*4.1. Workflow*

The workflow for parameterizing input for geochemical models described above is summarized in Figure 11. The main challenge with respect to geological characterization of saline aquifers is that hard data are often scarce. Thus, interpolation across large distances and/or burial depths from areas where data are available, to a less mapped, potential injection site are carried out. Absolute reactivity may often not be estimated, but taking into account facies changes (e.g., more clay and feldspar in finer grained facies) and diagenetic imprint (such as albitization with increasing temperature), some scenarios may be defined for initial geochemical bulk modelling including all phases and identification of main reactants. For the Johansen Formation the main cation donors were identified in bulk simulations as albite and Fe-chlorite [61]. Subsequently, estimating the facies-specific reaction potential through geochemical simulations takes a detailed description of the reactants into account. Solid-solutions must be specified and parameterized with suitable kinetic parameters [10,46] and assigned to one or more grain size/shape class occurrences (Figure 5).

**Figure 11.** Workflow for estimating mineralization potentials specific to sedimentary facies and mineral occurrence in reservoir characterization.

The total input reactive surface area (St) may be defined through petrographic studies and quantification of mineral assemblage as described here and/or according to other methodologies (e.g., [22,23,62]). Depending on the magnitude and kinetics of pH-changes during simulations (i.e., not too large fluctuations), simulations for each reactant (dissolving phase) may be run separately.

#### *4.2. Input Data*

XRD quantification procedures provide cheap and frequently available data. Direct use of bulk rock XRD data (sometimes lacking specification of sample type and treatment) as input for geochemical modelling of mineral trapping, with uniform grain sizes and associated reactive surface areas, is common procedure. Evaluating the effect of phase occurrence and grain shape/size may change the time scale and magnitude of mineralization potential significantly, as shown here. Without complementary geological knowledge about the depositional environment and burial history of siliciclastic rocks, reservoir scale estimations of reaction potentials may be grossly wrong. Appreciating the large uncertainty due to natural heterogeneity and lack of data, a range of geochemical models can be constructed to illustrate effects of alternative interpretations and data interpolation. Effects of system variability may also be explored by means of stochastic analysis: for example by Monte Carlo sampling from experimental surface area measurements [16].

In the Johansen Formation the content of reactive minerals varies in well 31/2-3, and until the assembly is confirmed in the actual injection area, facies related scenarios may be defined for sensitivity studies of trapping potential.

## *4.3. Upscaling*

Having defined reactive surface areas and mineral assemblages specific to each facies represented in the reservoir, bulk reservoir reaction potentials (dissolution + mineralization) may be estimated. By fluid flow simulations of CO2 injection and migration (e.g., Eclipse, TOUGH, and similar numerical tools), total dissolved volumes of CO2 within specific layers or in each defined facies setting may be quantified. The current facies model for the Johansen Formation (Figure 12) is based on acoustic impedance data, and must be verified and calibrated with well data before reactivity distributions can be predicted. Scenario based modelling of fluid distributions and dissolution potentials, however, indicate that the dissolution potential (e.g., 29% of 160 Mt injected CO2 after 1000 y) could be less than the mineralization potential (Figure 10) on the same time scales (100's of years), which would cause a catalytic effect of mineralization on further dissolution of the residual phase, where supply of CO2 would be the rate limiting factor. This is highly dependent also on reservoir pressure and temperature conditions, which still are to be measured in the proposed injection area.

Decoupling of models is an advantage in that fluid simulations may include realistic topography, fine grid sizes, proper equations of state (EoS) [63] and relative permeability curves (e.g., [51]) assigned to facies [28], all within the limits of computing capacity (CPU). Considering the residence time of CO2 at the dispersive plume front, as well as the dissolution of residually trapped CO2 left behind a migrating plume, the reaction potential in the injection area and along the predicted migration path may be evaluated. The formation water salinity has negligible effect on the mineralization potential, as the moles of cations in the solid phases are several orders of magnitude larger than in the water at any time, and the aqueous phase can be regarded as merely a transporter of CO2 during the mineralization [18]. Significant mineral precipitation may retard plume migration, increase the dissolution potential, immobilize CO2 and is considered the safest trapping mechanism [2,55,56]. The reaction potential is higher in fine grained facies (because of larger reactive surface areas and higher relative fractions of cation donor minerals), which in combination with higher fractions of residual CO2 would support more efficient immobilization. With a porosity of 25%, the estimated volumetric long term trapping potential would be 55 kg CO2 per m<sup>3</sup> reservoir.

