Integration of stochastic methods in the assessment of the risk generating phenomena. Study case - Tărnicioara tailings pond, Romania

Integration of stochastic methods in the assessment of the risk generating phenomena. Study case - Tărnicioara tailings pond, Romania 1,2 Diana Banu, 2 Sorin Mihai 1 Faculty of Geology and Geophysics, University of Bucharest, Bucharest, Romania; 2 Research and Development National Institute for Metals and Radioactive Resources, Bucharest, Romania. Corresponding author: D. Banu, banudianamaria@yahoo.com Abstract. This paper presents an acquisition methodology of the geological, geotechnical and hydrogeological data through direct and indirect methods, such as drillings and electrical tomography. Data from geophysical investigations and laboratory tests together with the data recorded by the existing monitoring system were used in numerical and analytical models to assess the stability of the tailings dam Tărnicioara during the closure. This paper focuses on the integration of the Monte Carlo method coupled with limit equilibrium method and on the point estimation method coupled with finite element method. The modeling methodology applied in the study focused on 3 cases: assessing the tailings dam safety state by applying analytic modeling, setting the warning and alarm benchmark in respect to water level oscillation recorded by the monitoring network and modeling the risk of tailings liquefaction due to earthquake loading. Key Words: tailing dams, internal erosion, slope stability, liquefaction, monitoring. Introduction. The awareness rising about the risks associated to the tailing dams permanent existence and with the communities requirements, the increasing rigors about the protection limits of the environment factors and the alignment requirement to the European regulations, has outlined a conceptual framework in which all parties must ensure the safety of these mining waste deposits. To ensure the criteria of long-term environment and the communities security, methodologies have been developed, suitable for achieving, processing, monitoring and modeling the behavior and the development of the risk generating phenomena (Cheng & Lau 2008; Knödel et al 2007). Worldwide, the deterministic approaches like the limit equilibrium method (LEM) and finite element analysis (FEM) are used conventionally to assess the stability of tailings dams in the mining industry (Mihai et al 2008). In the last years, the stochastic methods were introduced complementary to the analytical and numerical modeling techniques to assess stability of tailings dams (Valley & Duff 2011; Hammah & Yacoub 2009; Hamade et al 2011; Wolff 1996). The classical/deterministic approach implies a single set of input parameters values to generate one solution (unique values for the output parameters), while the stochastic approach takes into account the variability of the input parameters namely (Hammah & Yacoub 2009): - uncertainty (generated by the inability to determine accurately the main geotechnical parameters); - variability (random distribution of physical and mechanical data); - anisotropy (the modification of the rocks behavior and the sliding mechanism in certain directions). Tailings dams are subjected to many hazards that directly influence their stability. These hazards need to be properly identified and assessed. Each identified hazard was located in a risk matrix provided on Figure 1 (Vanden Berghe et al 2011). As can be seen in the Figure 1, the most important hazards concerns the tailings dams safety state on short and long term are: the high water level inside de dam s body, its heterogeneity and anisotropy and also an inadequate monitoring network or inaccurate modelling. A little researched phenomena such as the static liquefaction is considered a most common cause of flow failures of tailings dams, for example: Stava (ICOLD 2001), Sullivan Mine (WISE 2014; Davies et al 2002), Merriespruit (Fourie et al 2001), and Aznalcóllar (Davies et al 2002). 45

