Monitoring the deformation and strain analysis on the Ataturk Dam, Turkey

ABSTRACT Every man-made structure creates certain risks — dams are no exception. Most failures in man-made structures that have occurred could have been avoided if the structures’ behaviour had been inspected, monitored, and analyzed continuously, and if proper corrective measures had been taken in a timely fashion. The DSI (The General Directorate of State Hydraulic Works), which is the institution responsible for dam safety, has long used surveying methods to measure the displacements of geodetic points as a part of dam monitoring policy. In this study, we focus on the dam's mechanical behaviour throughout a time period of more than 10 years. These study results have been derived from a separate, ongoing project that has monitored deformation on the Ataturk Dam and is now determining the water level of the reservoir. The project results show that although the dam body has become more stable and the water load behind the dam has increased, the rate of displacement of the dam has declined significantly. From these results, it can be seen that the reservoir water level can be increased evenly over time and that 542 m is the maximum water level of the dam's reservoir.


Introduction
Structures such as dams are important for water supply, flood control, agriculture, and hydroelectric power production. Numerous dams are constructed worldwide for these purposes. With the development of construction technology, especially in the 1970 s, dams with huge capacities have been built. By the year 2000, there were 45.000 grand dams and 800.000 small dams constructed globally (Ponseti and L opez-Pujol 2006). Dams with crest heights taller than 300 m and reservoirs above 200 million m 3 have been constructed in the last century. The dam with the highest crest (335 m) is the Rogun Dam in Tajikistan (Singh 2013). Of these dams, the one with the largest capacity, The Synrude Tailings Dam, was built in Canada with a 5.40 billion m 3 reservoir capacity. But after that, a new dam was constructed in China, the Three Gorges Dam; it has the largest reservoir, with 39.3 billion m 3 greater than the Synrude Tailings Dam, and 22.500 MWs of hydroelectric power production capacity (Behr et al. 1998;Li and Wang 2011;Boulanger and Montgomery 2016).
These structures have certain risks, just like other man-made structures. Hazard assessment of the increasingly aging dams is mandatory, yet complex procedure. There are some challenges with these evaluations, especially if the original installation, design, and construction details are unknown or not defined clearly. A safety assessment of the dams involves filed surveillance, periodic safety observations, operation procedure tests, scheduled maintenance, and preparation for emergency situations. Risk-based dam safety procedures include traditional safety management applications with the integration of recognition on analysis of risks, and formal assessment of the uncertainties. Dam safety analysis relies on having a philosophy that is able to represent a wide range of risks and uncertainties in a practical way. This includes the engineering and scientific characterization of these parameters by using available techniques and monitoring physical performance in their natural environment. Such physical data present the most important issues while evaluating the safety and performance of both concrete and embankment formation dams (Liu et al. 2016). These issues are movement, water pressure, seepage, reservoir and tail water elevations, local seismic activity, total pressure, stress, strain, internal concrete and environmental temperature, and precipitation (Jansen 1983;ICOLD 1992). These physical data are measured and monitored frequently to acquire information necessary for analyzing and defining a problem. For example, the downstream movement of a dam due to high reservoir water pressure must be analyzed to determine whether the movement is distributed uniformly along or in the dam, the foundation, or both, and to assess whether the movement is increasing, decreasing, or stable. Such information can be used to determine proper corrective measurements (Manake and Kulkarni 2002;Chen et al. 2016;Kan and Taiebat 2016).
Ataturk Dam is 169 m in height and has a total volume of 84.500 hm 3 , whereas the reservoir's volume is 48.700 hm 3 and area is 817 km 2 . The approximately 900.000 km 2 area in Harran Plain, Sanliurfa is irrigated by the water provided by the dam using tunnels and pipe stations. Moreover, the drinking and usable water ofŞanlıurfa is obtained from the Ataturk Dam. The dam also feeds a hydroelectric power plant which provides a significant amount of energy to Turkey. The capacity of the dam is 2.400 MW with 8.9 billion KWh of annual power production. Ataturk Dam in the northwest ofŞanlıurfa, which is the biggest dam in Turkey and 11th worldwide according to its capacity, was evaluated with data derived with various techniques since 2006. Evaluation of the velocity and strain field of the dam is discussed and presented in this paper.

