NRAP-TRS-III-002-2014 Second-Generation Reduced-Order Model … Library/Research/onsite...

Second-Generation Reduced-Order Model for Calculation of Groundwater Impacts as a Function pH, Total Dissolved Solids, and Trace Metal Concentrations

7 April 2014

Office of Fossil Energy

NRAP-TRS-III-002-2014

Disclaimer This report was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights. Reference therein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States Government or any agency thereof. The views and opinions of authors expressed therein do not necessarily state or reflect those of the United States Government or any agency thereof.

This report (LLNL-TR-613412-DRAFT) has been reviewed by Lawrence Livermore National Laboratory (LLNL) and approved for public release.

Cover Illustration: Location of the model domain based on a site from the alluvium High Plains Aquifer in Kansas.

Suggested Citation: Carroll, S. A.; Mansoor, K.; Sun, Y. Second-Generation Reduced-Order Model for Calculating Groundwater Impacts as a Function of pH, Total Dissolved Solids, and Trace Metal Concentration; NRAP-TRS-III-002-2014; NRAP Technical Report Series; U.S. Department of Energy, National Energy Technology Laboratory: Morgantown, WV, 2014; p 40.

An electronic version of this report can be found at: http://www.netl.doe.gov/research/on-site-research/publications and https://edx.netl.doe.gov/nrap

Second-Generation Reduced-Order Model for Calculation of Groundwater Impacts as a Function of pH, Total Dissolved

Solids, and Trace Metal Concentration

Susan Carroll, Kayyum Mansoor, and Yunwei Sun

Atmospheric, Earth and Energy Division, Lawrence Livermore National Laboratory, 7000 East Avenue, Livermore, CA 94551

NRAP-TRS-III-002-2014

Level III Technical Report Series

7 April 2014

This page intentionally left blank.

Second-Generation Reduced-Order Model for Calculation of Groundwater Impacts as a Function of pH, Total Dissolved Solids, and Trace Metal Concentration

I

Table of Contents ABSTRACT ....................................................................................................................................1

1. INTRODUCTION...................................................................................................................3

2. METHOD ................................................................................................................................4 2.1 AQUIFER LITHOLOGIC MODEL ................................................................................6 2.2 LINKING LEAKAGE UNCERTAINTY TO THE STORAGE RESERVOIR AND

WELLBORE LEAKAGE MODELS ..............................................................................6 2.3 REACTIVE TRANSPORT MODEL ..............................................................................8 2.4 GEOCHEMICAL REACTIONS .....................................................................................9 2.5 UNCERTAINTY QUANTIFICATION ........................................................................10 2.6 IMPACT THRESHOLDS .............................................................................................12

3. RESULTS ..............................................................................................................................14 3.1 IMPACT OF CO2 LEAKAGE ON GROUNDWATER PH .........................................14 3.2 IMPACT OF BRINE LEAKAGE ON GROUNDWATER TOTAL DISSOLVED

SOLIDS .........................................................................................................................16 3.3 IMPACT OF BRINE LEAKAGE ON GROUNDWATER ARSENIC, CADMIUM,

CHROMIUM, IRON, LEAD, AND MANGANESE CONCENTRATIONS ...............18 3.4 FLUX OF CO2 TO ATMOSPHERE .............................................................................22

4. DISCUSSION ........................................................................................................................23

5. CONCLUSIONS ...................................................................................................................26

6. REFERENCES ......................................................................................................................29


II

List of Figures Figure 1: Plan view of Kimberlina model grid and possible leaky wells in High Plains

Aquifer. ....................................................................................................................................7

Figure 2: Example CO2 and brine leakage rates as functions of time using pressure and CO2/brine saturation shown in Figure 3. ..................................................................................8

Figure 3: Scaled reservoir pressure and CO2 saturation at the bottom of a leaky well. Note that the scaling factors for pressure and CO2 saturation are 0.657 and 0.452, respectively. .............................................................................................................................8

Figure 4: Flow chart of second-generation of groundwater impacts modeling. ............................10

Figure 5: Uncertain inputs considered in final reduced-order models. ..........................................12

Figure 6: Plots based on plume volumes for the impact and standard pH thresholds showing (top) time-based correlation coefficient analysis (R2) for emulated and simulated plume volumes, (middle) Sobol global sensitivity indices for the simulated model comparison of simulated and emulated results and (bottom) cumulative distribution profiles normalized to the number of emulations. ...........................15

Figure 7: Plots based on plume volumes for the impact TDS thresholds showing (top) time-based correlation coefficient analysis (R2) for emulated and simulated plume volumes, (middle) Sobol global sensitivity indices for the simulated model comparison of simulated and emulated results, and (bottom) cumulative distribution profiles normalized to the number of emulations. .................................................................17

Figure 8: Plots based on time-based correlation coefficient analysis (R2) for emulated and simulated plume volumes for the impact and standard trace metal thresholds. .....................19

Figure 9: Plots based on plume volumes for the impact and standard trace metal thresholds showing Sobol global sensitivity indices for the simulated model. .....................20

Figure 10: Plots based on plume volumes for the impact and standard trace metal thresholds showing cumulative distribution profiles normalized to the number of emulations. .............................................................................................................................21

Figure 11: Comparison of (top) the cumulative mass of CO2 input to the aquifer through wellbore leakage and (middle) the cumulative mass of CO2 released to the atmosphere. Color lines indicate the 0.05, 0.25, 0.50, 0.75, and 0.95 confidence levels (cl). ...............................................................................................................................22

Figure 12: Superposition of separated and merged plumes measured from two different simulated results plotted as plume volume. ...........................................................................25


III

List of Tables Table 1: Uncertain parameters and their ranges. Mean values for correlation length in y =

1,350 m, sand and shale porosity of 0.38 and 0.47, sand and shale van Genuchten m of 0.66 and 0.19, sand and clay van Genuchten α = 10-4.25 and 10-4.82 Pa-1, and CO2 diffusivity 10-9 m2s-1 ................................................................................................................5

Table 2: Uncertain trace metals and their ranges .............................................................................5

Table 3: Initial aquifer concentrations used in the simulations and estimated mean aquifer values, no-impact thresholds reported in Last et al. (2013). ..................................................13


IV

Acronyms, Abbreviations, and Symbols Term Description

3‐D Three‐dimensional

As Arsenic

CO2 Carbon dioxide

Cd Cadmium

cl Confidence level

CO2PENS Integrated systems/process model for simulating the fate of CO2

Cr Chromium

DOE Department of Energy

Fe Iron

EPA Environmental Protection Agency

LBNL Lawrence Berkeley National Laboratory

LLNL Lawrence Livermore National Laboratory

LP‐tau Quasi‐Monte Carlo method

m Meter

MCL Maximum contamination level

mg/L Milligrams per Liter

Mn Manganese

NATCARB National Carbon Sequestration Database and Geographic Information System

NRAP National Risk Assessment Partnership

NUFT Non‐Isothermal, Unsaturated and Saturated Flow and Transport Code

Pb Lead

pH ‐log{H+}

PSUADE Computer code of problem solving environment for uncertainty analysis and design exploration

R2 Linear correlation coefficient

ROM Reduced‐order model

TDS Total dissolved solids

T‐PROGS Transitional Probability Geostatistical Software

μg/L Micrograms per Liter

UQ Uncertainty quantification


V

Acknowledgments This work was completed as part of the National Risk Assessment Partnership (NRAP) project. Support for this project came from the Department of Energy’s (DOE) Office of Fossil Energy’s Crosscutting Research program. The authors wish to acknowledge Robert Romanosky (NETL Strategic Center for Coal) and Regis Conrad (DOE Office of Fossil Energy) for programmatic guidance, direction, and support.