**Figure 12.** Fluid distribution relative to sedimentary facies with different mineralization potentials: (**A**) Interpreted facies distribution related to porosity class and interpreted from seismic attribute analysis (acoustic impedance). Upper shoreface (high porosity and permeability) in dark yellow, lower shoreface (intermediate porosity and permeability) in light yellow. Non-reservoir siltstones and mudstones in green and grey. Model described in [28]; (**B**) fluid distribution at 100 years, after 50 years of injection 3 Mt/y CO2 through a well perforated in the lower half of the reservoir. In this case 16% of CO2 was dissolved, 38% residually trapped [28].

#### *4.4. Hydrogeochemical Trapping*

Mineral and ionic trapping reactions for CO2 are often disregarded in reservoir characterization and modelling schemes, due to slow kinetics. Geochemical studies indicate, however, that the dissolution of clay and silicate minerals and the subsequent precipitation of carbonates may be significant also on time scales less than 100 years (e.g., [10,24]), and approach immobilization potentials in the same size order as dissolution in pore waters [7]. Thus, it may be argued that mineralization potential and the associated increase in dissolution potential should be taken into account in general reservoir characterization schemes.

Immobilization of CO2 enhances storage security [2]. Between two otherwise suitable reservoir candidates (e.g., high injectivity, safe cap-rock, appropriate temperature and pressure conditions), the safer option would be the more reactive reservoir setting (e.g., mineralogy, salinity, pH, temperature) providing permanent storage of CO2 through dissolution-, ionic- and mineral- trapping, disregarding near-well pore-clogging by rapid salt precipitation in this context. Geological heterogeneity may cause plume spreading, increase dissolution and immobilization potential [28,64].

De-coupling of models for transport and reactions, as proposed here, introduce challenges with regards to timing and linked processes such as aqueous speciation of CO2 and pH-changes during silicate dissolution and carbonate precipitation [56]. Furthermore, the dynamics of porosity changes due to mineral precipitation may not be incorporated [65]. However, in coupled geochemical and transport models, reservoir geometries and geological heterogeneities are not accounted for, e.g., [7], which impose the most important control factors with respect to fluid distributions (Figure 12). Fluid flow models may highlight preferential flow paths within distinct facies, bypass zones, and plume separations due to layered heterogeneities. The reactive surface area is expected to be highest in rocks of sedimentary facies with smaller average grain size and higher clay content, as well as increasing with associated lower porosities and smaller pore throats. This relation is valid only down to effective

porosities <6% and associated permeabilities <100 mD [66], at which point it is no longer realistic that all pores are swept with CO2. Absolute estimations of mineralization potential by decoupled methods are not possible due to constrains of present day CPU capacity, but reservoir scale relative evaluations may be made by bulk volume calculations of residual and dissolved CO2 present in a given sedimentological facies setting at different time steps during fluid migration (Figure 12). The implementation of geological models in sensitivity studies may provide insight towards long-term reservoir behaviour.

The result shows that there is no significant difference between batch simulations using the complete reactive mineral assemblage, and the results of running separate simulations for the feldspars and chlorite and summing up the carbonatization potential (Figure 10). This indicates that pH of the simulations are also very similar, as pH strongly affects reaction rates. Such simplifications may be beneficial for running reactive-transport simulations of larger and more complex CO2 storage systems as the addition of kinetic reactions to flow simulations adds a substantial CPU load. All simulations were run at constant CO2 pressure, implying that the batch system is in communication with a CO2 source with sufficient CO2 to feed the necessary five moles required for a complete carbonatization of the feldspar-chlorite assemblage. The amount of CO2 required for a complete carbonatization at time scales when CO2 is still dominantly mobile (<100–1000 years) is about 2.5 to 3.8 moles/L Fmw (Figure 10). With a solubility of 1.28 moles/Kg Fmw in the Johansen Formation (SAFT v1 calculations: [59]), we need about 1.2 to 2.5 additional moles of CO2 per liter formation water for a complete carbonatization. This can be fed from CO2 trapped residually. The minimum volume of residual CO2 and percent residual required per liter pore water was estimated using a CO2 density of 680.13 kg/m3 [63]. The estimated amounts of residually trapped CO2 for 1.2 and 2.5 moles are then 7.2% and 13.9% respectively.