Deformation monitoring on dams
The monitoring of dam performance is a critical element in producing and maintaining a safe dam. Monitoring programmes typically consist of surveillance or visual observation, instrumentation, data collection, data evaluation, analysis, management, and decisions/response based on results (FEMA 2015). Surveillance and monitoring programmes on dams provide information for evaluating the dam's performance with respect to intention of design and expected behaviour. Monitoring the changes could affect the safety performance assessment of a dam, can assist the investigations and evaluations of abnormal dam performance, and determine if maintenance is necessary. Dams and their surrounding areas can be monitored with a variety of techniques to find out if horizontal and/or vertical deformations will arise in time. Because these data focus on particular areas of the dam that have moved, inspectors are alerted to look for potential problems during the visual inspection. Movement can cause damage or structural distress in concrete dams, and cracking and sliding in embankment ones. Dams and their environments should be monitored on a regular basis by available methods, which include geodetic and non-geodetic measurements. Non-geodetic survey devices consist of settlement sensors, inclinometers, extensometers, tiltmeters, plumblines, crack and joint measuring devices, stress-strain-joint and internal temperature meters in the baseplate of the embankment, and crossarms on the structure. Generally, movement and/or deformation of the structure are monitored and can be detected by geodetic methods. The type of dam and the predictive deformation are taken into consideration when determining the type of surveillance, time period of the surveillance, and related standards.

Deformation monitoring of Ataturk Dam by using geodetic methods
Geodetic methods for monitoring dam deformations range from terrestrial/conventional observations to space-based measurements like GNSS and interferometry. Different equipment and devices have been used based on the geodetic methods, such as theodolites for alignment measurements, total stations for conventional location measurements, levels for levelling surveys, GNSS receivers for space-based measurements, and SAR images for interferometric applications (Manake and Kulkarni 2002;USACE 2002;Tasci 2008;Kalkan et al. 2010;Yigit et al. 2016).
In this study, deformation monitoring of the Ataturk Dam has been performed periodically to determine the horizontal displacements using conventional (terrestrial) measurements from 2006 to 2013 and satellite-based (GNSS) techniques from 2007 to 2013 (11 periods for Terrestrial Measurements & 9 periods for GNSS, Table 1). The 25 reference stations outside the deformation zone have been established, and deformations on the dam embankment and the surrounding area have been evaluated with respect to these points. Potential deformation zones are represented with 217 object points (deformation points) (Figure 1(a,b)).

Terrestrial measurements
Terrestrial measurements may include conventional geodetic measurements which involve distance, horizontal, and vertical angle observations by optical/electronic total station instruments (Lambeck 1979). In this study, angle and distance measurements were performed with the TDA 5005 Total Station, the Wild T3000 electronic Theodolite, and the Distomat DI3000.
Although there were 25 reference stations, angles and distance measurements were carried out with conventional terrestrial methods on 13 of the stations closer to the dam in the study. Angles were measured with three complete sets, and distances were obtained conjugately. Two hundred and seventeen object points' coordinates were calculated from 13 reference stations to establish correlation between them. Angles to the object points were observed within the two faces of the instruments.

GNSS measurements
GNSS is a powerful technique to determine deformation on engineering structures, e.g. dams and bridges. GNSS has become more common due to its accuracy and cost-effective structure, making it preferable to classical surveying techniques in deformation analysis. Receivers that can record GNSS data were used in this project (Table 2). GNSS observations on the 25 reference stations were conducted with at least 10 h of observations, 5-s data interval, and 10 of elevation mask. On the other hand, object points were measured based on at least 5 reference stations for 40 min long with the same properties of data interval and elevation mask ( Figure 2).
Reference stations using both terrestrial measurements and GNSS survey are constructed as pillars with a force-centred design to eliminate blunder errors. The force-centred components were used to fix the GNSS antenna on object points ( Figure 3).
To process the GNSS data, Leica Geo Office software was used with the IGS final satellite orbit information. In the first step, the raw data were converted to RINEX format and the final orbits were downloaded. In the next step, reference stations were processed internally. After obtaining the control of the reference stations when they have no deformation between two periods, the object points were evaluated with respect to these reference stations. Finally, the results were transformed according to the first epoch of observations.

Conventional deformation analysis
Conventional deformation analysis (CDA) method in a geodetic network is explained in detail by Hekimoglu et al. (2010); it is based on the difference in coordinates between two observation epochs. These differences, called 'displacements,' occur when the expected movements are rejected by statistical tests. The reference network in this study is adjusted as a free network for each epoch.
The Gauss-Markov model is simply explained as follows: and where the subindices are, 1 = first epoch, i = second (last) epoch, A = design matrices, x = vector of estimated coordinate, P = the weight matrix, l = observation vector, and v = residual vector.
In the application of CDA, a global congruency test is implemented to evaluate if there is any significant displacement between two epochs. According to the model, if the coordinates of the corresponding points between two epochs changed according to expectations, they form the null hypothesis.