VI



1

ABSTRACT

The National Risk Assessment Partnership (NRAP) is developing a science-based toolset for the quantitative analysis of the potential risks associated with changes in groundwater chemistry from CO2 injection. In order to address uncertainty probabilistically, NRAP is developing efficient, reduced-order models (ROMs) as part of its approach. These ROMs are built from detailed, physics-based process models to provide confidence in the predictions over a range of conditions. The ROMs are designed to reproduce accurately the predictions from the computationally intensive process models at a fraction of the computational time, thereby allowing the utilization of Monte Carlo methods to probe variability in key parameters.

This report presents the development of ROMs designed to predict the evolution of several groundwater metrics over time in response to leakage of CO2 and/or brine. The ROMs are based on simulations from continuum-scale reactive transport simulations in which the inherent uncertainties in the groundwater system were propagated throughout the predictive process. Lawrence Livermore National Laboratory’s (LLNL) focus was on the assessing the magnitude of the trace element source term found in CO2-rich brines, the impact of leakage from multiple wells, and aquifer heterogeneity. Potentially variable parameters that were considered include aquifer heterogeneity, permeability, porosity, regional groundwater flow, CO2, total dissolved solids (TDS), and trace metal leakage rates over time. Aquifer heterogeneity was characterized from the proportions of geologic units identified from lithologic well logs from the High Plains Aquifer to derive spatial correlation lengths; variation in CO2 and brine flux was derived from first-generation reservoir and wellbore models; and TDS and trace metal concentrations were based on reported storage reservoir compositions. The overall fidelity of the ROMs was very good with linear correlations greater than 0.9 when directly compared with simulated results.

We conducted 250,000 emulations to fully sample the uncertainty and to identify trends that may apply to similar systems. The study captures current understanding of the uncertainty inherent in the storage reservoir, leakage pathways, and aquifer heterogeneity and provides a statistically robust dataset for comparison with future emulations and validation of the component systems’ models being developed by NRAP. However, our results are specific to injection of 5 million tons of CO2 per year into a high permeability storage reservoir with up to 10% leaking wells, and an overlying dilute aquifer made up of alternating layers of permeable sands and less permeable shale. The 50-year injection period was followed by 150 years of emulated monitoring with no mitigation of the leaks. Main findings from the emulation study include:

1. A significant number of emulated leakage scenarios did not change groundwater composition to values that exceeded the impact or standard thresholds for our model system

2. During the post-injection period, plume number increased with minimal volume change

3. Trace element storage reservoir source terms should be included in full-scale geochemical calculations because the number of plumes grows over the 200-year emulation period, even though a significant number of leakage scenarios did not yield plumes

4. Groundwater plumes defined by the impact thresholds were generally more numerous and in the case of the trace metal plumes significantly larger than those defined by standard thresholds (10 to 40 times)


2

5. Only minor amounts of CO2 are released to the atmosphere


3

1. INTRODUCTION

Increasing anthropogenic contributions of carbon dioxide (CO2) is a primary contributor to global climate change that has global repercussions to human and ecological health (Haines et al., 2006; McCarty, 2001; McMicheal et al., 2006; Walther et al., 2002). Deep underground storage of CO2 from stationary sources such as power plants and industrial processes is a promising strategy that may mitigate the effects of global climate change (IEA, 2008; IPCC, 2005; NETL, 2010).

Geologic storage potential of CO2 is promising, but long-term storage of CO2 in deep underground reservoirs requires a careful assessment of the reservoir integrity, well and fault susceptibility for potential leakage pathways and impact of leaks into the atmosphere or on shallow groundwater sources (Bachu 2008, Benson, 2007, Herzog et al., 2003). Potential impacts to groundwater quality are a focus for various regulatory agencies (e.g., U.S. EPA, 2010), because leakage of brine and/or CO2 into groundwater resources and subsequent geochemical transformations may impact water quality (Apps et al., 2010; Wang and Jaffe, 2004). Accurate prediction of these impacts is complicated by uncertainties and variabilities in data and key parameters needed to describe leakage sources and the properties of aquifer systems. The full range of potential groundwater impacts can be assessed though rigorous iterative numerical modeling, but a complete treatment of these uncertainties and variabilities would be prohibitively time consuming from a computational standpoint.

The National Risk Assessment Partnership (NRAP) is developing a science-based toolset for the analysis of the potential risks associated with changes in groundwater chemistry from CO2 injection. This quantification approach is based on simulating coupled physical and chemical processes to predict how the natural system behaves over time, and it includes uncertainty quantification explicitly. In order to address uncertainty probabilistically, NRAP is developing efficient, reduced-order models (ROMs) as part of its approach. These ROMs are built from detailed, physics-based process models to provide confidence in the predictions over a range of conditions. However, the ROMs are designed to reproduce accurately the predictions from the computationally intensive process models at a fraction of the computational time, thereby allowing the utilization of Monte Carlo methods to probe variability in key parameters.

The natural system studied here is an unconsolidated aquifer consisting of layers of permeable sand and impermeable shale and is based on the lithology of the High Plains Aquifer. Input parameters considered include aquifer heterogeneity, permeability, porosity, regional groundwater flow, number of leaky wells, and CO2, total dissolved solids (TDS), and trace metal leakage rates over time. Variation in CO2 and brine flux was derived from NRAP’s first-generation ROMs for reservoir behavior and wellbore leakage, and TDS and trace metal concentrations were based on reported storage reservoir compositions (Wainwright et al., 2012; Jordan et al., 2103; NATCARB 2012; Karamalidis et al., 2013). The goal of this report is to describe the development of ROMs that predict the behavior of a given groundwater system, to assess the fidelity of the reduced-order models to capture the predictions from detailed simulations and to assess the importance of considering both multiple leakage sources and trace-metal source terms from the reservoir in order to fully describe groundwater impacts. The results also serve as a calibration data set for the larger systems model.


4

2. METHOD

In this section we describe the science-based methodology used to quantify impacts of CO2 and brine leakage from deep underground storage reservoirs, transmitted along multiple wellbores into a shallow groundwater aquifer. The process involved using an uncertainty quantification code to generate input parameters for 1,000 reactive-transport models that were sampled over defined parameter ranges to simulate physical and chemical processes, to conduct a sensitivity analysis of all varied parameters, and to build ROMs that provided statistical approximations of specific output that capture the uncertainty and allow impact to groundwater chemistry to be assessed.

Second-generation efforts focused on the uncertainties of leakage from multiple wells and elevated arsenic, cadmium, chromium, iron, lead, and manganese concentrations in the CO2-rich brine on groundwater quality. The basic structure for the simulations used (1) the storage-reservoir ROM (Wainwright et al., 2012) to generate over pressure and CO2/brine saturations for input to (2) the wellbore-leakage ROM (Jordan et al., 2013) to generate the leakage source terms for (3) the physics-based reactive transport aquifer model. We adopted this structure because CO2 and brine leakage through multiple wells should be tied to the over pressure from storage-reservoir model. Use of first-generation reservoir and wellbore leakage ROMs to generate a spatially resolved and time dependent leakage source terms for the groundwater model is one major distinction between the first and second groundwater impact studies. This approach allows parameter uncertainties from reservoir and wellbore to be directly passed to the aquifer simulations. For the aquifer model we used similar uncertainties in geological architecture, rock properties, and leakage source term. Uncertain parameters and their ranges are listed in Table 1. Physical parameters including porosity, density, permeability and van Genuchten unsaturated parameters were taken from the USDA Rosetta database (Schaap et al., 2001). Trace metal parameters for reservoir brine concentrations were taken from Karamalidis et al. (2012).


