Geomembrane as an Upstream Impermeable Blanket of Embankment Dams-Laboratory and Numerical Study

The use of geosynthetics has been a common practice in geotechnical engineering when the improvement of local soil characteristics is necessary. This paper presents an experimental and numerical study of the performance of HDPE geomembrane as impermeable blanket installed upstream of embankment dams, a treatment technique for very permeable foundation. Data based on project, field and laboratory tests of the Brazilian Salto Hydroelectric Power Plant were presented to gather information about the constructive method and to support further studies. A small-scale seepage model was constructed to represent the transverse section geometry of a hypothetical embankment dam, whose geometry was adopted based on Salto Hydroelectric Power Plant. Likewise, a numerical model was elaborated referring to the small-scale seepage model to perform several parametric analyses. The obtained results indicated that the geomembrane sealing system acts increasing the flow path through the dam foundation, resulting in lower pore-pressures into the dam. Additionally, the hydraulic parameters in the soil mass may vary considerably in case of damage to the geomembrane. In general, the study has shown that the use of synthetic membranes can be a good solution for treatment of pervious foundations and possible defects can lead to the reduction of their performance.


Introduction
Embankment dams are normally large, complex, and costly works.Studies of alternatives to improve technical characteristics and to reduce construction cost of dams are important.One current alternative for improving soils, which has been common in geotechnical engineering works, is the use of geosynthetics.These materials have additional advantages compared to traditional materials as, for example, lower costs, shorter execution time and better ease of installation (Shukla, 2002;Koerner, 2012;Nicholson, 2015).The increasing use of geosynthetics in recent years has attracted the attention of engineers and researchers around the world, since those materials have peculiar characteristics when compared to traditional geotechnical materials.For this reason, new researches and design models became necessary, especially when geosynthetics are applied in embankment dams.
Geomembranes are among the most common types of geosynthetics.They are defined as a very low permeability synthetic membrane used with any geotechnical engineering related material, with the purpose of controlling liquid or gas migration in a human-made work, structure or system (ASTM, 2015).High density polyethylene (HDPE) geomembranes are commonly used for the construction of reservoirs as liners for water, chemical products, mining tailings, among others (Giroud & Bonaparte, 1989;Tao et al., 1996;FHWA, 1998;Whitfield, 1996;Poulain et al., 2011, for instance).However, the efficiency of these barriers depends on the integrity of the synthetic membrane (Giroud & Touze-Foltz, 2003), in addition to other aspects, like the service life (Needham et al., 2006).
Some researchers have studied aspects related to dam failure (e.g.Mahinroosta et al., 2015;Petaccia et al., 2016).Failures in dams and reservoirs protected by geomembranes have also been reported in the literature (e.g.Wu et al., 2008;Messerklinger, 2014;Dong et al., 2016;Bhowmik et al., 2018) indicating the need for further studies.
Additionally, geomembranes can also be applied for the construction of impermeable upstream blankets for embankment dams over permeable foundations.Scuero & Vaschetti (2004) pointed out that PVC geomembranes may be installed in the upstream portion of dams in order to minimize uncontrolled water presence inside the dam, improving safety.Among the existing literature about this kind of application, Cardoso et al. (2010) studied the project constraints and performed a numerical analysis that supported the application of the foundation treatment in the São Salvador Hydropower Plant.
This paper presents a research (Pierozan, 2014) whose main objective is to evaluate the efficiency of foundation treatment of embankment dams by upstream imper-meable geomembrane blankets.Firstly, data from field and laboratory tests of the Brazilian Salto Hydroelectric Power Plant are briefly presented to bring together information concerning the constructive method.A cross-section of the dam was selected to represent the overall behavior of the dam and to support further studies.Based on this crosssection, a small-scale seepage model was constructed to represent the behavior of a dam with geomembrane as upstream impermeable blanket, allowing the researchers to calibrate significant parameters.This small-scale model was built taking into consideration the characteristics of Salto Hydroelectric Power Plant, however it is a simplification of the prototype.Finally, a numerical model was elaborated based on the small-scale seepage model and considering the calibrated parameters, which permitted several parametric analyses.

Case History Used as a Reference for the Model
The Salto Hydroelectric Power Plant, located in Rio Verde, belongs to the Paraná River Basin in the state of Goiás -Brazil.The plant started its operation in 2010 with two generation units and an installed capacity of 116 MW. Figure 1A shows a photograph of the dam.This type of solution has been used in just a few dams around the world (e.g.Salto Hydroelectric Power Plant and São Salvador Hydroelectric Power Plant) and limited information regarding this kind of foundation treatment may be found in the literature (e.g.Cruz, 2004;Cardoso et al., 2010).For this reason, studies that consider the use of geomembranes as upstream impermeable blankets of dams are very important for the advancement of the knowledge on the topic.
The left side of the earth dam has a crest with approximately 580 m length and a maximum height of approximately 25 m.Due to the geotechnical properties of the dam foundation soil, an impermeable blanket was executed upstream on the left side of the dam.A HDPE geomembrane has been used as a liner to reduce the water flow throughout the dam foundation.The applied geomembrane is a 1.5 mm thick flexible synthetic HDPE membrane, textured on both sides.According to the dam designers, a textured geomembrane was chosen in order to avoid slippage between geomembrane and compacted soil.
The setting of the geomembrane on the foundation soil was made by means of a previously excavated anchor trench 0.80 m deep and 0.50 m wide, as presented in Fig. 1B.In both cases, backfill compaction was done by hand-operated equipment near the geomembrane.Several procedures were observed to prevent damaging the membrane.The geomembrane anchor trench at the upstream face of the dam is 2 m wide and was executed after the embankment construction.These geometric attributes were based on the experience of the engineers involved in the design and further explanation regarding the anchor trench at the upstream face of the dam is discussed in this paper.
The seams between geomembrane panels were executed by using a dual welding process through the application of heat melted polymer, forming an air-tight channel between the weld lines.From the channel, it was possible to identify eventual defects, check the quality of the sealing procedure and fix the eventually identified defects.
Aiming to protect the geomembrane and to mitigate the effects of an eventual local failure, due to some mechanical effect, a 50 cm layer of compacted soil was executed over the geosynthetic.Over the compacted soil layer, an additional protective 50 cm thick layer of compacted rockfill was constructed.