Hypothesis
and is tested against its alternative where E stands for 'expectation.' It is clear that the datum between two epochs must be the same. It is managed by taking the approximate coordinates of the identical stations as the same in both epochs. The adjustment of each epoch is carried out as a free network and the points' coordinates (assumed identical) are determined as unknown. Then, the influence of the null hypothesis on the least squares estimation (LSE), in the absence of correlations between epochs, results in (Pelzer 1971;Koch 1985;Niemeier 1985;Pelzer 1985;Cooper 1987) under the assumption of normally distributed observation errors. In the equations above, Q x1x1 and Q x2x2 = cofactor matrices; f = f 1 + f i = sum of the degrees of freedom at either epoch; d = differential vector of estimated coordinates; Q dd = cofactor matrix of d and Q dd + its pseudo inverse; h is the rank of the matrix Q dd ; and a = chosen error probability. Here, s 2 0 denotes the estimated variance component in the absence of the null hypothesis. If T > F h,f,a , then the null hypothesis is rejected. Therefore, the result of an unexpected displacement is obtained from the difference in the coordinates between two epochs. After that, its localization is carried out. There are numerous localization methods for deformation analysis in the literature. In this study, the S-transformation method was used (Baarda 1973;Niemeier 1985;Cooper 1987;Welsch et al. 2000).
After the adjustment and estimation for each epoch, the results were compared with sequential couple-epochs to decide if these epochs should be combined. The double period analysis is implemented with a stochastic test for each period.
The test value (f) in the Fisher distribution is defined as where m 2 1 and m 2 i are mean-square-error values of unit measure for the first and last (ith) period, respectively. The test value for Equation 7 was calculated with a 0.05 margin of error. Table values F (f 1 ; f i ), where f 1 and f i denote degree of freedom for the first and last period observations, respectively, were compared to see if the condition (f L < F L ) for the levelling results was met. When the levelling condition was procured, the displacement vectors were determined by Equation 3, and their test criteria were calculated according to the rule and their test criteria were calculated according to the rule test value (Hoover and Rockville 1984), We used the test values computed from Equation 9 to calculate the significant displacement (dPi > Ti).
Eleven epochs of conventional terrestrial geodetic observations and nine epochs for GNSS observations were carried out through 2006-2013. The same equipment and hardware were used through similar observation methods. First, double period analysis was used to find out whether stations on the reference network had moved or not. Then, observations on the object points based on the reference stations, which were accepted to have no movement, were adjusted, and double period analysis was implemented based on the results. Displacement vectors for the compared periods and their accuracy criteria were calculated according to the given formula (8). Calculated displacement was tested with the formula (9) to find out whether there was significant movement, and 90% of the object points were defined to have meaningful horizontal movement. These movement vectors' perpendicular components to the crest axis (radial displacement vector) are about 84% meaningful. The displacement vectors for the 30 object points having maximum deformation on the dam body are given in Figure 4 and Table 3, and all displacement vectors are shown in Figure 5.

