Use of above-zone pressure data to locate and quantify leaks during carbon storage operations
Introduction
A major concern associated with carbon capture and storage (CCS) in saline aquifers is the potential for leakage. While thorough pre-injection screening procedures have been developed to avoid caprock-leakage; it is important that, in the event there is a leak, we are able to quickly detect, locate and quantify it so that risk-assessment and remediation efforts can be initiated. In this work, we investigate the efficacy of history-matching above-zone pressure monitoring data in order to locate and quantify such leaks.
Much research has been dedicated to understanding the physical processes of leakage. Information and examples of flow through faults in CCS related operations are provided in Lewicki et al. (2007). The probability of a plume encountering a fault was investigated by Jordan et al. (2011) using a fault density distribution model. For their case study in the San Joaquin Basin, they found that a previously planned injection site had a 3% chance of encountering a large enough fault to cause significant leakage. Leakage through abandoned wells has also been studied by numerous authors, for example, Nordbotten et al. (2005) developed a semi-analytical model for flow through leaky wellbores. The likelihood of a plume encountering an abandoned well was investigated by Gasda et al. (2004), based on the distribution of wells in a well-studied area of the Alberta Basin in Canada.
There are several monitoring options for leakage detection in CCS operations, as detailed in Plasynski et al. (2011). Methods relying on surface detection of CO2 have been proposed by, e.g., Lewicki et al. (2007) and Fessenden et al. (2010). While these methods may be effective at determining when leakage has occurred, leakage should ideally be identified before it has reached the surface. Early detection monitoring techniques involving well data, seismic data, tracers, and satellite interferometry have been applied at both the In Salah and Sleipner storage projects. At the In Salah site, a combination of surface uplift data and measured CO2 breakthrough at a monitoring well confirmed that leakage had occurred. This leakage was largely attributed to one of the three CO2 injectors (Ringrose et al., 2009), which was subsequently shut in.
Because pressure responses in the subsurface propagate quickly, pressure data from monitoring wells may be used to detect, and possibly locate, leaks before CO2 has even reached them. Pressure transient data has been used to monitor and infer the migration of CO2 within the storage reservoir using multilevel pressure sampling at the Illinois Basin Decatur Project (Strandli et al., 2014). Using an approach developed by Strandli and Benson (2013), it was demonstrated that CO2 was trapped beneath a thin sealing layer far below the base of the regional seal created by the Eau Claire shale formation. The application of these data showed that multilevel pressure sampling, even within the storage reservoir itself, provides valuable information, which could potentially aid in detecting CO2 migration and predicting leakage.
The detectability of leaks using above-zone pressure-transient data was analyzed by Chabora and Benson (2009). They found that pressure transient data can indicate CO2 breakthrough in some cases, though their tests were performed on simplified aquifer models with known geology. The work of Sun and Nicot (2012) confirmed that pressure anomalies indicating leakage are detectable even when measurement error and spatial heterogeneity are considered. Sun et al. (2013b) used a probabilistic collocation method to determine when pressure data might be used to detect leakage. Their method considers the signal-to-noise ratio of pressure anomaly data and provides an effective means for detecting when a leak exists. A methodology to determine optimal monitoring well locations to minimize both the chance, and consequences, of not detecting a leak was presented in Sun et al. (2013a). They assumed a minimum pressure signal for leak detection and did not consider whether monitoring networks could predict the locations or leakage volumes of leaks via data assimilation.