5

Table 1: Uncertain parameters and their ranges. Mean values for correlation length in y = 1,350 m, sand and shale porosity of 0.38 and 0.47, sand and shale van Genuchten m of 0.66 and 0.19, sand and clay van Genuchten α = 10-4.25 and 10-4.82 Pa-1, and CO2 diffusivity 10-9 m2s-1

x Definition Minimum Maximum Unit

1 Sand volume fraction of aquifer 0.35 0.65 [‐]

2 Correlation length of aquifer in x 200.00 2,500.00 [m]

3 Correlation length of aquifer in z 0.50 25.00 [m]

4 Sand permeability of aquifer ‐13.00 ‐10.00 Log10[m2]

5 Clay permeability of aquifer ‐18.00 ‐15.00 Log10[m2]

6 Brine molality ‐3.00 1.00 Log10[mol kg‐1]

7 Trace metal molality See Table 2

8 Permeability of wellbore ‐14.00 ‐10.00 Log10[m2]

9 Leaky‐well distance 2,000.00 5,000.00 [m]

10 Number of leaky wells 1.00 5.00 [‐]

11 Pressure scale 0.00 1.00 [‐]

12 Saturation scale 0.00 1.00 [‐]

Table 2: Uncertain trace metals and their ranges (Karamilidis et al., 2013)

x Trace Metal Minimum Maximum Unit

7 Arsenic ‐7.98 ‐5.87 Log10[mol kg‐1]

7 Cadmium ‐8.87 ‐6.43 Log10[mol kg‐1]

7 Chromium ‐6.42 ‐4.02 Log10[mol kg‐1]

7 Iron ‐6.07 ‐2.79 Log10[mol kg‐1]

7 Lead ‐8.12 ‐4.74 Log10[mol kg‐1]

7 Manganese ‐5.13 ‐2.10 Log10[mol kg‐1]


6

2.1 AQUIFER LITHOLOGIC MODEL

Aquifer heterogeneity was characterized from the proportions of lithologic units identified from the High Plains Aquifer to derive spatial correlation lengths from the observed distributions. The High Plains Aquifer is representative of a typical sedimentary aquifer that might overlay a CO2 storage reservoir. The aquifer, also known as the Ogallala Aquifer, is one of the largest aquifers in the world covering about 450,000 km2 and spans eight states in the Great Plains. The aquifer accounts for approximately 27% of all irrigated land in the United States and about 30% of all groundwater used for irrigation (USGS, 2011). It is comprised mainly of unconsolidated or partly consolidated deposits of silt, sand, gravel and clay rock debris deposited ∼2–6 million years ago in the late Miocene to early Pliocene period when the Rocky Mountains were tectonically active. The underlying bedrock units consists of siltstone, shale, loosely to moderately cemented clay and silt, chalk, limestone, dolomite, conglomerate, claystone, gypsum, anhydrite, and bedded salt (Gutentag et al., 1984).

The top 300 m of the aquifer was used to develop a lithological model for the geological realizations of the numerical model. Lithologic descriptions were obtained from the Kansas Geological Survey, Water Well Completion Record (WWC5) database (KGS, 2011). Approximately 100,000 picks from 11,454 wells were used to create a regional (162 km × 142 km) three-dimensional (3-D) lithological model. All simulations were based on a smaller 10 km × 5 km portion of the lithological model to represent the extent of the numerical model. This smaller domain lies primarily in Haskel County and was selected due to the relative higher density of lithological picks located at a depth sufficient enough to represent the model depth of 240 m. A total of 48 wells lie within this sub-domain. The wells are a mix of domestic, feedlot, irrigation, public water supply and oil field water supply wells, which make up 468 lithological picks.

The lithology was separated into two category classification representing high and low permeability zones and used to developed geostatistically derived indicator models using the TPROGS software (Carle, 1999). Correlation lengths were derived from the transition probability approach based on the lithology identified in the well logs. The correlation lengths in the x- and y-direction varied uniformly from ∼200–2,500 m and the correlation length in z-direction varied uniformly from ∼0.50–25.0 m. A total of 1,000 conditional geostatistical realizations were developed based on randomly selected material-volume fraction and correlation lengths populated using the PSUADE uncertainty quantification (Tong, 2005, 2010).

2.2 LINKING LEAKAGE UNCERTAINTY TO THE STORAGE RESERVOIR AND WELLBORE LEAKAGE MODELS

We superimposed well locations in the aquifer over the storage reservoir to generate specific CO2 and brine fluxes into the aquifer over time using the first-generation leakage wellbore-leakage ROM (Figure 1). Among 165 deep wells, which penetrate the CO2 reservoir, 48 were within a 5,000-meter radius of CO2 injector and were considered as possible leaky wells. The locations of the possible leaky wells were based on actual deep oil wells at the site and were mapped onto reservoir model domain. There is a significant amount of uncertainty on the number of wells that are likely to leak in a depleted oil reservoir. In this study we assume that for a given simulation, 1 to 5 wells leak (~ 2–10%) to span the percentage of wells that were observed to have leaked in Canadian oil fields (Watson and Bachu, 2009). Leaky wells were assumed to fully penetrate the caprock and connect the storage reservoir and the shallow aquifer.


7

Leakage rate was driven by reservoir over pressure and CO2 and brine saturations and were used as input to the wellbore leakage model to calculate the flux into the aquifer. The first-generation storage reservoir ROM was based on two-phase flow and uncertainties in reservoir permeability and porosity, and caprock permeability based on CO2 injection at the Kimberlina-site geology (Wainwright et al., 2012). Pressure and CO2/brine saturation resulting from the 54 simulations represent the lumped parametric uncertainty as a function of space and time. The first-generation wellbore leakage ROM was based on multiphase and non-isothermal flow simulations and uncertainties in wellbore permeability, reservoir pressure and CO2/brine saturation, and wellbore depth (Jordan et al., 2013). The resulting CO2 and brine fluxes to the aquifer increased during CO2 injection into the reservoir over 50 years and then remained constant after some initial adjustments associated with the discontinuation of active injection into the storage reservoir (Figure 2). This behavior appears to mimic the CO2/brine saturation in the reservoir in the injection and post-injection periods (Figure 3). It suggests a long-term leakage source in the absence of remediating the leak.

Figure 1: Plan view of Kimberlina model grid and possible leaky wells in High Plains Aquifer.


8

Figure 2: Example CO2 and brine leakage rates as functions of time using pressure and CO2/brine saturation shown in Figure 3.

Figure 3: Scaled reservoir pressure and CO2 saturation at the bottom of a leaky well. Note that the scaling factors for pressure and CO2 saturation are 0.657 and 0.452, respectively.

2.3 REACTIVE TRANSPORT MODEL

We simulate the geochemical response of a dilute aquifer to CO2 and brine leakage through wells from a CO2 storage reservoir using a multiphase and multi-component reactive transport code. NUFT (Non-Isothermal, Unsaturated and Saturated Flow and Transport) code (Nitao, 1998; Hao et al., 2010) is a highly flexible non-isothermal, multiphase flow and transport model based on the finite volume method, and has been extensively verified and used for a variety of subsurface flow and transport problems including nuclear waste disposal, groundwater remediation, CO2 sequestration and hydrocarbon production. To solve mass and energy equations NUFT employs an integrated finite-difference spatial discretization scheme and the resulting non-linear equations are solved by the Newton-Raphson method. The reactive transport will be solved in


9

the unsaturated and saturated zones assuming full multiphase flow including water-CO2 partitioning between the gas and liquid phases.