#### *4.5. Kinetic Rate Uncertainties for Chlorite*

Kinetic data used for most CO2 storage simulations are taken from Palandri and Kharaka [46]. There has, however, been generated more recent data and some experiments have also been performed at conditions more relevant to CO2 storage (i.e., relevant CO2 pressures). Because the Palandri and Kharaka [46] review has incorporated data for all feldspars of interest to CO2 storage in sedimentary basins (anorthite, Na-rich plagioclases, albite, K-feldspar), and there are no more recent studies that change the rate constants or temperature dependencies, we have here focused on the variation in data for chlorites, and the few data of Fe-rich chlorites and the total lack of data for the Fe-endmember chlorite. Only rate data from experiments at acidic conditions will be compared here as they are most relevant for CO2 storage. A summary of the compiled chlorite data is given in Table 3.


**Table 3.** Kinetic data for chlorites.

(1) Experimental CO2 molar concentration (denoted with an 'M' after the value) or CO2 pressure. (2) Rate constant k (mol/m2s) at pH <sup>=</sup> 0 and 25 ◦C, assuming a rate equation of the form R <sup>=</sup> kSaH+n(1 − Ω) (see Palandri and Kharaka, 2004 [46]). (3) Rate order with respect to the proton activity.

Average values from Palandri and Kharaka [46] have been estimated from the published data prior to 2004. This compilation suggests a dissolution rate constant of the Mg-endmember chlorite (clinochlore-14A) of 7.76 <sup>×</sup> 10−<sup>12</sup> mol/m<sup>2</sup> s at pH = 0 and 25 ◦C (all rate constants will from here be discussed at this reference point), and with an apparent activation energy of 88.0 kJ/mol. Lowson et al. [50,67] examined the dissolution rates of an Fe-rich chlorite (molar Mg/Fe = 1.4) and found

a rate constant of 1.62 <sup>×</sup> 10−<sup>10</sup> mol/m2s, significantly larger than the average value listed by Palandri and Kharaka [46] for the Mg-endmember, but with a similar and even larger apparent activation energy (94.3 kJ/mol). Smith et al. [48] examined the clinochlore-14A end-member and found a rate constant comparable with [50] (1.23 <sup>×</sup> 10−<sup>10</sup> mol/m2·s), but with a much smaller apparent activation energy (25.1 kJ/mol). Finally, Black and Haese [68] recently found a clinochlore-14A rate constant of 9.55 <sup>×</sup> 10−<sup>13</sup> mol/m2·s, and with an even smaller apparent activation energy than in Smith et al. [48] (16.0 kJ/mol). In all studies, except for Black and Haese [68], a reaction order with respect to protons of about 0.5 has been found. However, the low value of 0.076 found by Black and Haese [68] was attributed to the inhibiting effect of bicarbonate on the dissolution rate, largely cancelling out the catalyzing effect of protons. It is clear that it is a large range in listed rate constants and apparent activation energies, and the work by Black and Haese [68] also suggest that CO2 and bicarbonate will significantly affect the pH dependency of the rates. It is therefore of interest to compare chlorite dissolution rates at conditions relevant for CO2 storage. The pH of a CO2 storage repository buffered by calcite dissolution is around 5 and quite independent of CO2 pressure and temperature [10,17].