Data source and analyses
The data used in this research were obtained from contractors documentation such as project drawings, topographic survey and results of field and laboratory tests.In Table 1, the main geometric characteristics of the left abutment of Salto Hydroelectric Dam are presented.
The instrumentation of Salto Hydroelectric Dam consists on standpipe piezometers, V-notch flow meters, water level indicators and surface marks.The V-notch flow meters are responsible for measuring the water flow from the internal drainage system.Five sections on the left side of the dam were instrumented.
The thickness of the permeable foundation soil layer was determined based on field permeability tests in boreholes located in several points on the left side of the dam foundation.The adopted hydraulic conductivities for foundation layers were the average values of several tests for each layer.The anisotropy relative to permeability, due to the constructive process of the embankment, was determined by the ratio between the horizontal hydraulic conductivity (k h ) and the vertical hydraulic conductivity (k v ), as suggested by Cruz (2004).
The vertical and horizontal average hydraulic conductivities of the embankment dam used in the analyses were determined from results of laboratory permeability tests.Based on tests performed in samples from undisturbed blocks of dam embankment, an average horizontal coefficient of permeability equal to 1 x 10 -5 cm/s and a vertical coefficient of permeability equal to 2 x 10 -6 cm/s were obtained.Thus, the horizontal permeability is approximately five times greater than the vertical permeability.The foundation bedrock consists mainly of basalt, covered by its weathered products.The average hydraulic conductivity of foundation bedrock is equal to 1 x 10 -6 cm/s.

Small-Scale Seepage Modelling
Laboratory tests for soil geotechnical characterization and small-scale modelling were performed at the CESEC/UFPR (Center for Studies on Civil Engineering/ Federal University of Paraná) facilities.The small-scale model consisted of a percolation tank filled with sand and other materials in the interest of representing the crosssection geometry of a hypothetical embankment dam, in the scale 1:100, taking into consideration some properties of Salto Hydroelectric Dam.

Geometric and boundary conditions of modelling
The geometry of the cross-section model was defined based on geometric characteristics of Salto Hydroelectric Dam, such as upstream and downstream slope inclination, crest width, dam height and thickness of foundation layers.It is important to highlight that the small-scale model was a simplification of Salto Hydroelectric dam instrumented cross-sections and, consequently, average parameters were adopted.The model itself is not equivalent to any of the instrumented sections and reproduces the overall observed behavior.The small-scale model did not have the objective to be the same as the prototype, since the prototype was anisotropic with properties ranging in the three dimensions (3D), and the small-scale model was isotropic and properties ranged just in two dimensions (2D).Considering that the small-scale model represents a hypothetical dam and not exactly Salto Hydroelectric Power Plant, the smallscale model has not the same hydraulic characteristics as the prototype.
Soils and Rocks, São Paulo, 42(1): 3-19, January-April, 2019.5 Geomembrane as an Upstream Impermeable Blanket of Embankment Dams -Laboratory and Numerical Study The length of the treated area with geomembrane on the foundation was defined from geometric data of several instrumented sections of the Salto Hydroelectric dam.Thus, the ratio between the length of upstream dam foundation treatment (L) and dam height (H) was reproduced in the small-scale model (L = 4H).The internal drainage system used in the model consisted of a vertical filter and blanket drain and had similarity with the dam prototype.
The boundary conditions imposed to the small-scale model were: a) reservoir water level, b) downstream water level and c) restriction to flow through the upstream impermeable blanket.The upstream water level was set to represent the maximum normal water elevation in Salto Hydroelectric reservoir (Elevation 446.50).

Model instrumentation
Internally, the tank was 250 cm long, 60 cm high and 45 cm wide, resulting in a volume of 0.75 m 3 .Flow visualization could be done by a lateral plexiglass wall.Water percolates through the embankment and foundation of the model as in the prototype.The flow was collected by a water outlet located at the end of the tank, allowing determination of the flow rate.
The seepage tank was instrumented with piezometers to determine the pressure head at different points.The interpretations of total heads were made by means of a reading panel.Figure 2 shows the piezometers location in the model.In this paper, piezometers in the embankment are named as PE and piezometers in the foundation are named as PF.

