1 Introduction

Geological sequestration of CO2 could play a crucial role in mitigating global CO2 emissions. It has the potential to substantially reduce emissions at fossil fuel energy production plants and other highly polluting industries, thereby contributing to the effort to limit global temperature increases to below 1.5 °C (IPCC 2023). It hinges on appropriate geological structures, commonly sedimentary basin rock formations with upper sealing layers, prevalent in the Earth’s crust and often saturated with brine, primarily composed of jointed rocks.

Fluid movement within rock formations involves matrix pores, fractures, or a combination of both (Iding and Ringrose 2010), with fractures often augmenting permeability (Berkowitz 2002; Cordero et al. 2019). In fluid dynamics, CO2 plume migration tends to follow paths of least resistance or highest fluid conductivity, favoring fractures over low-permeability matrices, especially in unconventional CO2 reservoirs (Senger et al. 2015) where matrix permeability is very low.

Even at considerable depths and high normal stress, fracture network permeability remains substantial, enduring up to 10 MPa of normal stress. Triaxial permeability tests (Van Stappen et al. 2018) confirm that despite aperture and permeability reductions of approximately 40 and 90%, respectively, fluid flow continues to be governed by fractures over the matrix. This persists due to fractures maintaining a residual aperture after surpassing the critical normal stress, preventing further closure (Indraratna et al. 1999).

During the injection of CO2 into a geological reservoir, it undergoes distinct phases (Bachu 2008; Metz et al. 2005), determined by CO2 phase and flow conditions. Initially, there is a dry area near the well, where CO2, often in a supercritical state due to pressure and temperature, displaces local brine during injection. The second phase involves a two-phase flow region, where immiscible CO2 bubbles or ganglia move through void spaces filled with brine. Afterwards in the distant regions, the CO2 plume dissolves into the brine, creating an acidic solution (Eq. (1)). This dissolution occurs at the interface of CO2 and local brine, with the denser solution flowing downward, forming gravity fingers (Gale et al. 2009; Riaz et al. 2005; Neufeld et al. 2010).

Fig. 1
figure 1

Fluid regions at a CO2 storage reservoir, with WAG stimulation (red rectangles refer to the CO2–water interface, where the acidic solution may occur

Full size image

The Water Alternating Gas (WAG) method, involving periodic water injection alongside CO2 (Fig. 1), (Nghiem et al. 2009) boosts injectivity rates, accelerates CO2 trapping, but fosters multiple CO2-brine interface fronts within the rock mass, increasing dissolved CO2 in water and promoting residual trapping, along with acidic flow conditions. Carbonic acid can also form beyond the CO2-brine interface (plume front), including within pore spaces where residual CO2 dissolves in water (Fig. 2).

Fig. 2
figure 2

Formation of carbon acid in pore space from residually trapped CO2

Full size image

The chemical reactions forming acidic solutions can alter reservoir characteristics (mechanical and hydraulic), particularly in sedimentary rocks rich in calcite like limestones, chalks, and sandstones (Eq. (2)), (Rohmer et al. 2016). These rocks commonly contain calcite, silica, and iron oxides as cementation agents. The reaction kinetics, influenced by dissolved CO2 concentration in local brine (Elkhoury et al. 2013), affect susceptible rock areas or volumes within the reservoir.

$${\text{H}}_{2} {\text{O}} + {\text{CO}}_{2} \to {\text{H}}_{2} {\text{CO}}_{3}$$
(1)
$${\text{CaCO}}_{3} \left( {\text{s}} \right) + {\text{H}}_{2} {\text{CO}}_{3} \left( {{\text{aq}}} \right) \leftrightarrow {\text{Ca}}^{{2 + \left( {{\text{aq}}} \right)}} + 2{\text{HCO}}_{3}^{{ - \left( {{\text{aq}}} \right)}}$$
(2)

Permeability in reservoir fracture networks is influenced by multiple factors (Liu et al. 2016) including aperture distribution, fracture roughness, length, intersection patterns, boundary stress, field scale, and hydraulic pressure gradients. Many laboratory experiments investigating CO2 geologic sequestration have focused on intact rock samples (cores) to examine the flow through the pore network and the relative permeability of CO2 (Izgec et al. 2007) (Levine et al. 2011) (Canal et al. 2014). Other studies have explored the formation of conduits, such as “wormholes”, through dissolved rock matrix (Smith et al. 2013) and the dissolution of minerals and overall geochemical response (Bemer and Lombard 2010) (Grgic 2011) (Ellis et al. 2013). Flow-through fractures, particularly in CO2 sequestration, have attracted research attention. For instance, a study analyzing CO2 storage in the Insalah project (Algeria) emphasized the pronounced CO2 dissolution within fractures (Iding and Blunt 2011). Larger void spaces in fractures create a substantial interface between CO2 and brine, accelerating CO2 dissolution. To maximize CO2 occupancy in these void spaces, rapid injection is necessary, further enhancing the CO2-brine interface.