The numerical model domain was comprised of a 3-D bi-modally distributed heterogeneous domain representing the unsaturated and saturated zones of the High Plains Aquifer. The model domain extended to 10,000 m × 5,000 m × 240 m with 1 to 5 leakage sources placed at a depth of 198 m based on known well locations. The leakage source term was estimated from first- generation reservoir and wellbore ROMS (see Section 2.2). Isothermal conditions were assumed with a generic temperature of 17°C in the entire domain. The uppermost portion of the model was set as atmospheric allowing for both saturated and unsaturated conditions. The east (minimum x) and west (maximum x) model boundaries were fixed in time. A hydrostatic-pressure gradient was imposed by changing the gravity vector. No-flow boundaries were assumed at the southern (minimum y), northern (maximum y) and bottom (minimum z) boundaries. A constant-pressure boundary condition was set on ground surface and at the aquifer bottom to maintain a hydrostatic initial condition with saturated and unsaturated zones. Regional groundwater flow was maintained by a 0.3% hydraulic gradient. The bi-modal lithologic units represent a high-permeability sand and low-permeability clay and were distributed throughout the domain as described in Section 2.2. The orthogonal numerical grid contained fixed cell widths in the x (Δx = 100.0 m), y (Δy = 100.0 m), and z (Δz = 4.8 m) directions. Since the regional groundwater flow of the Great Plains Aquifer in southwestern Kansas flows eastward, the mesh is structured to accommodate flow in the predominant x-direction. The grid dimensions were 100, 50, and 50 nodes in the x, y, and z directions respectively for a total of 250,000 nodes. Each simulation was executed for 20–60 hours each using the high performance computing facility at Lawrence Livermore National Laboratory (LLNL). Parallelization of the simulations and subsequent analysis were required, otherwise this work would not be achievable.

2.4 GEOCHEMICAL REACTIONS

We include a limited amount of chemistry to account for changes in groundwater pH due CO2 dissolution as well as dissolved sodium and chloride as indicators of TDS. We do not include trace metal geochemistry here because we simply want to assess the need for including storage reservoir source terms for certain metals. The dissolution of CO2 in groundwater promotes the following sets of reactions:

CaCO3(calcite)H Ca2 HCO3

These reactions promote the acidification of the system, which is then buffered by calcite dissolution. Background pH, TDS, and trace metal concentrations in the aquifer are listed in Table 3 and are based on mean values reported for the regional aquifer in 2010 or previous literature values (Last et al., 2013; Carroll et al., 2009).

CO2 (g)CO2 (aq)

CO2 (aq)H2OHCO3 H


10

2.5 UNCERTAINTY QUANTIFICATION

Reactive-transport models are defined as simulation of physical and chemical processes and ROMS (also called a response function) as emulations of statistical approximations of specific output that capture the uncertainty in the input parameters. Conceptually, this framework, as shown in Figure 4, is an iterative process whereby each new generation could have a reduced parameter set, define narrower bounds on the parameter distributions, and or increased complexity resulting in greater confidence in the risk profiles. In theory, the polynomial-based higher order models should be able to emulate the outcome of the numerical model but with less complexity, more thorough sampling of the parameter space, and significantly faster simulation times to generate risk-based profiles.

Figure 4: Flow chart of second-generation of groundwater impacts modeling.

The PSUADE code was developed for uncertainty quantification tasks such as forward uncertainty propagation, qualitative and quantitative sensitivity analysis, parameter exploration, risk analysis, and numerical optimization (Tong 2005, 2010). It employs the non-intrusive or sampling-based approach to UQ that does not require simulation codes to be modified and ensures that it can easily be integrated with a variety of application simulators. PSUADE is equipped with many response-surface generation and validation techniques. These techniques can be coupled with other UQ techniques such as numerical optimization and Markov Chain Monte Carlo methods for calibration and parameter estimation.


11

We used PSUADE to establish sampling points for the reactive transport simulations, to conduct parameter sensitivity analysis, and to develop ROMS for the simulated results. We used the Latin hypercube (McKay et al., 1979) method to produce 1,000 sample points for each of the groundwater models, a Sobol sensitivity analysis of the successful runs to test the overall model response to individual parameters for each time step, and LP-tau (Shukhman, 1994) to produce 250,000 sample points using the ROM to produce the necessary statistical output for UQ.

ROMs offer a cost-effective representation of a full-scale system. These emulations can represent objective functions captured by full physics simulations by considering the same inputs (x) and response variable (y) as presented in a higher-order polynomial:

y=a0+

i a

ix

i+

i

j a

ijx

ix

j+

i

j

k a

ijkx

ix

jx

k+…

where a represents the polynomial fitting coefficients and i, j and k are indices of the uncertain parameters. This reduced analytical form permits a greater magnitude of model simulations and encompass a more robust estimation of probability due to the increased parameter sampling density.

Wellbore permeability, leaky-well distance, pressure and saturation scale factors, as well as the three reservoir parameters, were represented by the uncertainties of wellbore leakage outputs to simplify the implementation of groundwater ROMs in CO2 PENS (Stauffer et al., 2009; Jordan et al., 2013). ROMs were developed using CO2 and brine leakage flux, their cumulative masses, and time as uncertain parameters (Figure 5). We treated time as a special dimension in ROM construction, because most all risk measures are time dependent. This allows a single ROM for each threshold to capture the dynamic process over the entire simulation period. Otherwise, multiple ROMs should be constructed separately in each time interval (Sun et al., 2013).


12

Figure 5: Uncertain inputs considered in final reduced-order models.

2.6 IMPACT THRESHOLDS

The ROMs generated in this study are specific to either “impact” or “standard” thresholds defined in Table 3. The impact threshold represents concentrations above the background water chemistry. It is calculated as the 95%-confidence, 95%-coverage tolerance from data collected in a 2010 USGS groundwater survey of 30 wells within the High Plains Aquifer from an area outside of lithologic model site. This data set was chosen because spatial and temporal data were not available from model site. The standard threshold refers to concentrations that exceed primary or secondary maximum contaminant levels designated by the U.S. EPA (2009). Primary drinking water standards are for trace metals, such as As, Cd, Cr, Cu, and Pb, among others, and are legally enforced for the protection of public health by limiting the levels of contaminants in drinking water. Secondary drinking water standards, which include standards for Fe, Mn, and Zn, are non-enforceable guidelines regulating contaminants that may cause cosmetic or aesthetic effects in drinking water.


13

Table 3: Initial aquifer concentrations used in the simulations and estimated mean aquifer values, no-impact thresholds reported in Last et al. (2013)

Analyte Initial Value used in Second‐Generation Simulations

Mean of Selected and Adjusted 2010

Datab

Impact Thresholdc US EPA Regulatory Standard

pH 7.600a 7.500c 6.625f 6.500

Total Dissolved Solids

570.000 mg/la,d 440.000 mg/ld 1300.000 mg/ld,e 500.000 mg/le

Arsenic 1.500 μg/l 1.500 μg/l 9.300 μg/l 10.000 μg/l

Cadmium 0.059 μg/l 0.059 μg/l 0.250 μg/l 5.000 μg/l

Chromium 1.000 μg/l 1.000 μg/l 3.900 μg/l 100.000 μg/l

Iron 5.400 µg/l b 5.400 µg/l b 43.000 µg/l b 300.000 µg/l

Lead 0.086 μg/l 0.086 μg/l 0.630 μg/l 15.000 μg/l

Manganese 0.350 µg/l d 0.350 µg/l d 7.000 µg/l d 50.000 µg/l

(a) Based on Carroll et al. (2009)

(b) Geometric mean except for pH, which is already a log value

(c) 95%‐confidence, 95%‐coverage tolerance limit based on log values except for pH, which is already a log value

(d) Rounded to two significant digits

(e) Threshold value exceeds regulatory standard, however using the regulatory standard may result in widespread false positives under field conditions

(f) Value is lower than no‐impact threshold estimated by Last et al (2013) because ROMs at higher threshold produced non‐physical results


14

3. RESULTS

In this section we evaluate the ability of the emulations to reproduce the simulated results, parameter sensitivity for each threshold, and compare the emulation results for impact and standard thresholds for pH, TDS, As, Cd, Cr, Fe, Mn and Pb (Table 3). The emulation results are specific to our model system which consists of a high permeability storage reservoir, up to 10% leaking wells, and an aquifer with alternating layers of permeable sands and less permeable shales. The results measure our current understanding of the contribution of the leakage source on groundwater quality given the inherent uncertainty in the storage reservoir, leakage pathway, and dilute aquifer for this model system. The emulated results provide a statistically robust dataset for comparison with future emulations and validation of the component systems models being developed by NRAP. We also analyze the simulated results to provide a measure to the amount CO2 that finds a pathway through the dilute aquifer to the atmosphere.