Figure 1. The risk matrix of dams stability (Vanden Berghe et al 2011). Probabilistic analysis helps to quantify risk and promotes greater understanding of problems. It increases the chances of success through more robust design of excavations, stabilization measures and improved monitoring decisions (Hammah & Yacoub 2009). The advantages of using stochastic methods in modeling the risk generating phenomena are: - the reduction of uncertainty by eliminating or decreasing the parameters measurement errors; - the variability evaluation of input/output parameters; - the anticipation of the most likely sliding mechanism; - highlighting the model sensitivity by direct correlations between parameters characteristics. Research in which different kind of probabilistic methods were applied to the modeling of tailings dams instability phenomena were conducted abroad by Benoît Valley and Damien Duff from Center for Excellence in Mining Innovation (Wolff 1996). The methods applied in Romania to assess the tailings ponds safety status are mainly classic type, such as the limit equilibrium methods with version as Fellenius, Simplified Bishop or Janbu and even grapho-analytical methods like Taylor circle s. The necessary support for these methods were made of the data achieved from geotechnical drills and piezometer network, visual observations about erosion or subsidence phenomena on external slope face (Mihai et al 2008). Lately the technical consultants have started to use modern methods especially for monitoring the tailings ponds on the closure and post-closure periods. Under the Phare 2006 national program it was possible the development of a pilot monitoring stations for three main regions with impact to the former mining activity respectively Vatra Dornei, Baia Mare and Deva. Tărnicioara tailings management facility has an operational system which started monitoring data from September 2010. These data consist of hydrostatic levels, inclinometer for slope on X/Y directions, water quality indicators (ph, Eh, heavy metals, etc.). All these data are stored in a data logger station which transmits information wireless to a central acquisition station located in Iacobeni (INCDMRR-ICPMRR 2011-2012-2014). Another pilot project developed was the Phare 2006/018 Applying the InSAR (Interferometric Synthetic Aperture Radar) techniques for monitoring the tailings dams surface and the nearby area. The combination of two or more InSAR images of the same area may obtain accurate topographic or motion information about the target area. This information is contained in the phase differences between the images, so the coherence of the images must be preserved in order to make an interferogram. This air born system is still tested for conditions noted above. In Figure 2 is presented the deformation map which highlights negative deformations (subsidence) with values between -2 to 6 cm, due to the water level decrease in the middle summer period. 46

These monitoring networks, investigation and modeling techniques will be developed for 6 tailings ponds in Vatra Dornei perimeter and further the main objective will be the integration into The National Integrated Monitoring System for Environmental Impacts. Figure 2. Short-term InSAR deformation map (Poncos et al 2012). Stochastic methods used. This research used two stochastic methods to be integrated into analytical limit equilibrium technique and into numerical finite element with shear strength reduction technique. These methods are the Monte Carlo (MC) and Point Estimate Method (PEM). The ultimate goal of a probabilistic slope stability analysis is to obtain the complete distribution of factor of safety values given a set of random (uncertain) input variables with specified statistical properties. From the distribution of factor of safety values, probability of failure can be determined. In this case, factor of safety is known as a response variable, and the algorithm used to calculate factor of safety as a response function (Hammah & Yacoub 2009). In the MC method, samples of probabilistic input variables are generated and their random combinations used to perform a number of deterministic computations. Information on the distribution and moments of the response variable is then obtained from the resulting simulations (Hammah & Yacoub 2009). If in a MC analysis we assume factors of safety to be distributed according to a normal distribution, the probability of failure and reliability index may be calculated from mean, μ, and standard deviation, σ (Wolff 1996). The PEM was originally developed by Rosenblueth (Rosenblueth 1975, 1981). As its name suggests, the PEM uses a series of point estimates point-by-point evaluations of the response function at selected values (known as weighting points) of the input random variables to compute the moments of the response variable. The principle of PEM is to compute solutions at various estimation points and to combine them with proper weighting in order to get an approximation of the distribution of the solution (Figure 3). The PEM implemented in Phase2 8.0 is the two-point estimate method for the first and second moment of uncorrelated variables. It needs 2n evaluations of the solution, where n is the number of random variables. 47