Simulating fracture networks is crucial for assessing their impact on CO2 flow and capture. Digital Fracture Networks (DFN) are efficient for such simulations (Iding and Ringrose 2010). However, modeling fractured reservoirs for CO2 flow is challenging. A study in Bigi et al. (2013) compared DFN and Analogue Model (A.M) methods, revealing that fracture plane intersections and realistic geometries based on field data affect field permeability anisotropy. DFN relies on statistical parameters for predicting permeability, while A.M. incorporates detailed field-based representations, including aperture, orientation, and roughness. They concluded that commercial DFN solutions, relying solely on aperture distribution, underestimated reservoir permeability by two orders of magnitude compared to A.M.

Fracture roughness and aperture variability significantly impact flow patterns (Elkhoury et al. 2013). Their study compared natural (rough) and artificially (saw-cut) fractured cores under reservoir conditions, revealing that flow rate, represented by the modified Damkohler number (Da), influences dissolution flow regimes. Lower flow rates favor channeling flow and elongated dissolute paths. Additionally, fracture network orientation, particularly vertical orientation, enhances natural convection mixing, known as gravity fingering (Hassanzadeh et al. 2007) (Mojtaba et al. 2014) (Emami-Meybodi et al. 2015). In vertically oriented fracture networks, gravity-driven fluid mass transport is facilitated (Rezk and Foroozesh 2019).

CO2 injection near the well utilizes both pore space and fractures due to significant pore pressure buildup within fractures. In contrast, remote free-phase CO2 plumes mainly migrate through fractures in low-pressure environments, unable to penetrate rock matrix surfaces (Oh et al. 2013). Moreover, inside low-permeability rocks with high entry pressure, fractures serve as the preferred short-term migration paths, supplementing primary routes. Notably, fractures’ void space can store substantial CO2 volumes, as exemplified by the Svalbard potential reservoir, where it may account for up to 2.5% of total storage capacity (Senger et al. 2015).

As mentioned, roughness is among the fracture properties that impact flow. In low Reynolds number flows, as anticipated within CO2 reservoirs, roughness is inversely correlated with fracture’s transmissivity (Crandall et al. 2010). Rough fractures reduce transmissivity while smooth fractures increase it. Although roughness can cause tortuosity and recirculation flow zones (Lee et al. 2014, Chen et al. 2017), these phenomena are unlikely to occur inside a CO2 reservoir, apart from the injection point area.

Aperture significantly impacts fluid flow, following the simplified Cubic law (Eq. (7)), which relates flow rate to aperture cubed (Witherspoon et al. 1980). Changes in aperture strongly influence flow, but natural rough fractures pose challenges in accurately measuring it. Mechanical aperture (geometric distance) doesn’t always match hydraulic aperture (flow-related). Various aperture types are proposed, including mean, hydraulic, mechanical, vertical, and apparent (Fig. 3). Additionally, the Cubic law’s flow rate prediction can be inaccurate for rough fractures with small aperture deviations, favoring mean aperture use (Brown 1987). In shear displacement within rough fractures, greater dilation and aperture boost transmissivity parallel to displacement, particularly in rougher fractures (Xiong et al. 2011).

Fig. 3
figure 3

Different definitions of aperture in rock fractures (after (Oron and Berkowitz 1998) as appeared in (Zhang and Chai 2020))

Full size image

Flow through a fracture network, especially for low matrix permeability and fractures with aperture that follows a log-normal distribution, is significantly influenced by stress and shear displacement, as confirmed by DEM (distinct element methods) simulations (Baghbanan and Jing 2008). However, it is important to note that extracting a permeability tensor for use in continuum media models can be challenging. Therefore, it is recommended to use a distinct element approach to obtain more realistic predictions of the hydraulic performance, which is strongly dependent on mechanical factors.

To our knowledge, not many studies have assessed the impact of CO2 geologic storage on fractured rock. We conducted a study on a fractured sandstone sample from a potential CO2 reservoir in North-West Greece. The sample had low matrix permeability and was rich in calcite, indicating carbonate cementation. We subjected the sample to an acidic CO2–H2O solution flow under high normal stress for 30 days. Prior to and after CO2 erosion, we measured water transmissivity exclusively. To estimate fracture aperture, we used the loading and unloading method for vertical displacement convergence before conducting the flow experiments. Our study aimed to enhance understanding of the hydraulic performance of fractures under the influence of CO2.