3.1 IMPACT OF CO2 LEAKAGE ON GROUNDWATER PH

We developed high-fidelity ROMs from the simulated data for the standard threshold pH 6.5, but not for the impact threshold of pH 7.0 because the ROMs yielded plume volumes that were smaller than those derived from more acid pH thresholds. We have redefined the impact threshold to pH 6.625 because we were not able to produce physically reasonable plumes at greater pH. ROM fidelity is plotted for the impact and standard thresholds as R2 versus time based on 14,620 sample points in Figure 6. Inspection of the R2 values at each time step shows good agreement between the emulation and simulation after 30 years for both thresholds (R2 > 0.9). Poor agreement during the first 30 years reflects too few sampling points for the 4th order polynomial fit to the simulated data. We limit the presentation of our results to times between 40 and 200 years where the fidelity of the ROMs are greatest.

Figure 6 plots plume volume as the cumulative distribution at representative time intervals for CO2 injection (40 years) and post injection (200 years) for both thresholds. The parameter uncertainty investigated in this study yields a bimodal distribution of leakage scenarios that either did or did not produce plumes. As expected, the plume number and volume is higher for the impact threshold pH 6.625 than for the standard threshold pH 6.5, because the impact plume captures a slightly larger range of acidity than the standard threshold. CO2 leakage changed pH from initial value of 7.6 to pH ≤ 6.625 and pH ≤ 6.5 for about 60% and 57% of the emulations, respectively. The number of plumes increased by an additional 15% at 200 years for both thresholds, suggesting a long-term term change to solution pH in cases with no mitigation of the leakage source as was the case for the model system studied here. Although the emulations suggest an increase in plume volume between the injection and post injection periods, volumes at 200 years are only 1.5 to 2 times the size of the plume at 40 years.

There is some variation in the Sobol global sensitivity indices of the variable parameters between the two thresholds. Plume volumes of both thresholds were most sensitive to cumulative CO2 mass (~1). This is consistent with CO2 solubility in groundwater and its subsequent dissociation of H2CO3 (aqueous) to HCO3

- and H+. The impact threshold was also sensitive to variations of volume fraction of the sand layer (~1.1) but the slightly more acid standard threshold was not. The collective uncertainty of our model system yields plumes that span between 106 to 108.7 m3.


15

Figure 6: Plots based on plume volumes for the impact and standard pH thresholds showing (top) time-based correlation coefficient analysis (R2) for emulated and simulated plume volumes, (middle) Sobol global sensitivity indices for the simulated model comparison of simulated and emulated results and (bottom) cumulative distribution profiles normalized to the number of emulations.


16

3.2 IMPACT OF BRINE LEAKAGE ON GROUNDWATER TOTAL DISSOLVED SOLIDS

We were able to develop a high-fidelity TDS ROM from the simulated data for the1,300 mg/L impact threshold but not for the 500 mg/L standard threshold, because it is lower than the initial aquifer TDS. ROM fidelity is plotted for the impact threshold as R2 versus time based on 14,620 sample points in Figure 7. Inspection of the R2 values at each time step shows good agreement between the emulation and simulation between 30 and 200 years (R2 > 0.9). Poor agreement during the first 30 years reflects too few sampling points for the forth-order polynomial to fit to the simulated data. We limit the presentation of our results to times between 40 and 200 years where the fidelity of the ROM is greatest. Sobol global sensitivity indices of the variable parameters shows that plume volumes were most sensitive to brine composition and sand volume fraction (~0.9), followed by the cumulative brine mass (~0.6), and mean brine flux (~0.3).

Figure 7 plots plume volume as the cumulative distribution at representative time intervals for CO2 injection (40 years) and post injection (200 years) for the impact threshold. Like the pH plume results, the parameter uncertainty investigated in this study yields a bimodal distribution of leakage scenarios that either did or did not produced plumes, and the plume number and volume is higher at 200 years than at 40 years. Brine leakage changed TDS from initial value of 570 mg/L to 1,300 mg/L for only 26% of the emulations at 40 years and 42% at 200 years. Although the emulations suggest an increase in plume volume between the injection and post injection periods, volumes at 200 years are at most 1.2 the size of the plume at 40 years. The resulting volumes span between 106 to 108.1 m3, slightly smaller to the range estimated for pH.


17

Figure 7: Plots based on plume volumes for the impact TDS thresholds showing (top) time-based correlation coefficient analysis (R2) for emulated and simulated plume volumes, (middle) Sobol global sensitivity indices for the simulated model comparison of simulated and emulated results, and (bottom) cumulative distribution profiles normalized to the number of emulations.


18

3.3 IMPACT OF BRINE LEAKAGE ON GROUNDWATER ARSENIC, CADMIUM, CHROMIUM, IRON, LEAD, AND MANGANESE CONCENTRATIONS

We were able to construct high fidelity impact and standard threshold ROMs for As, Cd, Cr, Fe, Mn, and Pb from the simulated data. Figure 8 plots R2 values at each time step and shows good agreement between the emulation and simulation between 20 and 200 years (and in some cases 10 and 200 years) for all thresholds. We limited the presentation of our results to times between 40 and 200 years to provide a consistent comparison among the trace metal, TDS and pH thresholds. Plume volumes appear to be most sensitive to the cumulative mass of brine leaked into the aquifer and trace metal concentration of the brine (Figure 9). The Sobol sensitivity analyses also suggest that the number of leaking wells also affects the plume size for many of the trace metals.

Figure 10 compares the cumulative distribution profiles for the trace metal impact and standard thresholds at 40 and 200 years. Very few of the leakage scenarios resulted in groundwater plumes. After 40 years of CO2 injection, only 8% of the scenarios yield impact plumes for As, 15% for Mn, 18% for Cd, 19% for Fe, 21% for Pb, and 24% for Cr. These numbers increase during the post injection phases to 15% for As, 26% for Mn, ~30% for Cd and Fe, 34% for Pb, and 38% for Cr. The overall size of the impact plumes is similar to those for pH and TDS, ~106 to 108.2 m3.

The emulations produced an even lower number of plumes for the higher standard thresholds, where only 2% of the scenarios yield standard plumes after 40 years of CO2 injection for Cd, 5% for Cr, and between 7–10 % for Pb, Mn, Fe, and As. The number of standard threshold plumes increased during the post injection phase to 8% for Cd and between 12–16% for Cr, Pb, As, Fe, and Mn at 200 years. The overall size of the standard plumes is about 10 to 40 times smaller than the impact plumes, ~105 to 107.2 m3.


19

Figure 8: Plots based on time-based correlation coefficient analysis (R2) for emulated and simulated plume volumes for the impact and standard trace metal thresholds.


20

Figure 9: Plots based on plume volumes for the impact and standard trace metal thresholds showing Sobol global sensitivity indices for the simulated model.


21

Figure 10: Plots based on plume volumes for the impact and standard trace metal thresholds showing cumulative distribution profiles normalized to the number of emulations.