A method to determine both the location and transmissibility of a leak from above-zone pressure monitoring data, collected over as little as a 50 day period, was initially proposed by Javandel et al. (1988). Their method assumes that the aquifer pressure can be characterized using the Theis equation (Theis, 1935), which is an analytical approximation of the pressure response from single-phase injection into an infinitely large, horizontal, homogeneous aquifer. An analysis of the stability and uniqueness of the inversion, using the Theis equation for pressure propagation, was carried out by Zeidouni and Pooladi-Darvish (2012a). The extension of that work (Zeidouni and Pooladi-Darvish, 2012b) used the inversion method to evaluate aspects that might improve the efficacy of monitoring. It was found that both pulsing injection and including additional monitoring wells reduces error in predicting the location of a leak. While the Theis equation has since been extended by Zeidouni (2014) to provide vertical pressure data, its application is restricted to homogeneous, horizontal aquifers with a single point of leakage. In more complicated (realistic) scenarios, simulation is required to forward-model the propagation of pressure. As a result, the data assimilation process is more difficult.
Data assimilation (also known as inversion, history-matching or model updating) refers to the determination of one or more geological models that match observed dynamic data to within some tolerance. Several studies have focused on data assimilation to improve the conceptual model of the storage formation and, sometimes, the overlying (above-zone) aquifer. Some of these used data collected from actual storage operations, while others used synthetic data to test methodologies. Many of the history-matching studies to date (e.g., Doughty et al., 2008, Ennis-King et al., 2011) considered simplified facies and ‘layer-cake’ models and included only a small number of geological parameters, which were generally matched by hand. In the case of Doughty et al. (2008), cross-well seismic and vertical seismic profile (VSP) data were collected (see also Daley et al., 2008) from the Frio formation near Houston, Texas. The forward model for history-matching involved both flow and seismic modeling, with seismic signals providing the observation data to be matched. In the case of Ennis-King et al. (2011), bottomhole pressure data from the depleted Naylor gas field in Southeastern Australia were used for history-matching.
In order to facilitate efficient data assimilation in an uncertain heterogeneous geologic setting with thousands of geologic variables, parameterization procedures can be applied to reduce the number of variables. One such technique, the Karhunen–Loève (K–L) expansion (Loeve, 1977), has been used in oil field applications (e.g., Oliver, 1996, Reynolds et al., 1996, Sarma et al., 2006), and more recently in the context of carbon storage (e.g., Cameron and Durlofsky, 2014). K–L expansion, which is analogous to principal component analysis (PCA), is most applicable in geologic settings with relatively smooth property variations, where the model parameters (e.g., porosity and permeability) are correlated according to their relative locations. Such a setting corresponds to a Gaussian or ‘two-point’ geostatistical model. A major benefit of the K–L expansion is that random samples of the K–L variables produce geologic realizations that honor the two-point geostatistics as defined by prior models (this is also the case for history-matched models generated via the K–L variables). By prior models, we mean realizations that honor the specified spatial statistics and any available ‘hard’ data, but not dynamic flow data (which is why history-matching is necessary). Extensions of the K–L parameterization include the kernel principal component analysis (KPCA, see Sarma et al., 2008) and optimization-based principal component analysis (O-PCA, see Vo and Durlofsky, 2014) methods. These treatments enable the representation of geologically-realistic systems characterized by multipoint (rather than two-point) spatial statistics. Examples of such systems include channelized or deltaic-fan geologies.
The use of data assimilation to locate a leak from simulated above-zone pressure data was addressed, in some aspects, by Jung et al. (2015). There, the accuracy in predicting the leak location was tested against several uncertain model parameters, including the size and permeability of the leak, the thickness and permeability of the caprock, and other storage formation and above-zone properties. They did not, however, test the effect of using more than two monitoring wells, nor did they consider different leak locations or heterogeneous geologies. A major conclusion from Jung et al. (2015) is that leak location estimates based on concurrent above-zone and storage formation pressure data are more accurate than estimates using only above-zone data, which are in turn much more accurate than estimates using only the storage-formation data. This claim is supported by Cameron (2013), though that work suggested that the modest benefit from including storage formation data may not be worth the capital expenditure (and risk of damaging the caprock) needed to acquire it.