Geotechnical materials
The materials used for simulating the embankment and foundation soil on small-scale models were submitted to some tests, such as particle size distribution, specific gravity, permeability and maximum and minimum void ratio.The material used as drain layer was submitted to particle size distribution and permeability tests.Based on the results of particle size distribution tests, the suitability of materials to be applied as filter and drain was evaluated by the Terzaghi filter criterion.
It is important to highlight that the small-scale model was intended to simulate the cross-section geometry of a hypothetical embankment dam, based on the geometry and some characteristics found on Salto Hydroelectric Power Plant.The purpose of the scale-model was to understand how the use of geomembrane as an upstream impermeable blanket would impact the flow through the dam and its foundation.The small-scale model (2D) is a simplification of the behavior observed on the prototype (3D).It was not possible to build the small-scale model with the same geotechnical characteristics of the prototype, since the geotechnical properties were not isotropic in the field.Considering the construction of the small-scale model, laboratory available materials were used, and they were not the same found in the prototype.

Small-scale model construction
For construction of the small-scale model, granular material was deposited in the tank by means of a technique known as "sand pluviation".This technique gives to the soil mass a standardized condition of compaction and permeability.It consists on promoting sand precipitation in pre- established conditions, in order to obtain a material as homogeneous as possible (Rad & Tumay, 1987;Brandon et al., 1991;Lo Presti et al., 1992).
The sand deposition flow rate was kept constant by using 5 mm opening funnels.Because soil compaction also depends on the material falling height in this method, a calibration curve relating falling height and material relative density was obtained.Based on the curves relating the sand fall height with the obtained soil density, it was selected a fall height equal to 12 cm, for both embankment and foundation materials in small-scale models.This fall height was adopted once small variation of the density has been verified for greater heights.
The dam slopes were drawn in internal faces of the tank walls to geometrically orientate the construction of the small-scale dam model, drainage system and foundation.Successive layers of gauze and paraffin were applied over the dam upstream soil to represent the impermeable membrane.Figure 3 presents some photographs of the model assembly.Wooden sticks were placed temporarily within the vertical filter (Fig. 3 -B) as leveling references.

Small-scale model simulations
Three distinct scenarios were simulated, allowing to evaluate the effect of an upstream impermeable membrane over the dam: a) no foundation treatment, b) use of geomembrane upstream of the dam and c) use of damaged geomembrane upstream of the dam.For each simulation, readings of total heads and percolation flow rates were made.Figure 4 shows a small-scale model sketch.
In the model simulating the existence of defects on the geomembrane, longitudinal openings were made in the sealing material upstream of the dam model to simulate defects that can occur during geomembrane installation and at the end of the construction of the dam.The objective of this simulation was to verify if the geomembrane sealing system would be able to maintain a minimum performance even with generalized failure.According to Nosko et al. (1996), this consideration is acceptable since most of the leaks occur during the procedure of covering the liner with soil or stone, while other types of defects that could influence the system proper behavior (e.g.seam failure between geomembrane rolls) may be identified and fixed at the same time as the geomembrane installation quality control.Other studies (e.g.Rollin et al., 1999;Rollin et al., 2002;Rollin et al., 2004) also presented similar conclusions.

Results of small-scale seepage modelling
The geometric characteristics of the model are summarized in Table 2 and illustrated in Fig. 5.It is important to observe that the dimensions adopted on the small-scale model were not exactly the same from Salto Hydroelectric Soils and Rocks, São Paulo, 42(1): 3-19, January-April, 2019.7 Geomembrane as an Upstream Impermeable Blanket of Embankment Dams -Laboratory and Numerical Study  Dam, once five sections were analyzed from the prototype and just one section was analyzed in the small-scale model.
The results of the characterization tests, maximum and minimum void ratio and permeability of materials are presented in Table 3.Based on grain size distribution curves of the materials, their suitability for using in filtration and drainage was verified.
Figure 6 shows the total head values obtained on the section of the small-scale models.The piezometers were divided into 3 arrangements to simplify the analysis of the results, considering that the arrangements are referring to the same cross-section of the dam and analyzing a distinct set of piezometers.
According to Fig. 6, the presence of nondamaged geomembrane upstream of the dam model reduced the total flow through the embankment and foundation by approxi-   5 x 10 0 mately 46%.On the other hand, the presence of a geomembrane with defects reduced the water flow by only 8% when comparing to the scenario without treatment.
Based on the results obtained in Arrangement 1 (Fig. 6), the presence of geomembrane upstream of the dam model leads to lower total heads.It can also be observed that the reduction of the hydraulic heads occurs primarily upstream from the chimney drain.The presence of defects leads to a lower efficiency of the foundation treatment.
The foundation treatment with upstream impermeable blanket had little influence on hydraulic heads over the dam embankment material, as shown in Arrangement 2. On the other hand, some influence can be noticed in the hydraulic heads at the interface between the dam embankment and foundation materials, as shown in Arrangement 3 (PE-2 and PE-4).