2 Theoretical Background

Flow-through fractures is governed by Navier–Stokes equations (inertial forces and viscous forces interactions, balanced by external pressure), in the general form (considering incompressible fluid):

$$\mu \nabla^{2} \vec{u} = \nabla p + \rho \left( {\vec{u} \cdot \nabla } \right)\vec{u}$$
(3)
$$\nabla \cdot \vec{u} = 0$$
(4)

where \(\vec{u}\) velocity vector, p pressure, μ is the dynamic viscosity, ρ fluid density, μ fluid viscosity, (Eq. (4) is the continuity equation representing mass conservation). Considering velocity gradient and velocity vector orthogonal to each other, Stokes equation is generated:

$$\mu \nabla^{2} \vec{u} = \nabla p$$
(5)

Usually, further simplification assumptions are made, considering the steady-state flow of incompressible fluid, ignoring time-dependent terms, and considering inertial forces negligible to viscous forces (slow flows) as well as gradual geometry (aperture) variations. The above leads to the Reynolds equation:

$$\nabla \cdot \left( {e^{3} \nabla P} \right) = 0$$
(6)

where e stands for aperture in which, flow takes place. For smooth parallel plate assumption of fracture surfaces, or very slow flows (with low Re—Eq. (8)), cubic flow derives:

$$Q = \frac{{ - we_{h}^{3} }}{12}\frac{1}{\mu }\nabla P = - T\frac{1}{\mu }\nabla P$$
(7)
$$Re = \frac{\rho Q}{{\mu w}} = \frac{{\rho u\overline{a}}}{\mu }$$
(8)

With Q for volumetric flow rate, w fracture’s width normal to flow direction, eh hydraulic aperture, T hydraulic transmissivity of the fracture, Re Reynolds number, ρ density of the fluid, u fluid velocity, \(\overline{a}\) mean aperture.

For the one-dimensional laminar flow of incompressible flow (water), Darcy’s law may be applied inside rock fractures:

$$Q = - k\frac{1}{\mu }\frac{A}{L}\Delta P$$
(9)

where A fracture’s area perpendicular to flow and k permeability (m2) of fracture.

For low flow rates (low fluid velocity), we can assume linear flow, with flow rate proportional to pressure gradient. For higher fluid velocities (e.g., near injection point) non-linear law should describe the flow rather than Darcy’s law.

Fracture permeability (from Eqs. (7) to (9) – assuming that pressure gradient exists only towards fracture’s length L) can also be written as:

$$k = \frac{{a^{2} }}{12}$$
(10)

The fracture’s aperture does not remain constant, as suggested by the parallel plate model, even for short sections along the flow path. Aperture distribution in fractures follows one of three types of distributions: normal ((Lee and Cho 2002, Su et al. 2020), power-law (Marrett et al. 1999), or log-normal (Snow 1970, Hakami and Nick Barton 1990, Renshaw 1995, Keller 1998)). In a flow model, the representation of fractures becomes more critical as the aperture distribution becomes narrower (i.e., with a lower standard deviation). Neglecting their presence can lead to inaccurate flow simulations (Gong and Rossen 2017).

3 Materials and Methods

3.1 Sandstone Fractured Core Samples

For our study, we collected sandstone samples rich in calcite from the Mesohellenic basin located in the North-West region of Greece. This formation is chronologically part of the Miocene period and is known to have potential hydrocarbon reserves (Kontopoulos et al. 1999). The sandstone blocks used in this experiment were obtained from surface outcrops.

The sandstone sample’s mineralogy was determined through XRD analysis, revealing the following composition: 47% calcite, 16% quartz, 14% feldspar, 8% clay, 7% mica, 5% chlorite, and 3% dolomite (Fig. 4). To obtain the necessary specimens for testing, a long core was wet-drilled in the laboratory and divided into two short cylindrical specimens measuring approximately 38 × 38 mm (~ 1.5 × 1.5 inches). The first specimen was subjected to the indirect tensile strength method, also known as the Brazilian method, resulting in a tensile fracture that split the specimen into nearly equal halves (Fig. 5). The second specimen was left intact and used to estimate the rock matrix permeability. The mechanical properties of the sandstone are presented in Table 1, which includes the observed water-weakening effect (Broch 1979; Vásárhelyi and Ván 2006).

