time-dependent coefficients only. The spatial information is contained in the two deformation shapes G and W obtained by the static analysis of the dam subjected to the gravity and water loads. It is implicitly assumed that the rate of creep deformation is uniform over the whole dam body. In reality, the rates of creep deformations could be different in the various material zones in the dam.

**Finite element analysis**

The following basic assumptions were made for the finite element analysis:

•A two-dimensional plane-strain model of the central cross-section of the dam (figure 2) and its foundation was used.

•The various zones in the dam were modelled as Mohr-Coulomb materials (elasto-plastic material).

•The Coulomb’s law of friction was assumed along the core-filter and filter-rockfill interfaces. The coefficient of friction of each sliding surface was treated as a constant over the whole dam height.

•Effects of inelastic deformations and uplift in foundation rock were assumed to be negligible.

•As this paper deals only with the post-construction deformations taking place under the self-weight of the whole dam, the gravity load was applied in one step to the whole structure. The buoyant unit weights were used in the case of the submerged zones.

•The full hydrostatic pressure from the reservoir was applied on the upstream face of the core. This is supported by the information obtained from the pore pressure measurements, which indicate that the seepage process has hardly developed through the clay core due to its very low permeability.

The finite element analysis was carried out with the computer program ADINA (ADINA R& D, 1996). The finite element model was nonlinear because of the elasto-plastic material model and the contact algorithm used to model slippage along the core-filter and filter-rockfill interfaces. In the dam analysis, first, the gravity load was applied to the whole dam, and then the upstream face of core was subjected to the water load. The water load displacements were obtained by subtracting the displacements due to the gravity load from those due to the combination of the gravity and water loads.

**Discussion on computational model**

The displacements computed with the help of the calibrated finite element model agree well with the observed displacement pattern based on the geodetic monitoring of the downstream slope as well as the crest region. It is clear that the dam behaviour can be simulated quite realistically using an elasto-plastic model in which the core is allowed to move relative to the filters along the material interfaces (figure 3).

The time-dependent or creep displacements occurring in the dam under almost constant loading conditions are nearly proportional to the displacements that occurred during the first impoundment. The results of the sensitivity studies indicate that the stiffness reduction of one particular zone alone cannot explain the observed time-dependent displacements in the whole dam. For instance, the decreasing stiffness of the core alone could not be the reason for the increasing settlements and horizontal displacements along the downstream slope; the creep deformations in the downstream shell itself are the most likely cause of the observed displacements of the downstream slope. On the other hand, the relatively large ongoing settlement of the crest cannot be the result of the creep processes in the downstream shell.

Creep deformation commonly occurs in a rockfill material due to particle breakage and rearrangement induced, for example, by wetting (Naylor et al., 1997; Justo and Durand, 2000). The post-construction settlement of the downstream shell could be taking place under the self-weight of the dam and the influence of the percolating rainwater; therefore, it could be a rather slow process. On the other hand, this process could have been mostly completed in the upstream shell already during the first impoundment, when it became submerged under the reservoir water.

The time-dependent deformations in the clay core could be related to the ongoing consolidation. As the permeability of the core is quite low, the dissipation of the excess pore pressures that developed during the dam construction is, however, progressing very slowly. Hence, the creep deformation could go on for a long time in the core.

**Prediction of future behaviour **

The displacements that have occurred in the dam since 1994, when the reservoir level first reached 535m asl, are primarily due to the creep deformations occurring under more or less constant loads. Some deviations from the constant loading condition have, however, occurred during the lowering of the reservoir by about 8m in 1997-2000 and the crest reinstatement works carried out in 1997-1999.

The time-dependent dam displacements since 1994 were assumed to have exponential asymptotic trends. Hence, the multipliers a and b were fitted with exponential asymptotic time functions by regression analyses (figure 4). These functions were then utilised to predict the expected future displacements.

The predicted displacements along the dam surface for the period July 1990 – December 2010 are shown in figure 5. The radial (horizontal) displacement at the top of the downstream slope would increase by about 0.5m over the period 2000-2010, provided that the reservoir level would not exceed about 535m asl. Figures 4 and 5 also show the possible effect of a rise of the reservoir level from 535m to 542m asl, in which case, the additional radial displacement over the same 10-year period would be up to about 2.3m at the top of the downstream slope. The incremental radial displacement due to the rise of the reservoir level decreases almost linearly from the top to the bottom along the downstream slope.