Numerical Analysis
Based on the results of the small-scale models, a hypothetical case with the same characteristics as the laboratory model has been simulated, in prototype dimensions.SEEP/W® software (GEO-SLOPE, 2012) was used to perform numerical analyses.This software provides twodimensional analysis of groundwater flow within porous materials.
The following assumptions were admitted for the numerical analysis: a) Geometric characteristics: The geometry of the dam and its foundation corresponded to that assumed in the small-scale models, except for scale and unit width (Width = 1 m).Table 4 presents the main geometric characteristics considered in the numerical analysis.
b) Saturated steady-state flow, governed by Darcy's Law.This assumption corresponds to a constant flow rate and volumetric water content at any position below the water table.The unsaturated flow at the downstream side was disregarded in the simulations.The seepage flow rate col-lected by the drainage system was measured in two sections: one between the vertical drain and the upstream blanket of the embankment, corresponding to the flow through the embankment (Section A-A', Fig. 7-A), and another between the horizontal drain and the foundation, corresponding to the flow through the foundation (Section B-B', Fig. 7-A).The sum of the two contributions resulted in the total flow; c) Boundary conditions: The adopted boundary conditions were (Fig. 7-B) a) total head equal to 33 m in the reservoir (12 m of foundation thickness plus 24 m of dam height, minus 3 m for the free board) and b) pressure head equal to atmospheric pressure in the vertical filter and in the blanket drain.These boundary conditions are reasonable since the unsaturated flow did not represent a considerable amount of the total flow, according to previous simulations (Pierozan, 2014).For this reason, the unsaturated flow was not considered when dealing with the numerical model.In the geomembrane region, no boundary condition was applied, in other words, the geomembrane was considered impermeable.
d) Material properties: The hydraulic conductivity of the foundation was equal to 1 x 10 -2 cm/s and the hydraulic conductivity of the embankment was equal to 2 x 10 -4 cm/s vertically and 1 x 10 -3 cm/s horizontally.These values have been adopted the same as for the small-scale tests Also, it Soils and Rocks, São Paulo, 42(1): 3-19, January-April, 2019.9 Geomembrane as an Upstream Impermeable Blanket of Embankment Dams -Laboratory and Numerical Study  was considered that the embankment material had hydraulic conductivity in horizontal direction 5 times greater than in vertical direction, which is the ratio found from laboratory tests on undisturbed samples of the Salto Dam embankment.
Once calibrated with appropriate parameters from physical and experimental data, the numerical analysis could simulate several not physically evaluated conditions to study some hypothetical cases with different boundary conditions.The behavior of the flow throughout the embankment dam and its foundation was evaluated in terms of flow rates, pressure heads and hydraulic gradients.

Validation and calibration
For the case with no geomembrane foundation treatment, the obtained flow rates were slightly higher than those obtained by the physical model.For anisotropy of the dam equal to 1, the predicted percolation flow rate has been 3% higher than the value obtained for the small-scale model.For anisotropy of the dam equal to 5, this difference was 16% and for anisotropy equal to 10, the difference was 33%.
The numerical analyses with geomembrane treatment also obtained flow rates higher than those obtained in the physical model.In this case, for anisotropy of the dam equal to 1, the predicted percolation flow rate was 10% higher than the value obtained for the small-scale model.For anisotropy equal to 5, this difference was 33% and, for anisotropy equal to 10, 62%.
The predicted values of hydraulic head were similar to those measured in the small-scale models (Fig. 6), with slight variations relative to the anisotropy coefficients.
According to the numerical analyses with no geomembrane foundation treatment, the differences between the predicted pressure heads in relation to the small-scale tests (Fig. 6), within the dam foundation for the piezometers PF-12, PF-13, PF-14, PF-15, PF-16, PF-17 and PF-18 varied between zero and 1% for the studied anisotropy coefficients.For the dam foundation in the location of piezometers PF-1, PF-2, PF-3, PF-4 and PF-5, the predictions have shown a difference between -30% and -8%, once the chimney drain highly influences this area.However, this variation is considered small for engineering practice.On the other hand, for the piezometers PF-6, PF-7, PF-8, PF-9, PF-10 and PF-11, this difference has ranged between -11% and 1%.Finally, for the embankment dam in the location of piezometers PE-1, PE-2, PE-3 and PE-4, this difference has ranged between -11% and -1%.
The use of geomembrane can also reduce the pressure head in the dam foundation, according to the numerical analysis.For piezometers PF-9, PF-10 and PF-11, this reduction has ranged between 17% and 29%, which is similar to that observed in the small-scale model (Fig. 6).Additionally, the numerical analyses also predicted a reduction of the hydraulic head between 4% and 8% for the piezometers PF-6, PF-7 and PF-8, located in the dam foundation below the vertical filter.For piezometers PF-6, PF-7, PF-8, PF-9, PF-10 and PF-11 and considering anisotropy of the dam equal to 1, the predicted pressure heads have reduced 12%, 4%, 5%, 8%, 10% and 7%, respectively, when comparing to the small-scale model.When considering the anisotropy of the dam equal to 5, the differences have been 12%, 3%, 5%, 8%, 9% and 65%, respectively.Finally, the differences have been 12%, 3%, 5%, 7%, 8% and 6%, respectively, for anisotropy equal to 10.
The region within the embankment, corresponding to piezometers PE-1, PE-2, PE-3 and PE-4, presented a pressure head reduction between 20% and 45% due to the presence of the geomembrane.For these piezometers, the simulations have diverged -10%, -10%, -8% and -9% when comparing to the small-scale tests, respectively.However, when the anisotropy of the dam was considered equal to 5, the differences have been 13%, -8%, 2% and -8%, respectively.Following the results, the differences have been 20%, -8%, 7% and -7%, respectively, for anisotropy of the dam equal to 10.
Based on these results, it can be inferred that the anisotropy coefficient from the embankment influences the percolation flow rates and the pressure heads, for the case studied.The anisotropy coefficient that leads to results closest to the values obtained in small-scale modelling is equal to 1.However, it must be considered that compacted soils in actual dams present anisotropy relative to permeability.For this reason and in accordance with Salto Dam results, the anisotropy coefficient adopted for the embankment in the numerical model was k h /k v = 5.This is acceptable, once in the field some factors cannot be simulated, such as heterogeneity.