In Cameron (2013), the methods of K–L expansion and particle swarm optimization (Kennedy and Eberhart, 1995) were shown to be effective in performing carbon sequestration data assimilation in problems with uncertain heterogeneous geology. It is the goal of this work is to expand upon the work of Cameron (2013), Sun et al. (2013a) and Jung et al. (2015) to provide an assessment for the predictability of leak locations and leakage volumes, given different numbers of above-zone monitoring wells, over different monitoring time-periods, in an uncertain heterogeneous environment.
This paper proceeds as follows. In the methodology section, we first describe the leaky aquifer model used in this study, then provide relevant theory on data assimilation and the K–L expansion, and finally present our formal data assimilation problem statement. In the results and discussion section, we assess the impact of (1) temporal data quantity, (2) the number and location of monitoring wells, (3) the use of multilevel versus single-level monitoring wells, and (4) the use of higher-resolution leakage models. We conclude with a summary of our major findings and suggestions for future work.
Section snippets
Methodology
To test our methodology, we consider a three-dimensional, heterogeneous, synthetic aquifer model, with injection into a storage formation, and leakage through an unknown location in the caprock. The model has three regions: a storage aquifer, a caprock and an (above-zone) overlying aquifer. Many model realizations, or ‘leakage scenarios’, are generated and simulated to quantify both short-term pressure propagation (for monitoring purposes), as well as long-term CO2 propagation and fluid leakage
Results and discussion
In this section five different ‘true’ models are considered in order to evaluate the efficacy of data assimilation using varying amounts of spatial and temporal monitoring data. By efficacy, we refer to the accuracy in predicting leak location, leakage fluid quantity during the injection stage, and long-term (500-year) CO2 leakage quantity. We first present data assimilation results where the amount of temporal monitoring data is varied and then present results where the number of monitoring
Concluding remarks
This work addressed a key issue for carbon storage project managers, that is, deciding how many pressure monitoring wells are required and how much time it will take to effectively locate and quantify leaks. We have performed over 100 data assimilations involving five different ‘true’ models in a variety of monitoring scenarios. The Karhunen–Loève expansion was used to reduce the optimization variable space, and particle swarm optimization was applied to minimize a pressure misfit function. We
Acknowledgments
We are grateful to the sponsors of the Stanford Smart Fields Consortium, the Stanford Center for Carbon Storage (SCCS), and the Global Climate and Energy Project (GCEP) for partial funding of this work. Special thanks also to Tapan Mukerji for useful discussions and suggestions.
References (50)
- et al.
Optimization of well placement, CO2 injection rates, and brine cycling for geological carbon sequestration
Int. J. Greenh. Gas Control
(2012) - et al.
Brine displacement and leakage detection using pressure measurements in aquifers overlying CO2 storage reservoirs
Energy Procedia
(2009) - et al.
The role of heterogeneity in CO2 storage in a depleted gas field: history matching of simulation models to field data for the CO2CRC Otway Project, Australia
Energy Procedia
(2011) - et al.
The effect of noise and Tikhonov's regularization in inverse problems. Part II: The nonlinear case
J. Appl. Geophys.
(2014) - et al.
The effect of noise and Tikhonov's regularization in inverse problems. Part I: The linear case
J. Appl. Geophys.
(2014) - et al.
Monitoring a large volume CO2 injection: year two results from SECARB project at Denbury's Cranfield, Mississippi, USA
Energy Procedia
(2011) - et al.
On the detection of leakage pathways in geological CO2 storage systems using pressure monitoring data: impact of model parameter uncertainties
Adv. Water Resour.
(2015) - et al.
Influence of small-scale heterogeneity on upward CO2 plume migration in storage aquifers
Adv. Water Resour.
(2015) - et al.
Upscaling of CO2 injection into brine with capillary heterogeneity effects
J. Pet. Sci. Eng.
(2015) - et al.
