Successful geologic sequestration of carbon in deep saline aquifers requires accurate predictive models of rock-brine-CO2 interaction. Often overlooked in siliciclastic-hosted saline reservoirs is the carbonate buffering of the groundwater. Carbonate minerals are ubiquitous, even in siliciclastic host rocks, resulting in some carbonate buffering. Geochemical modeling of rock-brine-CO2 systems often do not accurately predict the geochemical evolution of the system leading to significant doubts in predicting the performance of carbon repositories. New data from a simple NaCl brine-plagioclase hydrothermal experiment tests carbon sequestration in dawsonite and sensitivity to carbonate buffering. This is contrasted to a NaCl brine - siliciclastic rock system containing some initial bicarbonate buffering, analogous to most saline-aquifer sequestration targets, and show that critical errors are caused by incomplete or inaccurate characterization of the in situ geochemistry. We provide a methodology that accurately predicts the in situ condition using samples collected from brine-rock-CO2 experiments or well-heads in a carbon sequestration monitoring scenario.
 Once supercritical CO2 is injected into a brine aquifer, what chemical reactions will occur? Will these chemical reactions sequester carbon dioxide either as a dissolved phase or into a neo-formed mineral? And, importantly, can we accurately predict and monitor these processes in a field application? A variety of geochemical processes are implicit in these questions, including dissolution of supercritical CO2, formation and dissociation of carbonic acid, and dissolution and re-precipitation of minerals in acidic brine. In order to characterize these processes in the field so that the system can be correctly modeled and monitored, and its performance verified, an accurate characterization of the in situ aqueous geochemistry is required. In situ measurement of geochemistry in CO2 saturated aqueous systems requires the use of sophisticated devices such as U-tube [Freifeld et al., 2005], Leutert™ or other similar down-hole samplers [e.g., Goff et al., 1987]. However, routine carbon repository monitoring will require more practical, simpler and less expensive methods that utilize samples collected at the well-head. Thus, computer models are needed to recreate the in situ conditions based on surface measurements of fluids that have degassed CO2. Reconstructing the in situ geochemistry with models based on surface measurements is problematic since the critical parameters are often incorrectly measured or not obtained.
 In CO2-bearing systems, incomplete characterization of the groundwater requires assumptions about the geochemical system that can lead to incorrect evaluation of rock-water interactions. In particular, without proper characterization of the bicarbonate buffer in saline groundwater, geochemical predictions of rock-water interactions can be woefully inaccurate. Speciation, pH and mineral saturation state predictions, for example, are usually very sensitive to the natural bicarbonate buffer in the water. A 2% error in the bicarbonate content of brines could produce a 1 pH unit error when estimating in situ pH, greatly affecting the prediction of mineral saturation indices of carbonate minerals, which is critical for evaluating carbon sequestration in the mineral phase.
 Studies directed at understanding the potential of geologic basins for CO2 storage and the subsequent geochemical behavior of the mixed fluid (CO2 − H2O)-rock system fall into a variety of classes. These include field [Moore et al., 2005], natural analog [Bachu and Bennion, 2007; Carey et al., 2007], modeling [Gunter et al., 2000; Andre et al., 2007], experimental [Druckenmiller and Maroto-Valer, 2005; Lin et al., 2007], and coupled experimental and modeling studies [Shiraki and Dunn, 2000; Kaszuba et al., 2005; Rosenbauer et al., 2005; Suto et al., 2007]. Although these studies advance our understanding, the aqueous geochemistry is incompletely characterized in each. Insufficient data reporting (e.g., incomplete chemistry), inappropriate calculation of in situ pH, and incomplete quantification of CO2 and dissolved inorganic carbon in experiments yield model predictions of mineral precipitation and/or dissolution in conflict with experimental results. For example, Kaszuba et al.  observed neo-formed carbonates in run products from supercritical CO2-brine-rock experiments, but geochemical modeling predicted carbonates to be undersaturated. Suto et al.  reported the opposite phenomenon, predicting carbonate mineral precipitation but not crystallizing carbonates experimentally. Clearly, we must be able to accurately model well-constrained laboratory experiments before we can confidently predict the sequestration of CO2 in poorly-constrained natural systems.
 Here we present results from a new hydrothermal experiment that emulates a carbon sequestration scenario by injecting supercritical CO2 into NaCl brine hosted in a mineralogically simple (mono-mineralic) siliciclastic aquifer. We use an off-the-shelf geochemical code of a type that will be routinely used for carbon repository monitoring to calculate in situ conditions and predict mineralization and/or dissolution. We also revisit published experimental results of sequestration in a mineralogically complex (realistic) siliciclastic saline aquifer and recalibrate model predictions and demonstrate the errors stemming from assumptions based on incomplete geochemical analysis. From these two examples we 1) demonstrate the significance of small amounts of carbonate buffer in siliciclastic saline aquifers and 2) identify the key parameters that need to be measured in the fluid chemistry in order to accurately model these systems and develop protocols for reliable Monitoring, Measuring and Verification (MMV) of geologic carbon sequestration.