Parametric analysis and results
After validation and calibration, parametric analyses were performed to identify the influence of some factors on the internal flow through dam.In these analyses, some parameters were varied and others remained the same.In all analyses the following independent variables were kept constant: a) the dam geometry, b) the thickness of the foun-dation permeable soil and c) the hydraulic conductivity of embankment and foundation.The following evaluations have been made: 1) Use of a single geomembrane barrier as impermeable blanket, evaluating the geomembrane application in dams by the following simulations (Fig. 8 -A): (1a) No treatment for dam embankment and foundation; (1b) Treatment with geomembrane only for foundation, with anchor height equivalent to 15% of the dam height (geomembrane over surface A2-A3-A4); (1c) Treatment with geomembrane only for the dam upstream slope along all length (geomembrane over surface A1-A3); (1d) Treatment with geomembrane over the dam upstream slope and foundation (geomembrane over surface A1-A3-A4).
The height equivalent to 15% of the dam height is equivalent to that observed in Salto Hydroelectric dam.The geomembrane length on the foundation is also the same found in Salto Hydroelectric dam (L = 4H).
2) Use of a single barrier of compacted soil as impermeable blanket, with the same material and permeability coefficient of the dam (2 x 10 -4 cm/s), evaluating the foundation treatment through a single barrier of compacted soil with the following thicknesses (t) (Fig. 8 -B): (2a) t = 80 cm; (2b) t = 300 cm; Geomembrane as an Upstream Impermeable Blanket of Embankment Dams -Laboratory and Numerical Study (2c) t = 600 cm.The foundation treatment length was the same considered for the geomembrane (L = 4H), over surface B1-B2.
All the simulations considered the anchor height equivalent to 15% of the dam height (surface C1-C2).
The foundation treatment length was considered constant and equal to L = 4H, over surface D5-D6.
5) Longitudinal defects in the geomembrane, simulating a set of defects and evaluating the effect of an 80 cm thick compacted soil as a secondary barrier.Considering that the software used in the analysis supports only twodimensional simulations, these defects were simulated as a single longitudinal tear on the geomembrane, with unit width (1 m).The idea is to simulate a very adverse condition related to problems that can occur in the field such as during the placement of the cover soil over the geomembrane, since this constructive stage may result in geomembrane defects if not properly implemented.More reasonable parameters could be obtained with the use of three-dimensional modeling.
The results of parametric analyses are presented and discussed according to hydraulic head, flow rate and hydraulic gradient, as follows.