CO2–H2O mixtures in the geological sequestration of CO2. II. Partitioning in chloride brines at 12–100 °C and up to 600 bar
Geochim. Cosmochim. Acta
(2005)
CO2 plume tracking and history matching using multilevel pressure monitoring at the Illinois basin-Decatur project
Energy Procedia
Inversion of pressure anomaly data for detecting leakage at geologic carbon sequestration sites
Adv. Water Resour.
Optimal design of pressure-based, leakage detection monitoring networks for geologic carbon sequestration repositories
Int. J. Greenh. Gas Control
Assessing leakage detectability at geologic CO2 sequestration sites using the probabilistic collocation method
Adv. Water Resour.
Modeling above-zone measurements of pressure and temperature for monitoring CCS sites
Int. J. Greenh. Gas Control
Analytical model of well leakage pressure perturbations in a closed aquifer system
Adv. Water Res.
Leakage characterization through above-zone pressure monitoring: 1 – Inversion approach
J. Pet. Sci. Eng.
Leakage characterization through above-zone pressure monitoring: 2 – Design considerations with application to CO2 storage in saline aquifers
J. Pet. Sci. Eng.
Optimization and Monitoring of Geological Carbon Storage Operations (Ph.D. thesis)
Optimization and data assimilation for geological carbon storage
Time-lapse crosswell seismic and VSP monitoring of injected CO2 in a brine aquifer
Environ. Geol.
Site characterization for CO2 geologic storage and vice versa: the Frio brine pilot, Texas, USA as a case study
Environ. Geol.
CO2 leakage through an abandoned well: problem-oriented benchmarks
Comput. Geosci.
Chemical reaction modeling in a compositional reservoir-simulation framework. SPE 163677
Novel MVA tools to track CO2 seepage, tested at the ZERT controlled release site in Bozeman, MT
Environ. Earth Sci.
Cited by (22)
History matching for geological carbon storage using data-space inversion with spatio-temporal data parameterization
2024, International Journal of Greenhouse Gas ControlDeep-learning-based coupled flow-geomechanics surrogate model for CO<inf>2</inf> sequestration
2022, International Journal of Greenhouse Gas ControlCitation Excerpt :Uncertainty in these essential quantities can be reduced through use of surveillance/monitoring data in conjunction with data assimilation procedures. Several studies (Cameron et al., 2016; González-Nicolás et al., 2015; Jung et al., 2015), for example, have used pressure monitoring data for leak detection. Chen et al. (2020) used pressure and CO2 saturation data at monitoring wells to reduce uncertainty in plume location.
Bayesian well-test 2D tomography inversion for CO<inf>2</inf> plume detection
2020, International Journal of Greenhouse Gas ControlCitation Excerpt :For a large scale CO2 injection, such detection would focus on an overlying aquifer unit, where a leak (if it occurred) would create a pressure signal. Solutions have been given in the literature for the pressure changes that would be produced by CO2 leakage through wells (Nordbotten et al., 2005) or faults (Kang et al., 2014; Shan et al., 1995; Zahasky, 2014; Zeidouni and Pooladi-Darvish, 2012; Cameron et al., 2016). Indeed, each possible kind of leakage path should have a characteristic signature in the pressure (Cihan et al., 2013), from which it might be possible to identify the mechanism.
Data-space approaches for uncertainty quantification of CO<inf>2</inf> plume location in geological carbon storage
2019, Advances in Water ResourcesCitation Excerpt :Although the prediction of posterior well-based quantities for such systems has been accomplished in a DSI setting in Sun and Durlofsky (2017), it will be interesting to assess the predictive ability of DSI for QoI of the type considered here (such as top-layer CO2 saturation). The use of DSI for different problem setups, such as CO2 leakage detection under uncertainty as described by Cameron et al. (2016) and Jeong et al. (2018), should also be considered. In these applications, leakage may have already occurred, which will change the formulation (objective) of the optimization problem.
Temperature analysis for early detection and rate estimation of CO<inf>2</inf> wellbore leakage
2017, International Journal of Greenhouse Gas Control