Underground hydrogen storage: application of geochemical modelling in a case study in the Molasse Basin, Upper Austria

Gas production of this field started in June 2007, and the well was ceased at the end of June 2010. The well was shut in until October 2015. Thereafter, a hydrogen–methane mixture was injected in three subsequent periods. Two of these periods lasted for only a few days because of a shortage of hydrogen at the site. In the third attempt, hydrogen was injected into the formation for 3 months. Figure 1 illustrates the sequence of events for this reservoir.

X-ray diffraction (XRD) data were used to characterize the volumetric mineral compositions of the formation rock. The analyzed core sample was taken from a nearby well and is assumed to have a similar composition as the reservoir under investigation. The mineral composition is as follows: quartz (20 vol%), muscovite and clay minerals (47 vol%), subordinate plagioclase (4 vol%), K-feldspar (2 vol%), calcite (20 vol%), dolomite (9 vol%), ankerite (4 vol%), siderite (2 vol%), and pyrite (1 vol%). In this study, clay minerals were assumed to make up approximately 40 vol% of the total reservoir volume. The volumetric percentage of clay particles is given as follows: illite (59 vol%), smectite (29 vol%), chlorite (15 vol%) and kaolinite (3 vol%) (analysis performed by OMV Exploration & Production GmbH Laboratory, 2007).

The volumetric percentage of minerals was converted to mole values for each mineral to be used in the geochemical models. Using Eq. (1), the amount of each mineral was calculated as:where \({\rho _{\text{w}}}\) is water density, \({m_{\text{w}}}\) is water mass (basis = 1 kg), \(\phi\) is porosity, \({\rho _{{\text{mineral}}}}\) is the mineral density, SVF is the solid volume fraction of minerals, \({M_{{\text{mineral}}}}\) is mineral molar weight, and \({n_{{\text{mineral}}}}\) is the number of moles of each mineral.

The formation water was sampled from the well under investigation and is mainly dominated by K⁺, Cl⁻, HCO₃⁻, and Na⁺, with considerable amounts of SO₄²⁻, Mg²⁺, NO₃⁻, NH₄⁺, Ca²⁺ and some dissolved Fe²⁺ and Mn²⁺ (see Table 1).

The initial gas composition of the studied reservoir [sampled by RAG (Rohöl-Aufsuchungs Aktiengesellschaft)] consists of CH₄ (98.33 mol%), C₂H₆ (0.49 mol%), CO₂ (0.08 mol%) and N₂ (0.84 mol%). The maximum sulfide content (H₂S) is reported to be 5 mg/m³. The initial reservoir pressure is around 100 bars. The amount of CO₂ in the gas phase is incorporated in the geochemical models to account for the degassing effect during water sampling.

Geochemist’s workbench (GWB) (Bethke et al. 2018) was used as a geochemical modelling tool for this study. The software has numerous capabilities, such as the implementation of kinetic rate laws for mineral dissolution and precipitation reactions, complex association and dissociation, redox transformation, gas transfer, 1-D and 2-D reactive transport, and bio-reactive and colloidal transport (Aqueous Solutions 2016, http://​www.​gwb.​com/​).

The internal LLNL thermodynamic database (Lawrence Livermore National Laboratory) was used throughout the study. Consistent and comprehensive thermodynamic data are key to the accurate evaluation of the quality and accuracy of the geochemical model. Database consistency and completeness with respect to the mineralogy of the reservoir were checked. Aside from subordinate plagioclase, potassium feldspar, ankerite, illite, smectite and chlorite, data for the rest of primary minerals are present in this database. Our strategy for modelling was to look up the thermodynamic properties of the absent minerals and integrate them into the LLNL database. Proxy minerals were chosen for minerals for which thermodynamic properties could not be found. The thermodynamic properties of ankerite (Holland and Powell 1998) were integrated directly into the database. Montmorillonite was chosen to represent the smectite group. Albite and anorthite were chosen as proxy minerals for plagioclase. Microcline was selected as an alternative mineral for potassium feldspar. Finally, clinochlore and daphnite were chosen as the two end members of chlorite. Table 2 illustrates the final set of minerals used in the geochemical study.