Pressure head
The pressure heads obtained by the analyses considering or not the presence of the geomembrane (Case 1) are shown in Fig. 9.When geomembrane was used, the heads were significantly reduced, indicating that the geomembrane caused the reduction of pressure head in the soil.Foundation treatment with geomembrane increased the dam safety, since a pressure head and water flow decrease were observed.
The pressure heads measured in further simulations are presented in relation to the piezometer locations (Fig. 9).This procedure was adopted to synthesize the evaluated data.Slight variations were detected for piezometers PF-1 to PF-8, once they are installed in the dam foundation below the downstream embankment and are not suitable for evaluating the geomembrane performance.
Figure 10 presents the pressure heads of the embankment piezometers and analyzes are presented as follows.
With the presence of the geomembrane (Case 1), based on the results (Fig. 9), it is possible to understand that pressure heads are heavily influenced by the upstream embankment and foundation treatment with geomembrane.If the design purpose is to reduce the pressure heads in the embankment to ensure dam stability, Case 1c has the best cost-benefit.However, the installation of geomembrane over the upstream embankment may not be viable when the embankment has low hydraulic conductivity and the design purpose of the geomembrane installation is to minimize the flow rates along the foundation soil.In this case, Case 1b has the best cost-benefit.Case 1d may be the best solution for cases when both embankment and foundation have high hydraulic conductivity.
The use of a compacted soil barrier (Case 2) resulted in small pressure changes in the embankment, with exception of piezometer PE-4, which is located near the anchor trench.
In relation to the study concerning the geomembrane length relative to the total height of the dam (Case 3), small changes in pressure head were detected for the embankment piezometers (Fig. 10), except for piezometer PE-4.For this reason, for the studied case, it is possible to conclude that the length of the treatment does not heavily influence the upstream embankment stability, as long as the pressure heads remain below projected levels.
According to the results of Case 4 (Fig. 10), the pressure heads within the embankment have reduced when the geomembrane height along the slope increased.In relation to Case 1b, pressure heads of piezometer PE-1 have reduced 3%, 15%, 38% and 100%, for Cases 4a, 4b, 4c and 4d, respectively.Again, when comparing to Case 1b, pressure heads of piezometer PE-2 have reduced 13%, 25%, 40% and 43%, for Cases 4a, 4b, 4c and 4d, respectively.Pressure heads of piezometer PE-3 have reduced 6%, 65%, 76% and 94%, for Cases 4a, 4b, 4c and 4d, respectively, compared to the case with geomembrane with 15% embankment height.The same kind of analyses was made for the pressure heads of piezometer PE-4 and showed a reduction of 11%, 31%, 33% and 39%, for Cases 4a, 4b, 4c and 4d, respectively.However, considering the embankment soil has satisfactory permeability, the use of geomembrane along the embankment should just be long enough to construct the anchor trench, such as in Case 1b.Greater lengths may be adopted when the embankment stability may get reduced by the higher pressure heads.
According to Case 5 (Fig. 10), small oscillations of pressure head were detected for embankment piezometers PE-1, PE-2 and PE-3, with the presence of longitudinal defects in the geomembrane.Contrasting with Case 1b, PE-4 presented an increase of pressure head equal to 22% and 36% for Cases 5a and 5c.However, these values were just 0% and 11% for Cases 5b and 5d, respectively, which considered the existence of an 80 cm thick compacted soil as a secondary barrier.In this case, the compacted soil had an important role as a secondary barrier.
Figure 11 presents the pressure heads for foundation piezometers located below the embankment dam and analyses are presented as follows.
For Case 1, piezometers located below the upstream embankment of the dam are highly influenced by the geomembrane treatment (Fig. 11).Considering piezometer PF-9, the pressure heads have reduced in relation to Case 1a (without geomembrane) 26%, 4% and 41%, for Cases 1b, 1c and 1d, respectively.For piezometer PF-10, on the other hand, the pressure heads have reduced compared to Case 1a 29%, 4% and 44%, for Cases 1b, 1c and 1d, respectively.Piezometer PF-11 presented reduction of pressure heads, in relation to Case 1a, equal to 36%, 5% and 48%, for Cases 1b, 1c and 1d, respectively.For this reason, Cases 1b and 1d have better cost-benefit when the geomembrane purpose is to reduce pressure head in the foundation.
Considering the use of a compacted soil layer (Case 2), variations in pressure heads were detected for piezometers located below the upstream embankment of the dam (Fig. 11), however the effect of the geomembrane in pressure head reduction is lower than with the use of geomembrane (Case 1b).
The pressure head for foundation piezometers located below the upstream embankment of the dam (Fig. 11) presented small changes for piezometer PF-9 and more considerable changes for PF-10 and PF-11, which are located nearest to the treated area.In relation to Case 3a, which corresponds to a geomembrane length equal to the total height of the dam, the maximum reduction of pressure head (Case 3c) was 6%, 14% and 17% for piezometers PF-9, PF-10 and PF-11, respectively.
According to the results of Case 4, the pressure head within the foundation soil upstream of the dam had small variations in relation to the geomembrane length installed over the upstream slope of the dam.
For Case 5, an increase of pressure head was observed for foundation piezometers located below the upstream embankment of the dam (Fig. 11), when comparing to Case 1b.The increase of pressure head is a consequence of the flow rate increment on foundation soil below the dam.However, Cases 5b and 5d presented lower increase of pressure head, once an 80 cm thick compacted soil was considered as secondary barrier, with the same hydraulic conductivity of the dam.
Figure 12 presents the pressure head for foundation piezometers located upstream of the dam, and analyses are presented as follows.
Regarding the use of compacted soil barrier (Case 2), the reduction of pressure in the foundation is lower in magnitude than that observed for geomembrane application, indicating that the use of a single compacted soil barrier is less effective than the use of geomembrane for the evaluated case, even with high values of thickness (6 m).It must be considered that the studied soil liner has the same permeability of the embankment dam (2 x 10 -4 cm/s) and better results might be achieved with the use of soils with lower permeability.However, even when geomembranes are applied, a protective layer is normally recommended for geomembrane protection.In the specific case of this research, a compacted soil barrier was used with the purpose of acting as a watertight defense if the geomembrane is subjected to damage, also cooperating in the geomembrane protection.It is important to highlight that thick layers of polypropylene geotextile might be used with the purpose of protecting the geomembrane rather than compacted soil layers, since the installation of the synthetic layers may lead to lower risks of damaging the geomembrane.Besides, some researchers (e.g.Touze-Foltz, 2009) have shown that when the compacted soil over the geomembrane becomes saturated and is subjected to an applied load, the flow rate through the geomembrane defects may increase.Additionally, for Case 2, piezometers PF-12 and PF-13 exhibited considerable reduction of total head with the use of a compacted soil barrier.Contrasting with Case 1a, piezometer PF-12 had decrease of total head equal to 11%, 23% and 28% for Cases 2a, 2b and 2c, respectively.Also, when contrasting with Case 1a, piezometer PF-13 had decrease of total head equal to 7%, 19% and 19% for Cases 2a, 2b and 2c, respectively.However, the increase in thickness from 0.8 m to 3 m or 6 m may not be economically viable when considering the amount of material necessary and the other possibilities of foundation treatment, such as geomembrane.
For Case 3, pressure head reduction was detected in the foundation soil treated with geomembrane, upstream of the dam (Fig. 12).The magnitude of the reduction is directly proportional to the treated foundation extension.For this reason, more significant pressure head reduction was recorded for Case 3c, which corresponds to a geomembrane length of 3 times the total height of the dam.
In practical terms, the increase of the treated extension (Case 3) results in higher costs and the engineers should select a suitable treatment depending on the budget of the project.As an example, Salto Hydroelectric Dam has treated foundation extension variable according to the evaluated cross-section.
According to the results of Case 4, the pressure head below the upstream embankment of the dam had small variations in relation to the geomembrane length installed over the upstream slope of the dam.This was expected once this analysis considered the length of geomembrane over the upstream slope of the dam.
The occurrence of defects in the geomembrane (Case 5) increased the pressure head in the foundation soil upstream of the dam (Fig. 12), especially for piezometers located near the damage.For example, for piezometer PF-12 located near the dam, the pressure head from this piezometer (Case 1b), contrasted with Cases 5a and 5c, resulted in an increase of 23% and 37%, respectively.For piezometer PF-18, on the other hand, these values were just 4% for both situations.
Only a small increase of this pressure was recorded when an 80 cm thick compacted soil layer was placed above the geomembrane with defects (Fig. 12).As an example, the pressure head from PF-12 (Case 1b), contrasted with Cases 5b and 5d, resulted in an increase of 9% and 16%, respectively.Therefore, based on the results, the use of geomembrane associated to a compacted clay layer above it is an interesting solution to reduce the risk of the loss of efficiency of the system.