2. Carbon Sequestration in a Saline Aquifer and the Importance of Carbonate Buffer
 The brine-rock-CO2 experiment initially contained NaCl brine, albite and supercritical CO2. No carbonate minerals were present at the start of the experiment and no divalent metal cations that could lead to carbonate mineral precipitation were present in solution (Table 1). Thus, in addition to evaluating carbonate buffers in a model siliciclastic rock, this experiment tests the reactivity of plagioclase in brine and supercritical CO2 and the potential for sequestration of carbon in the mineral dawsonite. Dawsonite, NaAl(CO3)(OH)2, is considered a promising phase for the long-term mineral sequestration of CO2 in sedimentary basins [Hellevang et al., 2005; Benezeth et al., 2007]. Using established methods [Seyfried et al., 1987; Kaszuba et al., 2005], 1 M NaCl brine and pure albite were reacted for 123 days (2960 hours) at reservoir conditions (75°C, 20 MPa) to approach steady state. Supercritical CO2 was then injected into the system and the experiment continued for another 45 days (1079 h). Brine was periodically sampled from the ongoing reaction, and solids and quenched brine were analyzed at the conclusion of the experiment (Table 1). Dawsonite was not observed in the post-reaction solids based on SEM imaging and X-ray diffraction analysis of post-reaction solids, while minor illite alteration was observed (Figure 1). Geochemical speciation modeling using The Geochemist's Workbench (GWB) [Bethke, 2006] predicted clay saturation prior to CO2 injection, and under-saturation for all minerals (carbonates and silicates) post injection, accurately matching the experimental results. This mono-mineralic example suggests that 1) in the absence of carbonate buffers, brine pH and carbonate mineral saturation evolve independently of fluid-mineral reaction and solely as a function of injected CO2 and 2) mineral sequestration in dawsonite is unlikely. Below we contrast these results to an example from the literature that used a more realistic basin mineralogy that shows considerably different behavior based on our recalibrated model results.
Table 1. Albite + NaCl Brine + Supercritical CO2, 75°C and 200 Bara
In situ pH calculated using The Geochemist Workbench by reacting degassed fluid chemistry from 25 to 75°C and then titrating in the CO2 total less the dissolved HCO3− (∼17 ppm).
CO2 injected after sample −04; HCO3− not determined on post injection samples - only total system CO2; the pre-inject value of 17 ppm HCO3− is assumed; The water rock ratio at the start of the experiment was 118:1. We injected 15.4 grams of CO2 with water: rock ratio at 77:1 after sample −04.
 A critical error propagated in the literature that leads to disparity between modeling and experimental results is not completely defining pH, dissolved inorganic carbon, and/or quantity of supercritical CO2 injected into the geochemical system. The interdependence of these parameters and the importance for accurately determining each is evident by considering the equations governing aqueous carbonate speciation and pH control. Dissolved CO2 in fluid systems exists as H2CO3, HCO3− and CO32−, and the activity of these species in solution and the pH of that solution are a function of one another. The speciation of dissolved inorganic carbon can be described by the well known reactions and associated equilibrium expression (equations (1)–(3)):
Thus, the total dissolved inorganic carbon in solution can be expressed as:
By substituting between equilibrium expressions, an equation that describes the pH of a solution is:
where pH = −log(aH+) and pK = −log(K). Built into this equation is the impact of temperature (pK varies with temperature) and the buffering affect of bicarbonate on solution pH.