22

3.4 FLUX OF CO2 TO ATMOSPHERE

Minimal CO2 leaked from the upper reaches of the aquifer to the atmosphere in the layered aquifer studied here. Figure 11 plots the cumulative mass of CO2 from the leaking wellbores and to the atmosphere as a function of time for each simulation. Although there is a wide range, the median cumulative mass of CO2 released to the atmosphere is between 1,000 and 10,000 times smaller than the mass of CO2 introduced to the aquifer model through wellbore leakage. The layered structure of the aquifer tends to trap CO2 at a depth resulting in its continued dissolution in the groundwater. We were not able to produce a robust ROM that captured the simulated results and recommend that the risk be assessed as a defined percentage. We recommend that fixed values between 0.01 and 0.1% of the cumulative CO2 mass leaked into the aquifer from the source term (wells or faults) be used as a surrogate model for CO2 leakage to the atmosphere overlying a layered sandstone/shale aquifer.

Figure 11: Comparison of (top) the cumulative mass of CO2 input to the aquifer through wellbore leakage and (middle) the cumulative mass of CO2 released to the atmosphere. Color lines indicate the 0.05, 0.25, 0.50, 0.75, and 0.95 confidence levels (cl).

(a) (b)


23

4. DISCUSSION

The simulation/emulation study captures our current understanding of the uncertainty inherent in the storage reservoir, leakage pathways, and aquifer heterogeneity. However, our results are specific to injection of 5 million tons of CO2 per year into a high-permeability storage reservoir with up to 10% leaking wells, and an overlying dilute aquifer made up of alternating layers of permeable sands and less permeable shale. The 50-year injection period was followed by 150 years of emulated monitoring with no mitigation of the leaks. Main findings from the emulation study include:


2. During the post-injection period, plume number increased with minimal volume change


4. Groundwater plumes defined by the impact thresholds were generally more numerous and in the case of the trace metal plumes significantly larger than those defined by standard thresholds (10 to 40 times)


There appears to be range of leakage rates that do not change groundwater chemistry to values above the impact or standard thresholds. This statement is based on the high number of emulations that did not produce plumes (20–40 % for pH; 60–75% for TDS; 60–98% for trace metals) and sensitivity of all plumes on the cumulative brine or CO2 mass leaked into the aquifer. The cumulative brine or CO2 mass is a consequence of the number of leaky wellbores, wellbore permeability and depth, and reservoir pressure and CO2/brine saturation that feed the reservoir and wellbore leakage ROMs. This result supports the intuitive notion that wellbore characterization and management may be the most effective way of minimizing risks to quality of dilute aquifers. Site characterization should identify and refurbish all potential leaking wellbores a reasonable distance from injection well. This distance corresponds to a 5 km radius for this synthetic model, but is likely to vary depending on specific injected volumes and times and reservoir properties for specific CCS sites.

One conceptual model for the mitigation of groundwater quality caused by leakage of storage reservoirs is to stop injection, allow the pressure and leakage rates to fall off, and allow natural recharge and dispersion to dilute groundwater chemistry to levels below the prescribe thresholds. Our results show that any reductions in CO2 or brine leakage rates as a result of lower reservoir pressures together with natural recharge and dispersion did not fully mitigate groundwater quality after 200 years of emulation. The plumes were more abundant and less than 2 times larger. This result highlights the need to develop monitoring programs that detect wellbore leakage in the aquifer as quickly as possible so that wellbores can be remediated and leakage can be stopped. We hypothesize that the aquifer would recover more rapidly, if the smaller amounts of CO2/brine were leaked into the reservoir system. It also highlights the utility of measuring


24

groundwater against the impact thresholds, which would allow earlier detection of plumes because the impact thresholds are typically lower than the regulatory standards. We are currently conducting a simulation/emulation study to evaluate monitoring and mitigation strategies.

In this study we included the trace metal source term from the CO2-storage reservoir to assess if the range of trace metal concentrations determined by experiment (Karamilidis et al., 2013) would change the groundwater quality in the absence of any additional chemical reactions that might limit the brine or aquifer concentrations. Inclusion of the trace metal source term is particularly important if impact thresholds are the proposed metric for groundwater quality because emulations predict a greater probability and larger plume volume. Future studies should account for changes in dissolved trace metal concentrations caused by rock-water reactions along leakage pathways and within the dilute groundwater aquifer. Trace metal concentrations are likely to decrease as CO2 degasses, pH increases, and carbonate and iron-hydroxide minerals precipitate as temperature and pressure decrease along the reaction pathway. Iron, manganese and chromium concentrations will also change as the redox environment changes along the leakage pathway and within the aquifer. It is likely that sorption, oxidation, and the precipitation of iron and manganese hydroxides will lower trace metal concentrations as plume acidity is diluted in the aquifer.

Parameter sensitivity analysis of the simulated results can also be used to identify which parameters need to be constrained to lower the overall uncertainty of groundwater impacts for a given project. Better knowledge of the cumulative brine and CO2 mass leaking into the aquifer should narrow the uncertainty for groundwater impacts for the pH, TDS, and trace metal thresholds used in this study. As was mentioned above, the cumulative brine or CO2 mass is a consequence of the number of leaky wellbores, wellbore permeability and depth, and reservoir pressure and CO2/brine saturation that feed the reservoir and wellbore leakage ROMs. Of these parameters, the most poorly understood parameter is wellbore permeability because it is likely to change upon reaction with CO2 and brine mixtures (Mason et al., 2012, Walsh et al., 2012, Huerta et al., 2013; Scherer et al., 2011; Carey et al., 2007, 2009; Kutchko et al., 2007, 2008, 2009). Uncertainty could also be lowered by assessing the integrity of existing wellbores, especially depleted oil fields where a great number of wells penetrate the shallow, dilute groundwater and deep, saline storage zones. Characterization of the chemical source term for the reservoir will also narrow the uncertainty of estimated TDS and trace metal impacts. This can be done by measuring the down hole reservoir brine composition and conducting relevant experiments to measure the steady-state/equilibrium trace metal concentrations in the storage-brine after injection and reaction with supercritical CO2. Prior to conducting this study, we thought that aquifer heterogeneity would drive plume volumes. However, low sensitivities of spatial correlation lengths and lithology permeability on plume volumes suggest that this is not the case, at least for systems in which the lithology models are conditioned to site data, as was the case in this study.

A final topic for discussion is the notion that groundwater impacts from individual leakage sources can be added together to estimate the total volume of impacted groundwater. First-generation integrated assessment models have treated plumes resulting from leaking wells to be additive. Analysis of the simulated data set for pH plumes shows that this is not always the case. Numerical simulations #298 and #315 each had four leaky wells. The pH plumes for run #298 were small enough to develop separately over 200 years of simulation, while plumes of run #315 merged after less than 30 years of simulation (Figure 12). In the case where plumes from


25

different well locations merge with one another, linear superposition underestimates the impact defined by pH ≤ 6.5. Superposition leads to smaller volumes because acidity was more readily diluted to values below the threshold than merged plumes. The dilution effect should also hold for TDS and trace metals, but was not explicitly investigated.

Figure 12: Superposition of separated and merged plumes measured from two different simulated results plotted as plume volume.

Sample 298 Sample 315

(a) (b)


26

5. CONCLUSIONS

We developed reduced-order models that describe changes in dilute groundwater chemistry if CO2 and brine were to leak through multiple wellbores to an overlying layered sand/silt aquifer similar to the High Plains Aquifer, USA. We assessed the magnitude of the trace element source term in CO2-rich storage brines, the impact of leakage from multiple wells, and aquifer heterogeneity on groundwater quality. Input parameters included aquifer heterogeneity, permeability, porosity, regional groundwater flow, and CO2, TDS, and trace metal leakage rates over time. Aquifer heterogeneity was characterized from the proportions of geologic units identified from well logs from the High Plains Aquifer, variation in CO2 and brine flux was derived from first-generation reservoir and wellbore models, TDS and trace metal concentrations were based on reported storage reservoir compositions. The overall fidelity of the ROMs was very good with linear correlations greater than 0.9 when directly compared with simulated results.