As part of the UHS feasibility assessment, a multistep methodology is proposed (Fig. 2) to characterize the behaviour of the geochemical system in the presence of hydrogen. As at the time of the study, there were no data available to be benchmarked with the simulation results, several scenarios applying different assumptions were simulated. In the course of modelling relative adjustments and assumptions are incorporated into the models to have results that are more representative of what might happen in the field during a storage cycle. Even though we believe the final kinetic batch model has the highest relevance for the specific study, other scenarios as well as the long-term storage consequences under elevated conditions (longer time scales or elevated temperatures) should not be overlooked.

In this approach, we suggest different modelling steps to study the short- and long-term impacts of hydrogen on the reservoir; proper assumptions were considered for all stages of modelling. The modelling steps are: (1) the equilibrium batch model, which assumes instantaneous reactions (equilibrium) for both hydrogen and minerals; (2) the primary kinetic batch model, which considers mineral reactions to be kinetically controlled, while hydrogen reactions are assumed to occur at local thermodynamic equilibrium; this stage is composed of two modelling steps. In the first model, all the mineral reactions are taken from literature data, and in the second, reactions of pyrite and pyrrhotite are assumed to take place at equilibrium. The reason for doing this relates to experimental evidence indicating that these reactions are relatively fast in the presence of hydrogen, while the literature data do not consider this fact. Further discussions are given in the results section. The last step (3) is the final kinetic batch model, in which both minerals and hydrogen reactions are assumed to be kinetically controlled. A sensitivity analysis on the hydrogen reaction rate (low/moderate/high) and the assumption of having equilibrium/disequilibrium for redox pairs is made to investigate their significance on the results.

The results obtained from the equilibrium batch model assist in estimating the long-term consequences of hydrogen injection in the reservoir. However the model has its limitations to predict sensible results for the cyclic hydrogen storage, therefore, it should only be looked at as part of risk assessment study. From the primary kinetic batch model, we conclude that H₂ interactions with minerals require time scales much larger than a typical hydrogen storage cycle. Consequently, the interactions among hydrogen and brine components are recognised to be more relevant within the storage cycle of hydrogen. Because of the general lack of knowledge concerning the rates of hydrogen dissolution in brine, a sensitivity analysis on H₂ reaction-controlling parameters was performed in the final kinetic batch model. With the kinetic approach, a case was studied in which reactions of pyrite and pyrrhotite are considered at equilibrium. This imitates the reduction of pyrite to pyrrhotite, which can be significant at low-temperature conditions in the presence of hydrogen.

One of the main steps in geochemical modelling is the charge balance calculation to determine the accuracy of the brine analysis. A routine criterion to evaluate this is the charge balance error (CBE) of the reported cation and anion concentrations, as follows (Freeze and Cherry 1979):

The concentration unit in Eq. (2) is milliequivalents per litre. The conversion factor from milligram per litre to milliequivalents per litre is taken from (Zhu and Anderson 2002) and has been used for the charge balance calculation. According to Freeze and Cherry (1979), a charge balance error of less than 5% is acceptable for most laboratories. The result of the analysis is shown in Table 3. The calculated CBE is 17% for the water sample, which indicates that the initial water composition needs to be reassessed. In the course of modelling, the charge balance was compensated by varying the Cl⁻ concentration.