Strain analysis of dam
Strain is strictly related to displacement and is the displacement with the adjacent points in physical and mechanical bonds. Displacement can be considered as the difference in the Cartesian Coordinates of the same point at different times. The correlation between coordinates of a point at time t1 and time t2 under homogenous deformation can be explained with a simple affine transformation (Brunner 1979).
With this approach, variance for the relative location can be obtained by (Kakkuri and Chen 1992) du ¼ ðE þ wÞr þ t 0 (10) Figure 4. Location of the 30 object points having maximum displacement vectors.  where E is the symmetric strain tensor, w is the inverse symmetric strain tensor, t 0 is the translation solid block movement for all the points, and r is the coordinate. Also, translation can be explained by Symmetric strain tensor is obtained by Diagonal elements of the symmetric tensor define the dilation along the coordinate axes, and other elements off the diagonals define the small, angular disturbances through deformation according to the coordinate axes. (Do gan 2017). The inverse symmetric tensor is the differential rotation of the solid block due to deformation for the plane including coordinate axes.
If d u deformation vector is the result of the affine transformation within the epochs t 1 and t 2 , it can be calculated as (Brunner 1979) The matrix of the coefficients B is shown as If the Equations (14) and (15) readjusted according to (12) and (13) and were written in the Equation (16) (Prescott 1976;Brunner 1979;Lambeck 1988;Deniz 1990; Do gan 2017) Strain parameters are obtained by the least squares method. To calculate the strain analysis, the velocity field of the deformation area on the dam was derived from GNSS data ( Table 3). As given in Table 3, the results of the deformation analysis obtained from both GNSS and terrestrial measurements are very coherent. Differences between the two methods range from ¡2.4 cm to 1.2 cm.
The strain analysis of the dam that was conducted by using its velocity field was processed by Grid Strain Matlab Toolbox. The software can calculate the deformation (rate) field for an area using displacements/velocities of points covering the whole study area. Grid strain provides the intensity and direction of the principal components of strain tensor in each point of the grid, as well as the significance of the results obtained on the basis of spatial distribution and accuracy of available measurements (Pesci and Teza 2006;G€ ulal 2013).
After the process, the obtained results (the strain field tensors [consisting of the principal strain directions, Eigen vectors, and the Eigen values in each computation point], change in the area like the strain tensor trace, the prevalent Eigen value, and shear normalized strain trace) are shown as contours (Teza et al. 2008). The software automatically computes the grid length based on station baselines, generally on the spacing of the experimental points. The standard deviation of all the inner distances between the pairs of points were computed and assumed as the default values. The programme automatically defines the smoothing parameter, or scale factor, for the modification of the least square weighting matrix as given in Zhu-Jiang and Gang (1996). All the available data are involved in computation, but errors are rescaled using an appropriate function, which increases with distance. The detailed explanation of this programme is available in Teza (2006) andG€ ulal (2013).
2D strain analysis was executed in this study. The study area was divided into 10 £ 10 m grids, and strain fields for the corners of the grids were calculated. The arrows in Figure 5 show the temporal displacement vectors. Grid strain software calculates the amount of deformation between GNSS points iteratively. Data density and magnitude of the deformation can affect the displacement vectors given in Figure 5.
Maximum water pressure generally occurs in the middle of the dam body. Pressure in the middle is transferred to the edges with the help belt construction. When examining Figure 6, it can be seen that the maximum strains are in the middle and right-middle points. Vertical regimes are active, especially with the water pressure in these regions. This result is compatible with deformation analyses. Also according to Figure 6, the right-middle part of the dam body (ellipse) and the old stream bed have the largest deformation. Additionally, there is another zone to the left of the body (circle) where deformation occurred. These two zones are the regions where the most movement was observed after the construction of the dam, so deformation points in these regions appear dense when monitoring the deformation. (Figure 7).

Discussion and conclusion
In this study, to monitor the rate variation of the deformation, we used two methods (terrestrial and GNSS) to obtain the data as given in detail in Section 3. The dam is more stable with respect to about 20 years of terrestrial measurement results (Figure 8).
The determination of the strain field through the downstream and upstream surfaces should contribute to the security of a dam. Different results for the magnitude of the strain values on various  surfaces can be examined. Correlation between calculated strain fields obtained with different techniques can be analyzed to find out necessary precautions.
In this study, a total of 11 epochs of deformation observations using the terrestrial method and 9 epochs for GNSS methods were carried out between May 2006 and November 2013 under the 'Monitoring the Deformation on Ataturk Dam with Geodetic Method Project.' Data from the field work were evaluated through deformation and strain analysis, and the results are: According to the strain analysis, the biggest deformation was on the old stream bed. These results are compatible with the results of deformation analysis. According to the strain analysis results, the water level can reach a maximum retention level of 542 m.
According to the classic deformation analysis results for the 11th observation epoch, which was completed in November 2013, 91% of the object points have experienced significant horizontal movement for the last seven years. These results are also compatible, as with the results for embankment dams in these studies (Manake and Kulkarni 2002;Kalkan 2014;Kalkan et al. 2016).
According to these results, the largest movement was on the mid-section of the dam (Figure 4). On point 2082, which is on this section and has a close altitude with the crest, there is a total of 21.1 cm movement on the upstream-downstream direction between 2006-2013 (2nd period-90 months). The movement on the same point between 1997-2005 (1st period-84 months) is given as 87.2 cm (DSI 2005). The monthly average speed on the point for the first period is 10.4 mm, whereas it is 2.3 mm during the second. Average reservoir elevation was determined as 531.0 and 533.8 m for these periods, respectively, and the second period average was 3 m higher than the first. Despite the increase in water level, the deformation velocity on this object point decreased (Pytharouli and Stiros 2005).
In conclusion, it can be seen that dam embankment is now more stabilized, and the movement significantly decreased despite the water load behind the dam reservoir ( Figure 8). Based on this, the reservoir water level can be increased gradually over time and will approach 542 m, the maximum water level of the dam. This is important progress. Thus, a 1 cm increase in the reservoir water level is almost equal to 8 million m 3 of water. Continuing to monitor this dam, which is used for irrigation, potable water, and the generation of energy, using data from different techniques is an essential approach.