The additional settlement of the dam crest above the core top is expected to be about 1.1m over the period 2000-2010. The rise of the reservoir level from 535m to 542m asl has only a small effect on the settlement of the core top in the computational model.

**Comparison of predicted and measured deformations**

The present model for the prediction of the dam displacements was calibrated in 2001 on the basis of the available displacement data from the geodetic surveys completed until 2000. The dam displacements that have been geodetically measured since then are regularly checked against the values predicted in 2001. The new displacement data since 2001 up to now have shown very satisfactory agreement with the predicted values, as illustrated in figure 6 for benchmark 133 (position shown in figure 5) in the central cross-section.

**Summary and conclusions**

The main results and the conclusions are as follows:

•The ongoing displacements measured on the dam surface over any given period can be expressed as a linear combination a G + b W of the displacements G and W computed under the gravity and water loads respectively, where a and b are time-dependent coefficients or multipliers.

•The multipliers a and b were determined using the method of least squares, based on the geodetic measurements of the displacements along the downstream slope only. Nevertheless, the displacements of the crest region computed with the calibrated model, in the form of the combination a G + b W, are consistent with the measured settlements and horizontal displacements of the crest region, as well as the measured settlements of the top of the core.

•The good agreement between the measured and computed displacement patterns shows that the dam behaviour can be realistically simulated using an elasto-plastic model in which the core is allowed to move relative to the filters along the material interfaces.

•The large settlement of the crest region is related to the substantial post-construction settlement of the clay core, taking place primarily under the effect of the gravity load (self-weight). The core tends to move downwards relative to the substantially stiffer filter zones on both sides, in particular, along the interface between the core and the downstream filter.

•The horizontal (radial) displacements of the downstream slope of the dam can be predominantly attributed to the hydrostatic pressure exerted by the reservoir water on the upstream face of the clay core.

•Rockfill materials commonly undergo creep deformations because of settlements induced primarily by wetting. Such settlements result from crushing of sharp corners at the contact points, and the rearrangement of the rockfill pieces owing to the lubricating effect of water as well as the stress redistribution in the upstream shell during impoundment.

•The settlement of the downstream shell could be taking place under the self-weight of the dam and the influence of the percolating rainwater; therefore, it could be a rather slow process. On the other hand, this process could have been mostly completed in the upstream shell already during the first impoundment, when it became submerged under the reservoir water.

•In the clay core, the time-dependent deformation may be related to the ongoing consolidation process and possibly some reduction of the Poisson’s ratio. As the permeability of the core is quite low, the dissipation of the excess pore pressures that developed during the dam construction is progressing very slowly. Hence, the creep deformation of the core could go on for a long time.

•The results of the sensitivity studies clearly show that the stiffness reduction of one particular zone alone cannot explain the observed time-dependent displacements in the whole dam. For instance, the decreasing stiffness of the core alone could not be the reason for the increasing settlements and horizontal displacements along the downstream slope. The creep deformation of the downstream shell itself is the most likely cause of the observed displacements along the downstream slope. On the other hand, the relatively large ongoing settlement of the crest cannot be the result of the creep processes in the downstream shell.

•It is estimated that the additional radial (horizontal) displacement at the top of the downstream slope over the 10-year period from 2000 to 2010 will be about 0.5m if the reservoir level would not exceed about 535m asl.

•A rise of the reservoir level from 535m to 542m asl could cause the radial displacement at the top of the downstream slope to increase by up to about 2.3m during the period 2000-2010. The incremental radial displacement due to the rise of the reservoir level decreases almost linearly from the top to the bottom along the downstream slope.

•Based on the expected future settlements predicted using the present model, the additional settlement of the dam crest above the core top is expected to be about 1.1m during the period 2000-2010. The rise of the reservoir level from 535m to 542m asl has only a small effect on the settlement of the core top in the computational model.

•The present model for the prediction of the dam displacements was calibrated in 2001 on the basis of the geodetic measurements made until 2000. The dam displacements that have been measured biannually since then agree very well with the values predicted in 2001.

Author Info:

S. Malla, M. Wieland, R. Straubhaar, Poyry Energy Ltd , Hardturmstrasse 161, CH-8037 Zurich, Switzerland. Email: Martin.wieland@poyry.com

The authors are grateful to DSI for permitting publication of the paper. The investigations and studies described in this paper were carried out in cooperation with DSI and Dolsar Engineering Ltd. The contributions of other experts, who have participated in this project and are not listed explicitly, are greatly acknowledged. The opinions expressed in this paper are those of the authors and are not necessarily those of DSI