ECOTERRA - Journal of Environmental Research and Protection Figure 3. The computation principle of the point estimate method for two input variables (Valley & Duff 2011). Study case. The Tărnicioara tailings pond, Suceava County, Romania, is a high capacity valley pond and it is placed at the confluence of Scăldători and Tărnicioara creeks belong to Siret hydrographic basin. According to the Romanian National Register of Dams Rebar chapter for mining industry tailings, Tărnicioara tailings pond is classified as class III of importance by the regulatory standard STAS 4273/83 and C category of importance by NTLH 021 Technical Regulations (INCDMRR-ICPMRR 2011-2012-2014). This tailings pond was operated from 1975 to 2006, when it started the closure phase. The closure and the rehabilitation works weren t completed until early 2013 because of suffusion phenomena recurrences (INCDMRR-ICPMRR 2011-2012-2014). Tărnicioara tailings management facility has served as the main storage for the slurries resulted from the ores processing of Cu, Pb, Zn and barite in Tarniţa Processing Plant located about 6 km from the Ostra village, on the road linking Frasin and Leşu Ursului villages. The Tărnicioara stream water was diverted in the Brăteasa creek through a gallery and the Scăldători stream water undercrossed the pond through a closed concrete channel (Figure 4). Figure 4. Plan view of the Tărnicioara tailings pond (Source: Google Earth). 48

When the tailings pond exploitation was closed, the maximum height considered from the starter dike toe to the dam s crest was 78.50 m. The total amount of stored tailings was 15,668,000 tons equivalent to about 8,330,000 m 3. Tărnicioara tailings pond total footprint, including diverting dams and the main dam, is about 38 ha (INCDMRR-ICPMRR 2011-2012-2014). During the exploitation, the technological and pluvial water drainage was achieved through 5 decant wells. Through the 38 years of the tailings pond's existence, there were recorded many incidents and accidents, some of them even after its passing into reserve: the emergence of suffusion phenomena in the zone of decants wells (Figure 5) and also on the main downstream slope, erosional phenomena on downstream slope, the clogging of the Scăldători stream gallery under the tailings dam body (INCDMRR-ICPMRR 2011-2012-2014). Figure 5. Surface collapse due to suffusion phenomena around a former decant well 22/11/2013. Field and laboratory investigations. Due to the depositional process by sedimentation, tailings material may exhibit an anisotropic behavior with a lower shear resistance along horizontal planes. In this case, direct and simple shear tests are preferred to triaxial tests (Ballard et al 2008). In order to obtain the necessary data for modeling of the safety state, the following investigations have been executed during the years of 2011-2012: - two geotechnical drillings: FG1 (20 m depth, located on the dam s top berm) and FG2 (60 m depth, located on the pond beach); 25 samples were taken from 4 to 4 m from the two drills, required for laboratory determination of the geomechanical parameters required for stability calculations. The water level after stabilization in the two drills was 28 meters deep, measured from collar; - seven transverse resistivity profiles: the first one was located on the pond beach, the second on the dam s crest and the other five profiles on berms and on the starter dam. The resistivity technique used was ERI (Electrical Resistivity Imaging) and was performed by Wenner-Schlumberger method, with electrodes located at 5 m apart. The device used for this survey was a Pasi multielectrode station (Figure 6). These profiles were used to indicate the presence of water leaks from undercrossing gallery and from the unsealed decant wells, and also the emergence of the internal erosion phenomena which are difficult to detect by conventional methods. Geoelectrical profiles were calibrated with geotechnical drills soil type in order to specify the distribution of parameters along the axial profile used for modeling the safety state of tailings dam. 49

Figure 6. The resistivity profile located on the main starter dam s berm. The hydrostatic level data was achieved by time series using the existing piezometric network with 5 operating units, over a period of 8 months (Figure 7). Figure 7. Time series for water levels monitoring in piezometer P2 for an 8 months period. Grain size diagrams were used to identify the potentially liquefiable tailings using the guidelines for the study of non cohesive soil liquefiable properties (Indicative P 125-84, 1984). Angle of internal friction and cohesion were determined by the direct shear test performed with Shearmatic device made by Controls. 50