The study captures current understanding of the uncertainty inherent in the storage reservoir, leakage pathways, and aquifer heterogeneity and provides a statistically robust data set for comparison with future emulations and validation of the component systems models being developed by NRAP. However, our results are specific to injection of 5 million tons of CO2 per year into a high-permeability storage reservoir with up to 10% leaking wells, and an overlying dilute aquifer made up of alternating layers of permeable sands and less permeable shale. The 50-year injection period was followed by 150 years of emulated monitoring with no mitigation of the leaks. Main findings from the emulation study include:


2. Plume number and volume increase during the post-injection period


4. Groundwater plumes defined by the impact thresholds were generally more numerous and in the case of the trace metal plumes significantly larger than those defined by standard thresholds


We have demonstrated that emulations calibrated against simulated data are a useful methodology to assess the impact of leakage on groundwater quality. However, there are a couple limitations of this study in particular. Firstly, the emulated results are specific to a predefined thresholds. In this study we define two possible thresholds that represents a statistical change in groundwater chemistry or a change the meets or exceeds regulatory standards. Secondly, the models do not provide an indication of the relative plume size, nor do the models assess the role of groundwater usage (such as irrigation) on plume distribution and evolution. The reported volumes are the combined volumes for each leaking well and may span a larger spatial region than represented solely by the emulated plume volumes. We are currently using


27

post-imaging techniques to better understand the size of the impacted area and the relative position of the plumes to available drinking and agricultural wells within the High Plains model domain.


28



29

6. REFERENCES

Apps, J. A.; Zheng, L.; Zhang, Y.; Xu, T.; Birkholzer, J. T. Evaluation of potential changes in groundwater quality in response to CO2 leakage from deep geologic storage, Transp. Porous Med. 2010, 82, 215–246.

Bachu, S. CO2 storage in geological media: Role, means, status, and barriers to deployment, Prog. Energ. Combust. 2008, 34, 254–273.

Benson, S. Addressing long–term liability of carbon dioxide capture and geological sequestration. Technical Report. World Resource Institute (WRI) Long–term liability Workshop, Washington, DC, June 7, 2007.

Carroll, S.; Hao, Y.; Aines, R. Geochemical detection of CO2 in dilute aquifers. Geochemical Transactions 2009, 10, 4.

Carey, J. W.; Wigand, M.; Chipera, S. J.; WoldeGabriel, G.; Pawar, R.; Lichtner, P. C.; Wehner, S. C.; Raines, M. A.; Guthrie, G. D., Jr. Analysis and performance of oil well cement with 30 years of CO2 exposure from the SACROC Unit, West Texas, USA. International Journal of Greenhouse Gas Control 2007, 1, 75–85.

Carey, J. W.; Svec, R.; Grigg, R.; Lichtner, P. C.; Zhang, J.; Crow, W. Wellbore integrity and CO2-brine flow along the casing-cement microannulus. Greenhouse Gas Control Technologies 2009, 9.

Carle, S.F. T-PROGS: Transitional probability geostatistical software users' guide; University of California, Davis, CA, 1999.

Gutentag, E.; Heimes, F.; Krothe, N.; Luckey, R.; Weeks, J. Geohydrology of the High Plains Aquifer in parts of Colorado, Kansas, Nebraska, New Mexico, Oklahoma, South Dakota, Texas, and Wyoming; Technical Report Professional Paper, 1400–B; U.S. Geological Survey, 1984.

Haines, A.; Kovats, R.; Campbell-Lendrum, D.; Corvalan, C. Climate change and human health: Impacts, vulnerability and public health. Public Health 2006, 120, 585–596.

Hao, Y.; Sun, Y.; Nitao, J. J. Overview of NUFT: a versatile numerical model for simulating flow and reactive transport in porous media. In Groundwater Reactive Transport Models; Bentham Science Publishers, 2010.

Herzog, H.; Caldeira, K.; Reilly, J. An issue of permanence: assessing the effectiveness of temporary carbon storage. Clim Change 2003, 59, 293–310.

Huerta, N. J.; Hesse, M. A.; Bryant, S. L.; Strazisar, B. R.; Lopano, C. L. Experimental Evidence for Self-Limiting Reactive Flow through a Fractured Cement Core: Implications for Time-Dependent Wellbore Leakage. Environ. Sci. Technol. 2013, 47, 269–275.

IEA. CO2 Capture and storage: A key carbon abatement option; OECD Publishing, 2008; pp 264.

IPCC. IPCC special report on carbon dioxide capture and storage; Technical Report. Intergovernmental Panel on Climate Change; Prepared by Working Group III of the Intergovernmental Panel on Climate Change; Metz, B., Davidson, O., de Coninck, H. C.,


30

Loos, M., Meyer, L. A., Eds.; Cambridge University Press: Cambridge, United Kingdom and New York, NY, 2005; pp 442.

Jordan, A. B.; Stauffer, P. H.; Harp, D. R.; Carey, J. W.; Pawar, R. J. A Method for Predicting CO2 and Brine Leakage from Geologic Sequestration along Cemented Wellbores in System-Level Models. Los Alamos National Laboratory Rep. LA-UR-13-29243; 2013.

KGS, Kansas Geological Survey. Wells, logs, core, and other databases, 2011. http://www.kgs.ku.edu/PRS/petroDB.html accessed on 9/30/2011

Karamalidis, A. K.; Torres, S. G.; Hakala, A. J.; Shao, H.; Cantrell, K. J.; Carroll, S. A. Trace Metal Source terms in carbon sequestration environments. Environ. Sci. Technol. 2013, 47, 322–329. DOI: 10.1021/es304832m

Kutchko, B. G.; Strazisar, B. R.; Dzombak, D. A.; Lowry, G. V.; Thaulow, N. Degradation of well cement by CO2 under geologic sequestration conditions. Environ. Sci. Technol. 2007, 41, 4787–4792.

Kutchko, B. G.; Strazisar, B. R.; Lowry, G. V.; Dzombak, D. A.; Thaulow, N. Rate of CO2 attack on hydrated Class H well cement under geologic sequestration conditions. Environ. Sci. Technol. 2008, 42, 6237–6242.

Kutchko, B. G.; Strazisar, B. R.; Huerta, N.; Lowry, G. V.; Dzombak, D. A.; Thaulow, N. CO2 Reaction with hydrated Class H well cement under geologic sequestration conditions: effects of flyash admixtures. Environ. Sci. Technol. 2009, 43, 3947–3952.

Last, G. V.; Brown, C. F.; Murray, C. J.; Bacon, D. H.; Qafoku, M. P.; Sharma, M.; Jordan, P. D. Determining threshold values for identification of contamination predicted by reduced order models; PNNL-22077; Pacific Northwest National Laboratory, Richmond, WA, 2013.

Mason, H. E.; Du Frane, W. L.; Walsh, S. D. C.; Dai, Z.; Charnvanichborikarn, S.; Carroll, S. A. (2013). Chemical and mechanical properties of wellbore cement altered by CO2-rich brine using a multianalytical approach. Environ. Sci. Technol. 2013, 47, 1745–1752. DOI: 10.1021/es3039906.

McCarty, J. Ecological consequences of recent climate change. Conservation Biology 2001, 15, 320–331.

McKay, M. D.; Beckman, R. J.; Conover, W. J. A comparison of three methods for selecting values of input variables in the analysis of output from a computer code. Technometrics 1979, 21, 239–245. DOI: 10.2307/1268522

McMichael, A.; Woodruff, R.; Hales, S. Climate change and human health: present and future risks. The Lancet 2006, 367, 859–869.