As the charge imbalance error was higher than acceptable, the measured initial water composition needs to be reevaluated and corrected. Generally, prior to geochemical simulations, a proper chemical equilibrium state between brine and primary mineral assemblage states must be established to have a consistent and reliable initial system (Klein et al. 2013; Cantucci et al. 2009; De Lucia et al. 2012). Incorporating all detected minerals in the equilibration step leads to numerical issues and does not represent the water sample taken from the well. Thus, we calculated various scenarios to establish the initial state of the system. These scenarios enable us to consider different starting states for our models to account for incompatibilities in the mineralogy and brine sample (note that the rock sample is from a nearby reservoir). The purpose of establishing an “initial equilibrium state” is to identify mineral phases that control the concentrations of aqueous species, to make hypotheses about the equilibrium between brine and rock mineralogy and to estimate the concentrations of missing species (such as Al³⁺ and Si). Various representations of the mineral composition were equilibrated with the aqueous phase to determine the initial brine compositions. For each case, the calculated brine concentrations were compared to the measured values. Based on the reported natural gas phase in the reservoir, CO₂ with the fugacity of 0.08, was included in the models. With this step, the loss of dissolved CO₂ during the water sampling was taken into account. We further excluded NH₄⁺, NO₂⁻ and NO₃⁻ from the calculations as there is no indication of minerals containing nitrogen. Another reason for eliminating these species is due to not expecting the nitrate reduction by H₂ in the absence of bacteria, or any other specific catalysts (Fanning 2000; Truche et al. 2013a, b). It is worth noticing the co-existence of these redox species will require further discussions and investigations as it has strong association in term of oxygen fugacity in the system, likewise on the redox disequilibrium. Investigation of the latter is beyond the scope of this work. Small initial amounts of Al³⁺ and SiO₂ (aq) were added into the calculations to represent minerals containing these components (e.g., clay minerals). Table 4 presents four scenarios representing different mineral compositions from high to low complexity.

Imposing equilibrium conditions in all scenarios results in inconsistencies with regard to the Cl⁻ and HCO₃⁻ concentrations. In the first scenario, the assumed equilibrium state of both dolomite and calcite resulted in an overestimation of Ca²⁺ and HCO₃⁻ ion concentrations. Similarly, SiO₂ (aq) was overestimated because of the assumption of equilibrium for muscovite and microcline. In the second scenario, calcite was excluded from the equilibrium model, which led to a better match for Ca²⁺ and HCO₃⁻. In the third case study, microcline was eliminated from the equilibrium model, which improved the SiO₂ (aq) concentration. Scenario 4, which assumes equilibrium among the fewest minerals and the analytical brine (incorporated minerals represent entirely analytical brine components), gives the best match, and the Fe²⁺ and Ca²⁺ concentrations improved. Further studies considering hydrogen injection in the equilibrium batch models performed based on the initial states derived from scenarios 3 and 4.

The dissolution rate constants at standard conditions, activation energies and specific surface areas for minerals of interest were taken from literature (Table 5); most of the data were obtained from Palandri and Kharaka (2004). Mineral reactions often depend on pH, requiring different kinetic rates for acidic, neutral and basic reaction mechanisms. Normally, all three mechanisms are not incorporated in geochemical modelling applications, e.g., in applications of CO₂ storage; often, only neutral or acidic mechanisms are employed (Gaus et al. 2008). In the present case, the reservoir conditions are of high pH, as indicated by the initial brine composition and by the results of the equilibrium approach, which shows an increasing tendency of pH with the injection of H₂. For this reason, only the basic reaction mechanisms for minerals have been used in this study. For minerals for which no basic reaction data are published, data for neutral mechanisms have been used. Due to the scarcity of experimental precipitation rates in the literature, the precipitation rates for secondary minerals were set equal to dissolution rates. It is noteworthy to mention that the precipitation rates for some minerals can be slower than the respective dissolution rates. GWB implements the Lasaga type of reaction rate law:

where \(r_{k}^{ \to }\) is the reaction rate (mol/s), \({A_{\text{s}}}\) is the mineral’s surface area (cm²), \({k_+}\) is the rate constant (mol/(cm²s)), and \(Q\) and \(K\) represent the activity product and equilibrium constant for the dissolution reaction, respectively. The surface area is calculated from the specific surface area (cm²/g). The temperature dependence of the reaction rate constant is described by the law of Arrhenius:

Here, \(A~\)is the pre-exponential Arrhenius factor, \({E_{\text{A}}}\) is the activation energy, \(R\) is the gas constant and \({T_{\text{K}}}\) is the absolute temperature in Kelvin. Since we have no information on the effective reactive surface area, we use typical values of 10 cm²/g for non-clay minerals and 100 cm²/g for clay minerals.