The stability modeling by classical approach. The stability calculations were performed on a vertical geotechnical cross section along the downstream slopes axis, with a length of approx. 550 m. For this section two scenarios were considered: the static and the pseudo-static seismic, in which the seismic requests interfere at site area magnitude (Ks = 0.16). Deterministic methods have been used for the situation in which the dam s stability was computed taking into account the elevation of hydrostatic level in FG1 and FG2 wells, measured in August and September 2010 (the hydrostatic level was low due to drought and drainage works completed and functional). The geotechnical parameters used in stability calculations (volumetric weight, cohesion, friction angle) were determined by the Geotechnical Laboratory of the Faculty of Geology and Geophysics, Bucharest. Bidimensional calculation section was analyzed using Rocscience Slide software by five different analytical methods that satisfy the static equilibrium of forces or moments respectively Bishop, Lowe-Karafiath, Janbu), or the static equilibrium of forces and simultaneous moments as Spencer and Morgenstern-Price (Cheng & Lau 2008). The analysis was performed on about 5000 slip surfaces, each divided into 50 vertical strips; the convergence was 0.001, and the maximum number of iterations was set for 100. By comparison, the finite element numerical method (SSR - Shear Strength Reduction) was applied, using Rocscience Phase2 software. The differences between the safety factors values determined by analytical and numerical methods are explained by the approximation of the dilatation angle to 0 (the dilatation angle quantify volume changes that occur during creep), which leads to a reduction of the safety factor value (Table 1). For this reason it is recommended to use the analytical and numerical methods simultaneously. The stability calculation results Safety factor Table 1 Modeling methods Piezometric boundary (NH) NH measured in geotechnical drills Average values of the safety factor Lowest values of the safety factor Highest values of the safety factor Numerical modeling method NH measured in geotechnical drills Bishop Janbu Spencer Lowe- Karafiath Morgenstern- Price F static F dinamic F static F dinamic F static F dinamic F static F dinamic F static F dinamic 2.003 1.142 2.029 1.156 1.993 1.136 1.996 1.126 1.997 1.136 1.691 0.948 1.729 0.957 1.695 0.955 1.692 0,933 1.694 0.952 1.569 0.883 1.508 0.901 1.573 0.904 1.571 0.879 1.572 0.888 1.809 1.012 1.840 1.011 1.813 1.004 1.809 0.906 1.812 1.016 Finite Element Method SSR (Shear Strenght Reduction) F static F dinamic 1,87 1,02 The resulting values of the stability calculations are between 2.003 and 1.691 for the static hypothesis and between 1.142 (Figure 8) and 0.948 for the seismic hypothesis. For the September 2011 situation, the Bishop Method stability calculations show an aprons value for the static hypothesis (Fs = 2.003) and a corresponding value for the pseudostatic hypothesis (Fs = 1.142). 51

Figure 8. Pseudo-static stability analysis for Tarnicioara tailings dam. Safety factor Fs = 1.142 (Bishop Method, seismic coefficient Ks = 0.16). The stability modeling by Monte Carlo method. This modeling has been carried out in order to verify the feasibility of the monitoring system in operation. Hypothesis taken into account was the pseudo-static one to consider the worst case with a seismic load generated by a peak acceleration of 0.16 g in that area. In this case, the random variable will be the hydrostatic level with maximum and minimum values from the time series of each piezometer. Performing a sensitivity analysis, the interval between the minimum and maximum hydrostatic level is sampled into 50 values, and the safety factor was calculated by the limit equilibrium method. This procedure is repeated 5000 times (using MC sampling method) and is determining the probability distribution function of safety factors depending on the hydrostatic elevation level. From the safety factors statistical evaluation relative to the hydrostatic oscillation level, it can be observed that the static stability is adequate, but the pseudo-static safety factor has a subunit value (impending landslide in case of major seismic event according to regulations in force). The balance is satisfied for both cases where the hydrostatic level is less than 0.510 m (measured from the base of the borehole). A new MC simulation is performed assuming pseudo-static (for hydrostatic level ranged from a difference of 15 meters between the minimum and maximum values) on 5000 samples using the methodology described above for the existing network of piezometers. From the statistical variation diagram of the safety factor to the normalized value of hydrostatic level, reference values were imposed for the warning level (1 < Fs < 1.1) and for the alarm level (Fs 1.0) (Figure 9). The reference values were calculated for each piezometer and finally revealed that they have to be extended to a depth which allows the hydrostatic interception level (Table 2). The geotechnical properties of the tailings material that were used at analytical modeling are shown on Table 3. Reference values were calculated for each piezometer Table 2 Piezometer The actual tube The necessary tube NH value for the NH value for the length (m) length (m) warning level (m) alert level (m) P1 15 22 1.90 8.22 P2 15 24 1.87 8.91 P3 15 21 1.93 7.33 P4 10 18 2.32 7.73 P5 7 15 2.11 7.20 52

