(352c) CO­2 Storage Site Selection: A Case Study from Oklahoma, US

We present a site-selection case study for sequestering CO₂ in Oklahoma, US. The selection is made at two levels. First, a state-wide search is conducted hierarchically where we try to avoid areas with major faults and recent seismicity (2013 – 2018) as much as possible while remaining close to emission points and existing CO₂ pipelines (Figure 1). This narrowed our selection to two counties near the Oklahoma City metro area, Kingfisher and Canadian. The two counties are known for their oil and production from STACK play, which comprises a set of stacked reservoirs of various ages and depositional styles. Next, using a 3D seismic survey from Canadian County, we try to gain a deeper understanding of the structures and stratigraphy in the context of the STACK’s CO₂ storage potential. We use data attributes such as impedance and coherence to compare the different formations and realize that although over 3000 ft of Cambrian – Permian Age sediments are available for injection in theory, in practice they all have unique risks. For example, the deeper formations have high continuity but are dominated by faults and fractures while the shallower formations are stratigraphically compartmentalized but largely free of faults. In the ongoing work, we rank the formations by evaluating the pros and cons of injecting into various formations. Various criteria such as formation depth, reservoir pressure, composition, legacy wells, and stratigraphic architecture determine the fate of the injected plume.

In this article, we demonstrate our ranking strategy through two representative sets of dynamic injection simulation examples. The simulator is physics-based and uses realistic relative permeability and capillary pressure curves from experiments on the closest analogs. The simulations account for the variation in relative permeability endpoints and entry pressure, as well as other uncertainties such as the expected changes in pore architecture and the impact of various injection rates on near-wellbore phenomena such as halite precipitation. The first set of simulations is done for the Pennsylvanian section comprising formations of Desmoinesean, Missourian, and Virgilian ages, that are dominantly siliciclastic. In the study area, the Pennsylvanian section drapes over the deeper structures and is overlain by an anhydrite-rich Permian regional seal. Although the Pennsylvanian sections can sustain good fluid flow due to the high porosity (5% – 15%) and high permeability (100s of mD), their complex internal stratigraphic architecture leads to highly uneven plume dispersion. The other set of simulations is done for the Meramec Age formation, a self-sealed and mixed siliciclastic-carbonate “tight” formation with primary storage in the pre-existing fractures. The primary trapping mechanism in the Pennsylvanian is capillary storage while in the Meramec is mineralization.

The 3D seismic volume used in this study is from El Reno, a city to the west of the Oklahoma City metro. It is hereafter referred to by the same name, the El Reno survey. Data were acquired using IVI type vibroseis source that simultaneously performed 4 linear 10 – 100 Hz sweeps each 12s long with a start and end taper of .3s and a peak ground force of 70% of the weight of the vibroseis. In all the surveys, the receiver type used was a SM24 10Hz resonance frequency geophone. Data were acquired with a 2ms sample interval and a bin spacing of 25 m (82.5 feet). The survey area is ~ 125 square miles comprising 1428 inlines (north-south) and 564 crosslines (east-west). The data were processed to zero-phase reverse polarity and imaged using an advanced orthorhombic pre-stack time migration (PSTM) to account for both the vertical transverse anisotropy (VTI) and horizontal transverse anisotropy (HTI). Data have a bandwidth of 15–65 Hz with a 35–40 Hz dominant frequency.