Generally, kinetic parameters are obtained in the absence of hydrogen, which makes their precision and applicability in hydrogen storage questionable. The rate constant of pyrite reduction by hydrogen is obtained from the experimental work of Truche et al. (2010). This particular study offers the most relevant and practical values for pyrite dissolution and pyrrhotite precipitation in the presence of hydrogen. The experimental kinetic rate constant for pyrite reduction to pyrrhotite by H₂ was reported in the unit of (mol/m²/h^0.5). This unit needed to be converted to (mol/m²/s). Truche, 2009 reproduced the dependency of the square root of time virtually by increasing the simulated reaction times. In our study, we reproduced this dependency by testing models with pyrite kinetic rate coefficient up to two orders of magnitude higher than the base model. Likewise, we made two case studies for pyrrhotite; in one the rate constant is set to be the same as pyrite and in the other one pyrrhotite reactions are assumed to be at equilibrium. No significant difference was observed in these case studies. In the last part, for pyrite and pyrrhotite, which are the minerals that may most likely react with hydrogen, an extra case study is considered and discussed in the simulation results section.

The prediction of the geochemical reactivity of hydrogen via equilibrium batch modelling is crucial to assess the potential long-term impacts on UHS on one hand, and on the other it allows to test different hypotheses concerning uncertain parametrization with simpler calculations. In these calculations, equilibrium of the aqueous phase with the injected H₂ is maintained; likewise, mineral reactions with the formation water are considered at equilibrium. Two starting points—i.e., initial conditions—were considered corresponding to case studies 3 and 4 from Table 4. The initial state is equilibrated with injected hydrogen with a partial pressure up to 7.5 bars, corresponding to the operational H₂ partial pressure during the injection operation. The reactions are monitored as a function of hydrogen fugacity. The formation of CH₄ (aq) has been suppressed as its formation is not realistic and would lead to misinterpretations. As a further assumption, all other redox couple reactions are treated at equilibrium. As a consequence of hydrogen injection, primary minerals dissolve partially into the formation water, modifying the formation water composition, which leads to the precipitation of other mineral phases. Here, we present outcomes of a model, in which the calcite formation is excluded as it gave more sensible results. The behavior of pH shows first a sharp peak, followed by a smooth increase, as shown in Fig. 3. In case study 3, a slightly higher increase in pH value has been observed. The mineral reactions in the system explain this observation.

A detailed view on the mineral phase, mineral dissolutions and mineral precipitations alongside the variation in pH is provided in Fig. 4. The mineral reactions included in each case study are enumerated in Table 6.

The injection of hydrogen, and consequently the change in pH value, results in the formation of pyrrhotite from pyrite in reservoir conditions (Fig. 5). This reaction is understood to occur in the presence of hydrogen (Truche et al. 2013a, b, b; Betelu et al. 2012). The relationship between pyrite and pyrrhotite can be expressed by the general redox equation (Hall 1986): 2FeS ↔ FeS₂ + Fe²⁺ + 2e⁻.

Changes in aqueous species are expected and could be correlated with mineral dissolution and precipitation reactions. Figure 6 shows the changes in hydrogen sulfide, calcium, potassium, acetate and bicarbonate components as a function of the injected H₂ fugacity. There is a noticeable decrease in a few aqueous species (HCO₃⁻,) with increasing H₂ fugacity because of mineral precipitation. The dissolution of muscovite can explain the increase in K⁺ in the system. The increase in sulfide and calcium concentrations are in agreement with the observed experimental trend in both species in the presence of hydrogen (Truche et al. 2010). The more pronouced production of H₂S at the beginning of the reaction was similarly in agreement with the experimental results (Truche et al. 2010).