 These parameters are extremely difficult to measure within supercritical CO2-brine-rock laboratory experiments that emulate sedimentary basin pressures and temperatures. This problem is analogous to the difficulty of collecting in situ fluids from monitoring wells. Fluid samples withdrawn from experiments cool and degas CO2, processes which drastically alter pH. For example, Kaszuba et al.  report degassed brine chemistry, including measured pH, major ion chemistry and total in situ CO2. The dissolved carbonate species (HCO3− and CO32−) were not measured, but CO32− can be ignored because at pH < 9 it exists in insignificant amounts [Drever, 1997]. The authors used these data with GWB to add CO2 into a bicarbonate free solution and calculate in situ pH and mineral saturation indices (SI) (pH ≈ 4.4 and carbonate mineral SI ≪ 0, see Figure 2, box A). Their omission of the initial bicarbonate concentration in the modeling resulted in predictions that did not agree with experimental products. Figure 2 illustrates the sensitivity of carbonate mineral SI and fluid pH to the initial quantity of bicarbonate in solution. We revise the Kaszuba et al. in situ chemistry calculations and mineral saturation index predictions by estimating the amount of bicarbonate buffering in solution two ways. First we assume the bicarbonate in solution is the pre-CO2 injection total CO2 (∼5 mmolal) as determined from sampling and analysis of experimental fluids. This estimate is a minimum because the authors did not measure dissolved carbonate species after CO2 injection, and the solution may have acquired additional buffering due to subsequent mineral-fluid interaction. Forward modeling of their rock-brine-CO2 experiment using GWB was used to estimate this additional buffering and suggests that ∼8 mmol of dissolved HCO3− should be in solution and accounted for during mineral saturation index calculation. Forward modeling was conducted by first reacting (equilibrating) the starting mineralogy and brine chemistry reported by Kaszuba et al.  at 200°C, followed by titrating in the amount of CO2 injected in their experiment. This estimate is considered a maximum since the geochemical model assumes equilibrium is reached. Both the lower and upper estimates are adequate to increase in situ pH by approximately 0.5 pH units, pushing calcite and siderite to near equilibrium (SI = 0) and dolomite and magnesite to supersaturation (SI > 0) (Figure 2, box B). These revised estimates are consistent with the precipitation of early magnesite and later reaction to siderite reported by Kaszuba et al. . While this concentration of bicarbonate is small compared to the total system CO2 (2–3% of the total), the disconnect between modeling predictions (no carbonate precipitation) and experimental results (neo-formed carbonate) hinges on this subtle difference. The presence of a small amount of bicarbonate buffer must be explicitly accounted for in computational models.
 Again, our mono-mineralic experiment shows that dawsonite is undersaturated at the experimental conditions (Figure 2). The initial dissolved carbonate in this experiment was quite small, controlled by solution equilibration with atmospheric CO2 prior to sealing the experiment. However, Figure 2 indicates that the solubility of dawsonite is sensitive to the initial bicarbonate content, and that only ∼5 mmol initial bicarbonate is needed to push dawsonite to saturated conditions. The difference between our experiment and that of Kaszuba et al.  is that our experiment is a very simple system of one pure silicate (albite) containing a mono-valent metal cation and the latter replicated a realistic siliciclastic host rock containing a complex mineralogy that provided divalent metal cations and small but significant bicarbonate buffering during rock-water-CO2 reaction. If our experiment had contained a more realistic starting mineralogy including other silicate minerals and/or a small amount of carbonate mineral, the solution would have acquired some bicarbonate buffering and achieved conditions favorable for dawsonite precipitation. Most siliciclastic sedimentary formations are not mono-mineralic and small amounts of carbonate minerals, as detrital grains and as secondary cements, are common, and based on this study, are important for initial carbonate buffering, subsequent rock-brine interaction and mineralization after CO2 injection.
3. Implications for MMV of Carbon Sequestration
 Accurate and complete characterization of the aqueous chemistry in siliciclastic saline reservoirs targeted and utilized for geologic sequestration of CO2 is critical for reliable predictions and measurements of performance. A commonly overlooked problem in saline reservoirs is the sensitivity in geochemical model predictions of pH evolution, mineralization and dissolution to carbonate buffering in groundwater. Even in these saline systems, dissolved carbonate is present in solution because of the ubiquitous presence of some carbonate minerals and cements in the majority of siliciclastic rocks. Based on the examples presented in the previous sections, we have shown the importance of considering geologically realistic scenarios. Also, to accurately construct geochemical models for prediction, evaluation of experiments and evaluation of actual field applications, the following parameters must be accurately measured or calculated: 1) major and minor ion chemistry 2) dissolved carbonate species concentration (i.e., HCO3− and CO32−), 3) total dissolved CO2 and 4) in situ pH. Because water samples will cool and degas CO2, with associated changes in pH, during extractions from experiments or wells, it is difficult to determine the in situ pH and dissolved carbonate distribution. By measuring the total amount of CO2 (by capturing water and gas quantitatively) and measuring the pH and dissolved carbonate in degassed samples, in situ geochemical conditions (pH and bicarbonate buffering) can be accurately calculated. Using this method, we have shown that reliable predictions of mineralization and dissolution, critical for the successful application of geological CO2 sequestration, are possible with off-the-shelf geochemical modeling software. In terms of large scale sequestration efforts and routine MMV, this approach is attractive because samples can be collected at the surface from nearly any monitoring well, providing an inexpensive and relatively simple alternative and/or addition to in situ sample collection and analysis using sophisticated and expensive down-hole devices.
 We thank Bill Carey for discussion related to geochemical modeling of CO2 saturated systems, Dale Counce for analytical chemistry and Ren Wan and Steve Chipera for X-ray diffraction analysis of experimental solids. Funding for this work was provided by Los Alamos National Laboratory LDRD-DR 2004042DR. This paper was improved by the constructive comments and corrections from two anonymous reviewers.