Tailings material geotechnical properties used for analytical modeling Table 3 Material name Color Unit weight Sat. Unit weight Cohesion Phi (kn/m 3 ) (kn/m 3 ) (kpa) (deg) Starter dike 20.00 21.5 0.0 30 Upstream berm 21.19 22.54 6.5 28 TailingFineSand 23.87 26.13 1.7 30 TailingSiltySand1 23.96 25.42 5.6 29 TailingSiltySand_NP2 20.87 21.35 14.8 24 Base Soil marl clay 19.50-18.0 22 Figure 9. Alarm levels for piezometers monitoring network; green line=warning level, red line=alarm level (assuming pseudo-static hypothesis) The stability modeling using PEM (risk assessement at liquefaction). Liquefaction is a loss of bearing capacity phenomenon of saturated sandy granular materials under the action of cyclic dynamic loads due to increased water pore pressure. Due to uncertainty of determination the degree of shear parameters reduction under the influence of seismic loading, the PEM method was used and applied to the evaluation of stability by finite element method (run with numerical analysis software Rocscience Phase 2). The advantage of a finite element approach in the analysis of slope stability problems over traditional limit equilibrium methods is that no assumption needs to be made in advance about the shape or location of the failure surface, slice side forces and their directions. In the SSR (Shear Strength Reduction) finite element technique, elasto-plastic strength is assumed for slope materials and material shear strengths are progressively reduced until collapse occurs. In this study was considered the Mohr-Coulomb material model or failure criteria (Mihai et al 2008). For Mohr-Coulomb material model, six material properties are required. These properties are the friction angle φ, cohesion C, dilation angle ψ, Young s modulus E, Poisson s ratio ν and unit weight of soil γ (Figure 10). Young s modulus and Poisson s ratio have a profound influence on the computed deformations prior to slope failure, but they have little influence on the predicted factor of safety in slope stability analysis. The change in the volume during the failure is not considered in this study and therefore the dilation angle is taken as 0. 53

The modeling was done in two stages: - the pseudo-static evaluation of the tailings dam stability considering the seismic acceleration at its maximum value (Ks = 0.16 g) and the pore pressure increasing in the liquefiable layer Tailing_SiltySand1 (Figure 11). We may also observe the modification of the landslide kinematics phenomena; - the post-seismic evaluation in static hypothesis and shear resistance reduction (cohesion and friction angle) to 25% of baseline values with a deviation of ± 5% on a normal distribution (Figures 12, 13 and 14). The critical surface in post-seismic evaluation is now located above the nonliquefied strata Tailing_SiltySand2. Figure 10. Numerical model of the finite element meshing and material distribution. Figure 11. Pseudo-static analysis using finite element method - main dam. Total displacement distribution, Safety factor Fs = 1.06 (seismic coefficient Ks = 0.16). 54

Figure 12. Post-seismic evaluation using the finite element method - main dam. Shear deformation distribution, Safety factor Fs = 0.88. Figure 13. Post-seismic evaluation using the finite element method - main dam. Total displacement distribution, Safety factor Fs = 0.88. Finite element analysis coupled with the stochastic PEM method gives us the possibility to predict displacements due to liquefaction triggered by a seismic event. In our case, liquefaction occurs in tailings with material code Tailing_SiltySand2, when the water pressure becomes equal with the total stress component and the effective stress will be zero. In this case, the silty sand will be in a liquid state with very low shear strength. The extent of the liquefied zone is about 53 m behind the dam and the maximum deformation occurred near the dam s crest with a value of about -57 cm (Figure 14). Due to the presence of low effective stress, the liquefaction initiates near the slope surface while the increase of effective stress with depth reduce liquefaction. An integrated monitoring system may be designed with the help of the existing piezometric network used for achieving data on the excess pore pressure and the InSAR regular deformations measurements focused on fixed landmarks for calibration with the numeric model. 55