The findings from the equilibrium batch models indicate that potential redox couples can play a big role in the consumption of hydrogen and increases in pH. However, the likelihood of their occurrence within hydrogen storage time cycles needs further investigation. Defining the rates of redox reactions in the presence of hydrogen requires laboratory data which are not yet widely existing within the various range of temperatures and hydrogen partial pressures. However, there are valuable experimental data for higher ranges of temperatures in presence of hydrogen, e.g., aqueous sulfate reduction by H₂ at 250–300 °C under 4–16 bars H₂ partial pressure (Truche et al. 2009), nitrate reduction in the presence of H₂ and specific catalyst (stainless steel 316L and Hastelloy C276) at 90–150 °C under 0–10 bars H₂ partial pressure (Truche et al. 2013a, b, b), pyrite reduction into pyrrhotite at temperatures higher than 90 °C and under pressures higher than 10 bars H₂ partial pressure pressures (Truche et al. 2010), and carbonates reduction (Berndt et al. 1996; McCollom and Seewald 2001). Many redox couples are unlikely to reach equilibrium (Sigg 2000). Nordstrom (2002) states that redox disequilibrium is the rule and that many redox species in water will not attain an equilibrium state freely. The main redox couples contributing to the consumption of hydrogen are identified as CH₄–HCO₃⁻, HS⁻–SO₄²⁻, and CH₄–CH₃COO⁻. The CH₄–CH₃COO⁻ redox couple seems irrelevant in the model as a very high formation of acetate is unreasonable (Seewald et al. 2006; Truche et al. 2010). In the presented equilibrium model, this redox pair was decoupled. There are some data available indicating that H₂-induced redox reactions (pyrite reduction and the precipitation of pyrrhotite) can be substantial at low temperatures (Hall 1986; Betelu et al. 2012; Truche et al. 2013a, b). As the kinetic rates for the reactions of acetate/bicarbonate and methane/bicarbonate redox pairs are not reported in the literature, in the kinetic batch models, these redox pairs are decoupled. Assuming an equilibrium for the HS⁻/SO₄²⁻ redox pair may be more relevant for the reaction of pyrite–pyrrhotite; however, previous studies show at the low temperature of the studied system this reaction must be decoupled as well (Kiyosu and Krouse 1993; Cross et al. 2004; Truche et al. 2009). As decoupling of the HS⁻/SO₄²⁻ redox pair requires additional constraints, we have made disequilibrium assumption by adjusting the kinetic redox reaction with a negligible rate constant to this redox pair reaction to allow the pyrite–pyrrhotite reaction. The decoupling reaction among these redox pairs lessens the disturbance of the pH state; likewise, a higher amount of H₂ stayed in the gas phase.

Quantification of the chemical interactions related to a storage cycle requires a kinetic approach. Compared to equilibrium models, accurate kinetic data are scarce and are difficult to acquire, especially for complex systems as in the present case. A hydrogen storage cycle is typically limited to seasons—less than 1 year. Many of the mineral reactions occurring in equilibrium batch models would not occur on those relatively short timescales. Therefore, to understand what is more likely to occur within a hydrogen storage cycle, the integration of mineral reaction rates (listed in Table 5) was essential. Kinetic modelling indicates how fast the system reacts to a perturbation of its geochemical equilibrium state. In this section, we only show the results of the kinetic models based on the initial system of case study 3, which we consider to be the most relevant. In the batch kinetic simulation, a typical cycle of hydrogen injection (we assume 12 months) with hydrogen partial pressure of 7.5 bar was considered. In the primary kinetic batch model, the mineral kinetic reaction rates were integrated in the model, while hydrogen dissolution in brine was assumed to be instantaneous.

Variations in pH and changes of mineral quantities in contact with hydrogen are displayed in Fig. 7. A minor pH increase is observed; likewise, changes in mineral quantities, which proceed extremely slowly, are only notable for primary minerals; none of the secondary minerals were formed in this case. As it was discussed earlier, two models were tested for pyrrhotite reaction rate; in one the rate constant is set to be the same as pyrite reaction rate and in the other reactions of pyrrhotite assumed to be at equilibrium. The results did not show any significant difference.