The temperature varies from reservoir to reservoir. At 37 ◦C (e.g., the Utsira CO2 storage site), differences are only modest for the Mg-endmember, with the largest rate constants from Smith et al. [48] being approximately seven times larger than the lowest from Palandri and Kharaka [46]. The Fe-rich chlorite is at these conditions suggested to react approximately six times faster than the average found for the Mg-end-member (Table 3). At 75 ◦C the rate constant from Palandri and Kharaka [46], having much higher apparent activation energy, passes the value for two other studies on the Mg-chlorite. At this condition, the Palandri and Kharaka [46] rate constant is approximately 16 times that of the Black and Haese [68]. The rate constant for the Fe-chlorite, having even larger activation energy, is at 75 ◦C almost 100 times larger than for the average of the Mg-chlorites. This fast reactivity of the Fe-end-member fits well with studies of other Fe-rich clay minerals, such as glauconite [69], and poses a challenge in predicting the short term (<100 years) reactivity and mineral trapping potential of the Fe-chlorites, being very common in North Sea reservoirs [34,35,44,61]. The large variations in data for Mg-chlorite and the lack of data for the Fe-endmember result in a significant uncertainty in estimated dissolution rates, on top of the large uncertainties in reactive surface areas. This calls for more rate studies on chlorite dissolution, preferentially done at CO2 storage conditions (including realistic CO2 pressures).

## **5. Conclusions**

The reactive surface area depends on grain size and shape, porosity and permeability, and varies according to sedimentary facies and diagenetic imprint. More accurate, or relevant ranges, of input values for reactive specific mineral surface areas as used in geochemical modelling of long term mineralization potential for CO2 may be estimated by combining optical, physical and chemical quantification methods, and relate mineral morphology to grain size in estimations from weight %. Implementing sedimentary facies variations in reservoir models provides for volume estimations of fluid distributions in various parts of the reservoir, which may be applied for evaluating spatial variability of mineralization potentials in CO2 storage reservoirs.

Na-plagioclase and Fe-chlorite are the main cation donors for mineral trapping of CO2 in the Johansen Formation. Reaction rates of chamosite in reservoirs ~100 ◦C are likely significant on shorter time scales (100's of years), and relevant for estimation of immobilization potential and increased dissolution. The bulk reactive mineral content (feldspar and chlorite) as well as reactive surface area per weight fraction is higher in fine-grained facies. Simulations suggest that chlorites in ooids or dense aggregates may reduce the short term (<100 years) mineral trapping potential by up to 20%, compared to more reactive occurrences like grain coats. Feldspars are suggested to have the largest impact on long-term (1000–10,000 years) mineral trapping. Intact lithic fragments are less reactive, while diagenetically altered grains may be more reactive. In our geometric model, fine-grained facies have four times larger specific reactive surface areas compared to medium-grained sand, and the mineral trapping rates are correspondingly faster.

**Author Contributions:** A.S. and H.H. wrote, edited and reviewed the manuscript, and developed the concept, idea and methodology; H.H. performed the numerical simulations; A.S. made input for the models; A.S. handled project administration and secured funding.

**Funding:** This work has been partly funded by the Research Council of Norway in projects CO2 Upslope under grant # 268512 and SUCCESS under grant # 193825/S60. SUCCESS (SUbsurface CO2 storage—Critical Elements and Superior Strategy) is a consortium with partners from industry and science, hosted by Christian Michelsen Research, and is an Environment-friendly Energy Research (FME)-center assigned by the Research Council of Norway. The CO2-Upslope project: Optimized CO2 storage in sloping aquifers, is funded by the CLIMIT-programme. Gassnova SF and the Research Council of Norway collaborate on the CLIMIT-programme which finances projects within Carbon Capture and Storage (CCS).

**Acknowledgments:** The authors would like to thank CLIMIT and the Norwegian Research Council for funding. We acknowledge Equinor and the Norwegian Petroleum Directorate for access to sample the Johansen core, with the mineralogy presented in previously published works forming the basis for these simulations. The authors thank Equinor for sharing QemScan data from thin sections, and we appreciate discussions with C. Kruber and R. Meneguolo. We are also grateful to R. Miri, H. Dypvik, and P. Aagaard at UiO for fruitful discussions. Finally, the authors thank the two anonymous reviewers for their insightful comments.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.