Figure 14. Post-seismic evaluation using the finite element method - main dam. Yielded elements distribution, Safety factor Fs = 0.88. Conclusions. This study integrates for the first time the stochastic methods with analytical and numerical modeling of risk triggering phenomena on tailings dams. These methods are of great interest as modeling and prognosis tools in an integrated monitoring system for the geohazard phenomena that may be develop on active or closed tailings facilities. Based on complex data obtained in the field, laboratory and from the existing monitoring systems three situations were analyzed, respectively: the tailings dam safety state by applying the analytic modeling, the increase of pore pressure inside the dam and the risk of tailings liquefaction due to earthquake loading. In the first case, by analyzing the stability of the Tărnicioara tailings deposit, it appears that for the static analysis, the stability varies between 2.003 and 1.691 (being in accordance with current regulations) and for the dynamic analysis it varies between 1.142 and 0.948, indicating that at additional loading represented by an earthquake with a representative area magnitude, the landfill is below the stability limit. In the second case was analyzed the necessity to impose warning and alarm benchmarks using the existing piezometers network and their data on a period of 8 months. This modeling was done by applying Monte Carlo procedure to limit equilibrium method. Analyzing the estimation of these levels makes it clear that the current piezometers network can not reveal the warning level for the seismic hypothesis (worst case), because the piezometers have improper depths which do not allow the interception of the hydrostatic level throughout all the statistical achievable values. In the third case the post-seismic evaluation results modeled by point estimation method applied to the finite element shear strength reduction method, revealed a high degree of risk due to the liquefaction of silty sand deposit coded Tailing_SiltySand2, partially located below the piezometric level. This modeling gave us the possibility to predict the liquefied zone inside de dam s body and the distribution of shear deformation and displacement. As a conclusion, these stochastic methods allow their integration into a complex system for tailings ponds monitoring and their interfacing with advanced airborne technologies like InSAR. The national research such as the interferometric SAR (InSAR) applied in Phare 2006/018 project, developed on this study case, provides very good data for detecting vertical ground displacement on a centimeter scale. The probabilistic (stochastic) analysis allows the risks quantification and promotes a better understanding of the problems of structural stability for mining and processing waste deposits, leading to proper monitoring decisions and stabilization measures which 56