NATCARB. National Carbon Sequestration Database and Geographic Information System, www.natcarbviewer.com (accessed June 2012).

NETL. 2010 Carbon sequestration atlas of the United States and Canada, 3rd ed.; Technical Report; U.S. Department of Energy, Office of Fossil Energy, National Energy Technology Laboratory, 2010.


31

Nitao, J. J. User's manual for the USNT module of the NUFT code, version 2 (NP-phase, NC-component, thermal); UCRL-MA-130653; Lawrence Livermore National Laboratory, Livermore, CA, 1998.

Schaap, M.; Leij, F.; van Genuchten, M. Rosetta: A computer program for estimating soil hydraulic parameters with hierarchical pedotransfer functions. J. of Hydrology 2001, 251, 163–176.

Scherer, G. W.; Kutchko, B.; Thaulow, N.; Duguid, A.; Mook, B. Characterization of cement from a well at Teapot Dome Oil Field: Implications for geological sequestration. International Journal of Greenhouse Gas Control 2011, 5, 115–124.

Shukhman, B. Generation of quasi-random (LPτ) vectors for parallel computation. Comp. Phys. Commun. 1994, 78, 279– 286.

Stauffer, P.; Viswanathan, H.; Pawar, R.; Guthrie, G. A system model for geologic sequestration of carbon dioxide. Environ. Sci. Technol. 2009, 43, 565–570.

Sun, Y.; Tong, C.; Trainor-Guitton, W. J.; Lu, C.; Mansoor, K.; Carroll, S. A. Global sampling for integrating physics-specific subsystems and quantifying uncertainties of CO2 geological sequestration. Int. J. Greenh. Gas Control 2013, 12, 108–123. dx.doi.org/10.1016/j.ijggc.2012.10.004

Tong, C. PSUADE User’s Manual; LLNL-SM-407882; Lawrence Livermore National Laboratory, Livermore, CA, 2005.

Tong, C. Self–validated variance–based methods for sensitivity analysis of model outputs. Reliab. Eng. Syst. Safe. 2010, 95, 301–309.

U.S. EPA Code of Federal Regulations, 40 CFR Part 141 and 143, EPA 816-F-09-0004, 2009.

U.S. EPA. Federal requirements under the underground injection control (UIC) Program for carbon dioxide (CO2) geologic sequestration (GS) Wells; Technical Report; U.S. Environmental Protection Agency, 2010.

U.S. Geological Survey (USGS). National water–quality assessment (nawqa) program: High plains regional ground water (hpgw) study, 2011. http://co.water.usgs.gov/nawqa/hpgw/HPGW_home.html accessed on 9/1/2011

Wainwright, H.; Finsterle, S.; Zhou, Q.; Birkholzer, J. Modeling the Performance of Large-Scale CO2 Storage Systems: A Comparison of Different Sensitivity Analysis Methods; NRAP-TRS-III-002-2012; NRAP Technical Report Series; U.S. Department of Energy, National Energy Technology Laboratory: Morgantown, WV, 2012; p 24.

Walsh, S. D. C.; Du Frane, W. L.; Mason, H. E.; Carroll, S. A. Permeability of welbore-cement fractures following degradation by carbonated brines. Rock Mechanics and Rock Engineering 2012, 46, 455–464. DOI: 10.1007/s00603-012-0336-9

Walther, G. R.; Post, E.; Convey, P.; Menzel, A.; Parmesan, C.; Beebee, T.; Fromentin, J. M.; Hoegh-Guldberg, O.; Bairlein, F. Ecological responses to recent climate change. Nature 2002, 416, 389–395.


32

Wang, S.; Jaffe, P. Dissolution of mineral phase in potable aquifers due to CO2 releases from deep formations, effect of dissolution kinetics. Energy Conversion and Management 2004, 45, 2833–2848.

Watson, T. L.; Bachu, S. Society of Petroleum Engineers: Drilling & Completion: Evaluation of the Potential for Gas and CO2 Leakage Along Wellbores; SPE Paper 106817; 2009.

NRAP is an initiative within DOE’s Office of Fossil Energy and is led by the National Energy Technology Laboratory (NETL). It is a multi-national-lab effort that leverages broad technical capabilities across the DOE complex to develop an integrated science base that can be applied to risk assessment for long-term storage of carbon dioxide (CO2). NRAP involves five DOE national laboratories: NETL Regional University Alliance (NETL-RUA), Lawrence Berkeley National Laboratory (LBNL), Lawrence Livermore National Laboratory (LLNL), Los Alamos National Laboratory (LANL), and Pacific Northwest National Laboratory (PNNL). The NETL-RUA is an applied research collaboration that combines NETL’s energy research expertise in the Office of Research and Development (ORD) with the broad capabilities of five nationally recognized, regional universities—Carnegie Mellon University (CMU), The Pennsylvania State University (PSU), University of Pittsburgh (Pitt), Virginia Polytechnic Institute and State University (VT), and West Virginia University (WVU)—and the engineering and construction expertise of an industry partner (URS Corporation).

Technical Leadership Team

Jens Birkholzer LBNL Technical Coordinator Lawrence Berkeley National Laboratory Berkeley, CA

Grant Bromhal NETL Technical Coordinator Lead, Reservoir Performance Working Group Office of Research and Development National Energy Technology Laboratory Morgantown, WV

Chris Brown PNNL Technical Coordinator Pacific Northwest National Laboratory Richmond, WA

Susan Carroll LLNL Technical Coordinator Lawrence Livermore National Laboratory Livermore, CA

Josh White Lead, Induced Seismicity Working Group Lawrence Livermore National Laboratory Livermore, CA

Diana Bacon Lead, Groundwater Protection Working Group Pacific Northwest National Laboratory Richmond, WA

Tom Daley Lead, Strategic Monitoring Working Group Lawrence Berkeley National Laboratory Berkeley, CA

George Guthrie Technical Director, NRAP Office of Research and Development National Energy Technology Laboratory Pittsburgh, PA

Rajesh Pawar LANL Technical Coordinator Lead, Systems/Risk Modeling Working Group Los Alamos National Laboratory Los Alamos, NM

Tom Richard Deputy Technical Director, NRAP The Pennsylvania State University NETL-Regional University Alliance State College, PA

Brian Strazisar Lead, Migration Pathways Working Group Office of Research and Development National Energy Technology Laboratory Pittsburgh, PA

NRAP Technical Report Series

Sean Plasynski Deputy Director Strategic Center for Coal National Energy Technology Laboratory U.S. Department of Energy Jared Ciferno Director Office of Coal and Power R&D National Energy Technology Laboratory U.S. Department of Energy Susan Maley Technology Manager Crosscutting Research National Energy Technology Laboratory U.S. Department of Energy Regis Conrad Director Division of Cross-cutting Research Office of Fossil Energy U.S. Department of Energy

NRAP Executive Committee Cynthia Powell Director Office of Research and Development National Energy Technology Laboratory U.S. Department of Energy Alain Bonneville Laboratory Fellow Pacific Northwest National Laboratory Donald DePaolo Chair, NRAP Executive Committee Associate Laboratory Director Energy and Environmental Sciences Lawrence Berkeley National Laboratory Melissa Fox Program Manager Applied Energy Programs Los Alamos National Laboratory Roger Aines Chief Energy Technologist Lawrence Livermore National Laboratory George Guthrie Technical Director, NRAP Office of Research and Development National Energy Technology Laboratory

NRAP-TRS-III-002-2014 Second-Generation Reduced-Order Model … Library/Research/onsite...

Documents

Transcript of NRAP-TRS-III-002-2014 Second-Generation Reduced-Order Model … Library/Research/onsite...