In the next model calcite was excluded from the secondary minerals and a kinetic rate was assigned to the redox pair of HS⁻/SO₄²⁻. For this model, we implemented a zero^th order equation with rate constants ranging from 5E−9 mol/s (Berta et al. 2018) to 5E−20 mol/s (considering slower rate constant for abiotic reactions). Only for rate constants lower than 1E−12 did our kinetic simulations converge and produce meaningful results. The kinetic rate of 1E−12 leads to the same results as what is already depicted in Fig. 7. Moreover, for the rate constant of 1E−20 and decoupling, no pH increase was observed, still the minerals’ change shows the same change illustrated in Fig. 7. As it has been discussed in the previous chapter, the major uncertainties in the kinetic model results are the assigned reaction rates, which are taken from laboratory experiments. These experiments are generally performed in the absence of a hydrogen partial pressure. Truche et al. (2013a, b) state that abiotic hydrogen redox reactivity is kinetically restricted and that many of the potentially hydrogen-induced redox reactions (e.g., sulfate and carbonate reduction) stay insignificant at low temperatures. However, there could be exceptions: a reduction of pyrite into pyrrhotite is one such possible exception. pH controls the extent of the reaction through alkaline conditions, which may promote pyrrhotite precipitation at lower temperatures and low hydrogen partial pressures. For these reasons, we ran another case study in which reactions of pyrite and pyrrhotite were considered at equilibrium. In this model, primary minerals and pyrrhotite alter the most during 1 year of hydrogen injection (Fig. 8).

Mineral reactions proceed faster than in the previous case, and although the changes are still minor, the pH increase is high and can be problematic (Fig. 9).

Despite the uncertainty in the equilibrium assumption for reactions of pyrite and pyrrhotite, this case study represents more realistic and reliable outcomes as these reactions are likely to occur in the presence of hydrogen. However, experimental data on reaction kinetics in the presence of hydrogen would be desirable and would reduce the uncertainties.

The equilibrium assumption for hydrogen reactions in brine in the primary kinetic batch model results in a large pH increase that seems implausible in short timescales. In the final model, we consider the most realistic scenario, and we used the implemented gas transfer option to account for the hydrogen kinetic reaction rate. The equilibrium assumption for pyrite and pyrrhotite reactions is valid in this model. Literature data on hydrogen gas dissolution kinetics are not available, and we, therefore, defined three categories for kinetic parameters to account for slow, moderate, and fast reactions. Here, the presented terminologies as slow, moderate and high rates (shown in Table 7) serve as rough estimates for the speed of reactions in the presence of hydrogen in a typical storage cycle (12 months) with the operational hydrogen partial pressure of 7.5 bars. These values are determined based on numerous simulations performed to determine the impact of kinetic rate on the pH rise for this particular model. It should be noted that these values are approximations based on simulations in our specific case study and may not be generalised. Varying the kinetic rate-controlling parameters classifies scenarios that can possibly occur in our reservoir. The gas transfer is a rate-expression option that describes the dissolution of gases from an external reservoir. The rate \(r_{k}^{ \to }\) (mol/s) at which a kinetic gas dissolves into the fluid is calculated via the built-in equation:

\({n_{\text{w}}}\) accounts for the solvent mass (kg), \({k_+}\) is the rate constant (mol/cm² sec), \({A_{{\text{sp}}}}\) is the specific contact area (cm²/kg water) between the reservoir and fluid, and \({f_{{\text{ext}}}}\) and \({f_{{k^ \to }}}\) are the external and in-fluid gas fugacity. For convenience, we defined a new variable α (mol/kg water sec), which is the product of \({A_{{\text{sp}}}}\) and \({k_+}\).

The results of the selected case study (α [mol/(kg(water)s)]: 1E−7; injected H₂ fugacity: 7.5 bar; time span: 1 year) are shown below. This case study accounts for quite a high reaction rate of hydrogen. Compared to the primary kinetic model, assigning a kinetic rate for hydrogen solubility controls the increase of pH in the system (Fig. 10).

Minerals abundance change is negligible and at a lower amount compared to the previous case (Fig. 11).

The effect of the H₂ dissolution rate on the pH increase is tested within 1 year of hydrogen injection. Furthermore, the influence of redox pair equilibrium is investigated for the different models (Table 7).

In the case of assuming disequilibrium for redox pairs, when the hydrogen reaction rate remains relatively low, the pH increase is minor, and hydrogen gas acts like an inert gas in the system. This is valid for the case where the equilibrium assumption is considered for redox pairs. In the cases of the moderate and high hydrogen rate constants, a higher increase in pH is observed; however, this amount remains insignificant. It is worth noting that the changes in mineral abundances in all cases are negligible. The equilibrium assumption for redox pairs in these cases triggers hydrogen consumption and results in the considerable pH increase.