are more reliable and hold a much lower risk degree versus the strict application of deterministic methods. Acknowledgements. This work has been supported from the strategic grant POSDRU/159/1.5/S/133391, Project Doctoral and Post-doctoral programs of excellence for highly qualified human resources training for research in the field of Life sciences, Environment and Earth Science co-financed by the European Social Fund within the Sectorial Operational Program Human Resources Development 2007 2013. References Ballard J. C., Vanden Berghe J. F., Jewell R., Pirson M., 2008 Some lessons from a recent tailings dam failure. Proceeding of the Fourth International Conference on Forensic Engineering, Institution of Civil Engineers, London, UK. Cheng Y. M., Lau C. K., 2008 Slope stability analysis and stabilization: new methods and insight. CRC Press, 246 pp. Davies M., McRoberts E., Martin T., 2002 Static liquefaction of tailings fundamentals and case histories. Proceedings of Tailings Dams 2002, ASDSO/USCOLD, Las Vegas, 23 pp. Fourie A. B., Blight G. E., Papageorgiou G., 2001 Static liquefaction as a possible explanation for the Merriespruit tailings dam failure. Canadian Geotechnical Journal 38(4):707 719. Hamade T., Mitri H., Saad B., Pouliot S., 2011 Stochastic analysis of tailing dams stability using numerical modeling. In: Proceedings of 2011 Pan-Am CGS Geotechnical Conference, Toronto, Ontario, Canada, Oct 2-6, Paper No. 348, pp.1-8. Hammah R. E., Yacoub T. E., 2009 Probabilistic slope analysis with the finite element method. Asheville: the 43rd US Rock Mechanics Symposium and 4th US-Canada Rock Mechanics Symposium, 8 pp. INCDMRR-ICPMRR, 2011-2012-2014 [Hydro-geotechnical study for analyzing the current overall safety state of the Tărnicioara tailing pond (phase 1, 2 and 3)]. Bucharest, contract no. 203/01.05.2011 Conversmin [in Romanian] Knödel K., Lange G., Voigt H. J., 2007 Environmental geology - handbook of field methods and case studies. Springer Berlin Heidelberg New York, pp. 128-133. Mihai S., Deak S., Deak G., Oancea I., Petrescu A., 2008 Tailings dams and waste-rock dumps safety assessment using 3d numerical modeling of geotechnical and geophysical data. Proceedings of the 12th International Association for Computer Methods and Advances in Geomechanics 1-6 October, Goa, India, pp. 4212-4220. Poncos V., Serban F., Teleaga D., Ciocan V., Sorin M., Caranda D., Zamfirescu F., Andrei M., Copaescu S., Radu M., Raduca V., 2012 Water induced geohazards measured with spaceborne interferometry techniques. Geophysical Research Abstracts Vol. 14, EGU2012-4654. Rosenblueth E., 1975 Point estimates for probability moments. Proc Nat Acad Sci USA 72(10):3812-3814. Rosenblueth E., 1981 Two-point estimates in probabilities. Applied Mathematical Modelling 5(5):329-335. Valley B., Duff D., 2011 Probabilistic analyses in Phase2 8.0. CEMI - Center for Excellence in Mining Innovation, 6 pp. Vanden Berghe J. F., Ballard J. C., Wintgens J. F., List B., 2011 Geotechnical risks related to tailings dam operations. Proceedings Tailings and Mine Waste 2011, Vancouver, BC, 6-9 November, 11 pp. Wolff T. F., 1996 Probabilistic slope stability in theory and practice. In: Uncertainty in the Geologic Environment, ASCE, New York, pp. 419-433. *** Google Earth URL: https://www.google.com/earth/. *** ICOLD, 2001 Tailings dams - Risk of dangerous occurrences, Bulletin 121. Commission Internationale des Grands Barrages 151, bd Haussmann, 75008 Paris. 57

*** Indicative P 125-84, 1984 [Technical Guidelines for the study of non cohesive soil liquefiable properties], INCERC, Bucharest [in Romanian]. *** STAS 4273-83, 1983 [Water constructions works, integrating with importance classes]. Romanian Institute for Standardization [in Romanian]. *** NTLH-021 Technical Regulations, 2002 [The methodology for setting the dams categories of importance]. Ministry of Water and Environmental Protection [in Romanian]. *** WISE, 2014 Chronology of major tailings dam failure. Available at: http://www.wiseuranium.org/mdaf.html. Accessed: December, 2015. Received: 26 August 2015. Accepted: 20 December 2015. Published online: 30 December 2015. Authors: Diana Banu, Faculty of Geology and Geophysics, University of Bucharest, Traian Vuia Street, no. 6, District 2, Bucharest, Romania; Research and Development National Institute for Metals and Radioactive Resources (INCDMRR), Carol I Blv., no. 70, District 2, Bucharest, Romania, e-mail: banudianamaria@yahoo.com Sorin Mihai, Research and Development National Institute for Metals and Radioactive Resources (INCDMRR), Carol I Blv., no. 70, District 2, Bucharest, Romania, e-mail: sorinmihai2005@yahoo.com This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution and reproduction in any medium, provided the original author and source are credited. How to cite this article: Banu D., Mihai S., 2015 Integration of stochastic methods in the assessment of the risk generating phenomena. Study case - Tărnicioara tailings pond, Romania. Ecoterra 12(4):45-58. 58