The study is broken into three parts – seismic interpretation, model building, and plume simulation. Seismic interpretation follows the standard practice of well-to-seismic-tie, which allows the interpreter to relate the log tops to seismic wiggles, followed by fault and horizon picking. Although the study area is densely covered by wellbores of various vintages, wells containing sonic and density logs, which are needed for simulating synthetic traces are scarce. This limited our well dataset to two boreholes only, hereafter referred to as A and B. Both are vertical wells that mainly sample the Meramec, which has been the recent focus of the operator’s exploration efforts. Formation depths in both wells were provided by the operator driven by the subsurface knowledge accumulated from decades of exploration in the region. The well-tying is done in the time domain with the understanding that if the depth-to-time conversion of the well log (e.g., using a check shot or sonic log) is accurate, the waveform character of the synthetic trace will align with the corresponding waveform character of the seismic data. When the correlation is not apparent, a match is achieved in an interpretive manner by “stretching” and “squeezing” the individual wavelets in the synthetic seismic trace and/or bulk shifting it which implies changing the interval velocities. Theoretically, such adjustments are justified because of the inherent conversion inconsistencies between stacking and interval velocities. Here, we focused on correlating the wavelet characters corresponding to the tops of the Hunton, Osage, and Chester formations, obtaining an overall match of 79% in Well A and 85% in Well B.

Following the seismic interpretation, a geo-cellular model was constructed. These models, essential for flow simulations, are a collection of various petrophysical properties such as porosity, permeability, water or hydrocarbon saturation, and lithofacies within a regularized grid pattern. Geocellular modeling requires various data types, including geophysical, geologic, petrophysical, and engineering, to be integrated at the reservoir scale. The process is challenging for two reasons – not all the modeling parameters can be determined at all locations, and the size of the simulation grid is often much larger than the size of the samples used for lab measurements. As a result, the flow parameters such as porosity and permeability, must be significantly upscaled depending on the size of the cell chosen for simulation. For example, to achieve the simulation completion within a reasonable time, the cell size is often kept at 10m X 10m X 10m. In comparison, samples used for lab measurements are cm scale. In this case study, we addressed the conundrum of upscaling by populating the cells statistically following a facies-based expected value.

Preliminary results from the numerical experiments show that CO₂ trapped by dissolution, capillary trapping, and mineralization is significantly affected by the various uncertain reservoir properties and decision parameters. The vertical permeability anisotropy significantly impacted the plume migration throughout the reservoir. We observed significant dissolution in the higher permeability region, whereas capillary trapping was dominant in the low permeability zones. As a result, the mobile phase CO₂ disappeared in the reservoir faster in high permeability zones compared to low permeability zones. The faults and fractures resulted in significant channeling of the injected CO₂ within the formation. We observed a substantial reduction in the pH of the aquifer brine from the initial value of around 7. The pH near the well bore rapidly dropped to 4.45, gradually reducing to 4.13 during the active injection period. The pH change was only seen close to the wellbore for the injection layer, and there was no pH change far away from the injection site. However, the pH in the upper layer close to the caprock was significantly depressed beyond the wellbore due to the buoyant CO₂ spreading throughout the formation. This reduction in pH had significant implications for Meramec formation but a limited impact on the Pennsylvanian formation. For each case, the capillary pressure enhanced the mineralization, which can be attributed to the occlusion of ScCO₂ in pores due to capillary forces rather than being displaced as a continuous plume. Furthermore, the rate of dissolution and the transport of dissolved CO2 is governed by the pore-scale configuration of different phases, which is controlled by the capillary pressure. Our simulation results also showed a significant impact of divalent ions (Ca⁺⁺ and Mg⁺⁺)on the mineralization and dissolution of minerals such as calcite and dolomite. We also observed that the highest injection rate led to the least fraction of mobile CO₂ in a reservoir. This is mainly because relative permeability hysteresis and capillary pressure trap a more significant fraction of the injected CO₂ at a high injection rate. We also identify locations on the reservoir that are suitable for safely containing injected CO₂ based on the reservoir porosity and permeability. Finally, we present preliminary results for the loss of injected CO₂ from neighboring leaky abandoned wells. Our results showed that injecting in the sweet spot resulted in a rapid dissolution of CO₂, leading to a reduced probability of diffusion through the existing wells. The result from the study provides critical insights for the proper selection of CO₂ storage sites for secure long-term storage of the injected CO_2, accounting for various reservoir and operation uncertainties.