Fig. 4
figure 4

SEM image of sandstone thin-section (after (Koukouzas et al. 2018) whom we provided the same sandstone material used in this study) clay matrix marked with yellow and porosity with blue

Full size image
Fig. 5
figure 5

Cylindrical Specimens of this study (picture taken on grid “mm” paper), with embedded sampling area (Mesohellenic basin), and proximate thermo-power-plants

Full size image
Table 1 Sandstone’s characteristics
Full size table

Quantifying roughness of rock fracture can be valued either by widely used empirical method (joint roughness coefficient-JRC (Barton and Choubey 1977)) or statistical methods (fractal dimension evaluated by variogram analysis—(Huang et al. 1992; Mandelbrot 1982)). In our study we didn’t have access to a 3d laser scanner, our primary objective was not the investigation of roughness impact on CO2–H2O flow. To assess the surface roughness in the direction of flow, we employed a profilometer to measure profiles along six lines that were parallel to flow (Fig. 6) allowing us to extract JRC for each profile. For more accurate selection of JRC value, we digitized the 6 profiles, and we estimated the Z2 coefficient (dimensionless parameter used to quantify roughness (Myers 1962)) by employing Eq. (11). Later Z2 value was correlated to JRC by Eq. (12) (Tse and Cruden 1979). The sample is characterized as undulated rough, results can be seen in Table 2.

$$Z_{2} = \sqrt {\frac{1}{{M\left( {Dx} \right)^{2} }}\mathop \sum \limits_{i = 1}^{M} \left( {z_{i + 1} - z_{i} } \right)^{2} }$$
(11)
$${\text{JRC}} = 32.2 + 32.47\;\log Z_{2}$$
(12)

where L length of interest, Dx interval length, M number of intervals, z vertical distance.

Fig. 6
figure 6

Roughness profiles of the fractured sample

Full size image
Table 2 Roughness results
Full size table

3.2 Experimental steps

The experimental procedure consisted of the following steps:

  • The initial mechanical aperture of the fracture was estimated by conducting normal loading–unloading cycles and measuring the displacement (closure).

  • A micro-CT scan was performed on the fractured sample in its unaltered state.

  • The water permeability of the fracture under normal stress was measured for increasing steps of normal stress, while the rock matrix permeability was measured in an unfractured sample of equivalent size.

  • The fractured sample was exposed to a low flow rate of CO2–H2O acidic solution flow for 30 days under normal stress, resulting in erosion of the fracture surface.

  • The water permeability of the altered fractured sample was measured again for comparison purposes.

  • A micro-CT scan of the altered fracture was conducted to extract the aperture distribution and compare it with the initial condition.

3.3 Aperture closure measurement

To estimate the initial mechanical aperture of the fractured sandstone sample, we followed a procedure described by (Bandis et al. 1983). The procedure involves loading and unloading the fracture by vertical stress for three cycles, with the asymptote of the loading–unloading curves representing the total closure of the fracture’s aperture. This value corresponds to the initial mechanical aperture of the fracture when the surfaces are matched under their own weight. The loading and unloading were performed using a servo-mechanically controlled 10t GDS VIS Loading system, which recorded digital acquisition data for vertical displacement and load.

The fractured specimen was molded into a cubic shape using cement mortar, with the fracture plane left un-mortared. The fracture plane was carefully aligned parallel to the loading plates, and the sample was inserted between the plates of the loading system. The fracture was loaded and unloaded three times, with five ultimate normal stresses (σn) selected at 1, 2, 3, 4, and 5 MPa. The load–unload (L–U) test was repeated at least five times for each stress. After each L–U test, the fracture halves were carefully separated and left to match again under the specimen’s weight only. The average maximum displacement (fracture closure) was extracted for each test to estimate the mechanical aperture, taking into account the solid rock’s deformability for calculating the final “closure” of the mechanical aperture.

3.4 Water permeability of matrix and fracture transmissivity measurements

To assess the importance of pore network fluid flow in the fractured sample, we conducted a permeability test on an intact cylindrical sample using the constant head pressure method. The sample was placed vertically in a hassler-type core-holder from Vinci Technologies™ and permeability tests were conducted at 5 different confining pressures (4, 6, 7, 9, and 10 MPa). The sample was subjected to varying differential water pressure (ΔP) (2.5–5–7.5 MPa), which was achieved by adjusting the inlet pressure only, while maintaining the outlet pressure atmospheric (0 Pa). The volumetric flow rate of the fluid was determined by measuring the mass of water outflow using a high precision KERN lab scale, while the inlet pressure was applied using a GDS high precision pressure/volume controller.