Seepage flow rate
For the hypothetical case when geomembrane was not used (Case 1a), the obtained flow rates were 255.1 L/(h.m) and 1371.1 L/(h.m) through the embankment and the foundation, respectively.Figure 13 presents the flow values from all the simulations and the reductions in total flow compared to Case 1a.Flow reduction in the embankment was detected even when geomembrane was applied over the upstream slope of the dam with just the necessary length for anchoring it, making them work together.Since the permeability of the foundation is 50 times higher than the permeability of the embankment, most of the percolation occurred through the soil foundation, justifying the use of geomembrane only on permeable foundation soil (Case 1) for the studied case.
Predicted values considering only the use of compacted soil as a barrier (Case 2) showed that the flow rates were greater than the ones with the use of geomembrane.For 0.8 m thick soil layer, the total flow rate has increased 43%, whereas for 3 and 6 m thick layers, the increase was equal to 26% and 14%, respectively, when compared to Case 1b.Therefore, even considering that the compacted soil layer has no cracks, its efficiency is lower when compared to the use of geomembrane without defects for the studied case.For this reason, the construction of thick layers of compacted soil might not be economically feasible when compared to solutions with geomembranes and each case must be evaluated by the engineers.
Based on the simulations that considered the variation of the geomembrane length (Case 3), it was found that the seepage flow through the foundation depends on the length of the water percolation path.Although there is a reduction in flow through the embankment, its magnitude is small when compared to the flow through the foundation.
According to the obtained results from numerical analyses that considered the length of geomembrane over the upstream slope of the dam (Case 4), when the length increases the flow rate through the embankment decreases.However, the decrease in total flow is small, because most of the flow occurred through the foundation and the geomembrane is covering most of it.For this reason, it is sufficient to adopt just the necessary length for anchoring the geomembrane on the slope.
The obtained seepage flow rates considering the existence of one longitudinal tear showed that if there is not a compacted soil layer over the geomembrane (Case 5a), the total flow increases approximately 25% when compared with the geomembrane without defect (Case 1b).However, if an 80 cm thick protective soil layer is placed over the geomembrane (Case 5b), the flow rate increases only 6% in comparison with the case of geomembrane without defect (Case 1b).Predicted values of total flow, when three defects were considered in the geomembrane, indicated that the flow rate increase was approximately 32% (Case 5c) higher than that obtained for the geomembrane without defect (Case 1b), however this value was only 14% considering the presence of a protection layer (Case 5d).Thus, the presence of a layer of lower permeability than the foundation above the geomembrane is of great relevance for the proper performance of the system.

Hydraulic gradient
According to the results presented for Case 1, the hydraulic gradients in most of the embankment were between 0.8 and 1 for the analyses without the presence of geomembrane.Considering the use of geomembrane on foundation soil, the hydraulic gradient in the embankment, specifically near the anchor trench area, reached values above 1.5, as shown in Fig. 14 (A).For this reason, the anchorage area must be carefully constructed to avoid unexpected percolation between the geomembrane and the compacted soil.
In Case 2, the hydraulic gradients were found between 0.5 and 1 for most of the embankment for the case when an 80 cm thick compacted soil liner was employed.In the simulations considering soil liner 3 and 6 m thick, gradients in the embankment were between 0.8 and 1.2.High hydraulic gradients were observed near the connection with the embankment dam, with values ranging between 1.4 and 1.8, as shown in Fig. 14 (B), which suggests that the anchorage region between the embankment dam and the compacted soil liner must have high execution process control.
The hydraulic gradients observed in Case 3 ranged between 0.8 and 1.0 in most of the embankment, when a geomembrane with length L = H = 24 m was used.For higher lengths, the hydraulic gradients were greater, with values between 1 and 1.5 in the embankment, with greater values near the anchor trench area.In all the studied cases,  hydraulic gradients lower than 0.5 were observed in the foundation.
Regarding Case 4, it was observed that the installation of the geomembrane over the upstream slope of the dam reduced the hydraulic gradient in the embankment.However, the flow through the embankment corresponds to a small portion of the total flow collected by the drainage system.High hydraulic gradients in anchor trench areas, with values greater than 1.5, were observed in all analyses, with small differences due to different lengths.
The presence of defects in the geomembrane (evaluated in Case 5) led to the increase of hydraulic gradient in the soil foundation when compared to the case of the geomembrane with no defects.However, the values were below 0.5.Particularly in the damaged areas, high hydraulic gradients were found, which can contribute to the formation of percolation paths below the geomembrane.This is confirmed by the high values of pressure head of some piezometers below the defects, shown in the last section (Fig. 6).

Conclusions
This paper presented an experimental and numerical study of the performance of HDPE geomembrane as an impermeable blanket installed upstream of embankment dams.Data from the Brazilian Salto Hydroelectric Power Plant contributed to the development of a small-scale seepage model and the numerical modelling.After the laboratory tests, some parameters were calibrated and some numerical analyses of those tests have been performed.Additionally, some hypothetical conditions varying the presence of defects in the geomembrane, the impermeable blanket thickness and the length of the geomembrane were carried out.
For the small scale model, the obtained results indicated that the geomembrane impermeable blanket increased the percolation path through the dam soil foundation, reducing the pressure head in both embankment and foundation, as well as the flow collected by the drainage system.The higher reduction in pressure head occurred in the dam foundation, below the geomembrane and in the dam embankment upstream of the vertical filter.At the downstream side of the dam, on the other hand, a small variation of pressure head was observed.It indicates that the flow in the downstream side is controlled by the chim-ney drain.The tests also have shown that if the applied geomembrane has damage, the total flow collected by the drainage system and the pressure heads are similar to those observed in the test with no geomembrane.
Based on the results of the scaled and the numerical model, it can be inferred that the anisotropy coefficient from the embankment influences the percolation flow rates for the case studied.The anisotropy coefficient that leads to results closest to the values obtained in small-scale modelling is equal to 1.However, it must be considered that compacted soils in actual dams present anisotropy relative to permeability.For this reason and in accordance with Salto Dam results, the anisotropy coefficient adopted for the embankment in the numerical model was k h /k v = 5.
Regarding the use of impermeable blanket of compacted soil as a single barrier with a horizontal permeability of 1 x 10 -3 cm/s and vertical permeability of 2 x 10 -4 cm/s (same material of dam), even considering it without cracks, the seepage reduction was not as significant as that found when geomembranes were used.Even for a 6 m thick layer, the performance of compacted soil was worse than that obtained by using geomembrane.The flow rates were greater than those calculated by software simulation with the use of geomembrane.Thus, even considering that the compacted soil layer has no cracks, its efficiency in some cases is lower when compared to the use of geomembrane without damage for the case studied.
For the analyses that considered the length of the geomembrane, pressure head reduction was detected in the foundation soil treated, upstream of the dam.Thus, the magnitude of the reduction is dependent on the treated foundation extension.It was also found that the seepage flow through the foundation depends on the length of the water percolation path.
The analyses considering the length of the geomembrane over the upstream slope of the dam have shown that the pressure head in the foundation soil had small variation in relation to the geomembrane length installed over the upstream slope of the dam, once most of the flow occurred through the foundation.However, the pressure head within the embankment have reduced when the geomembrane length along the upstream slope increased.Considering that most of the total flow occurred through the foundation, increasing the length of the geomembrane over the upstream slope of the dam is not effective for reducing the magnitude of this parameter.For this reason, it was sufficient to adopt just the necessary length for anchoring the geomembrane on the slope.
For the analyses with the occurrence of defects in the geomembrane, a compacted soil layer over the geomembrane resulted in good performance for the foundation treatment.In the simulation that considered longitudinal tears in the geomembrane without the compacted soil layer, the flow increased approximately 25%, when compared to the simulation with undamaged geomembrane.However, considering a soil with lower permeability than the foundation protection layer in addition to the geomembrane liner, the flow reduction was only 6%.
Relating to the hydraulic gradients, the use of the geomembrane on foundation soil resulted in gradient values higher than 1.5.For this reason, it is recommended to carefully construct the anchor trench to avoid unexpected percolation between the geomembrane and the compacted soil.Major care must also be taken in case of defects in the geomembrane.
Based on all the results of this analysis, the use of geomembrane for treating permeable foundations can provide a good solution to reduce flow and pressure loads on dams.

Figure 1 -
Figure 1 -Salto Hydroelectric Power Plant: (A) Aerial view and (B) Geomembrane anchor trench at upstream dam face and foundation.

Figure 2 -
Figure 2 -Piezometers location in the small-scale model: (A) Model cross section setup and (B) Photo from laboratory tests.

Figure 5 -
Figure 5 -Dimensions of the small-scale model.

Figure 6 -
Figure6-Total hydraulic head obtained in tests with small-scale models.

Figure 7 -
Figure 7 -Numerical model: (A) Seepage flow rate collected by the drainage system and (B) Boundary conditions of the numerical model.

Figure 9 -
Figure 9 -Pressure head for Case 1 (in meters of water column).

Figure 11 -
Figure 11 -Pressure heads for foundation piezometers located below the upstream embankment of the dam.

Figure 12 -
Figure 12 -Pressure heads for foundation piezometers located upstream of the dam.

Figure 13 -
Figure 13 -Flow rate reduction in relation to all the studies.

Table 1 -
Main characteristics of the left abutment of Salto Hydroelectric Dam.

Table 3 -
Results of laboratory tests on material used on smallscale models.