The transmissivity of the fracture was re-measured using the Constant Head method under a steady pressure gradient (ΔP). The water flow rate was determined by measuring the outflow mass of water as previously described. The inlet pressure (Pin) was set to 50 kPa while the outlet pressure (Pout) was left atmospheric (0 kPa). To prevent sudden normal stress on the fracture and avoid damage, confining pressure (Pc) was applied in steps (0–100–200–300–500–750–1000–2000–3000 kPa). For each step, the flow was allowed to equilibrate while flow measurements were continuously captured. Except for the lower confining pressure steps, where Pin and Pc were of the same order of magnitude, the applied Pc was considered the effective normal stress on the fracture’s surface since fluid pressure became negligible (Eq. (13)).

$$\sigma_{n}^{\prime } = P_{c} - u_{w} \to ^{{u_{w} = P_{in} - P_{out} \left\{ {P_{out} = 0} \right\}}} \sigma_{n}^{\prime } = P_{c} - P_{in}$$
(13)

where σʹn the effective normal stress, uw water–CO2 solution pressure, Pc, Pin and Pout the confining, inlet and outlet pressure consequently.

3.5 CO2–H2O solution flow through fracture

The fractured specimen was subjected to a CO2–H2O solution flow (acidic flow) at similar temperature and confining pressure to reservoir conditions. The sample was saturated with water and inserted vertically in the core-holder, with the fracture plane parallel to the gravity vector. To prevent surface damage, a 13 MPa confining pressure was applied within an hour, and the specimen was wrapped in layers of PP, Aluminum, PTFE, and Nitrile sleeve to prevent CO2 migration into the confining fluid (Fig. 7). The CO2-saturated solution was prepared with 7% w/w CO2 (Rochelle and Moore 2002; Bachu and Adams 2003)) and left to equilibrate for 24 h inside a piston accumulator. The solution was then injected through the fracture at a flow rate of 2 ml/h (0.033 ml/min or 48 ml/day), controlled by a pressure–volume controller. The low flow rate mimics slow flows at the boundaries of the CO2 plume inside the reservoir, and the outlet pressure is kept at 7.5 MPa by a back pressure regulator to maintain CO2 in supercritical conditions.

Fig. 7
figure 7

CO2–H2O flow Experimental setup. (1) fractured sandstone cylinder, (2a) spacers, (2b) paper filter, (3) nitrile sleeve, (4) oil for confinement pressure, (5) PP tape, (6) aluminum foil, (7) PTFE wrap, (8) pressure transducer, (9) back pressure regulator, (10) high precision lab-scale, (11) piston accumulator, (12) pressure/volume controller, (13) deep tube CO2 tank, (14) Hassler type core holder

Full size image

Experimental measurements were taken over a 30-day period during the flow of CO2–H2O solution. The data collected included the inlet pressure, outlet pressure, and water-mass outflow. A schematic of the experimental setup can be found in Fig. 7.

To analyze the change in transmissivity (\(k \cdot A\)) during the reactive flow, the Darcy’s law for flow in a length L was calculated using Eq. (7). For the parallel plate flow model, the transmissivity can be accurately calculated as:

$$T = kA \equiv \frac{{wh^{3} }}{12}$$
(14)

where w is the width of parallel plates and h is the distance (height) between the plates.

In order to avoid relying solely on mean values of the fracture dimensions (such as cross-sectional area) and to better understand the impact of the reactive flow on the fracture, we opted to observe the normalized changes in transmissivity (Ti/T0) within the fracture throughout the duration of the experiment. Here, Ti represents the transmissivity at a given time and T0 represents the initial transmissivity, as shown in Eq. (16).

$$T_{i} \equiv kAQ = \frac{kA}{\mu }\frac{\Delta P}{{\mathop{L}\limits_{\rightarrow} }}kA = \frac{ - Q}{{\Delta P_{i} }}L\mu$$
(15)
$$\frac{{T_{i} }}{{T_{0} }} = \frac{{\frac{ - Q}{{\Delta P_{i} }}L\mu }}{{\frac{ - Q}{{\Delta P_{0} }}L\mu }} = \Delta P_{0} \frac{1}{{\Delta P_{i} }}$$
(16)

Fluid flow rate Q remained steady throughout the experiment, CO2–H2O solution was kept under the same conditions, and fracture’s length was considered constant, so transmissivity change is inversely proportional to Pressure gradient ΔPi during the experiment.

3.6 Micro-Ct of the sample, image processing

The fractured specimen was initially scanned using an industrial computer tomograph (Werth Tomoscope HV 225 Compact) to digitize its geometry before exposing it to CO2–H2O flow. The specimen was self-weight matched/interlocked and the scanning resulted in 662 slices with a pixel size of 50 × 50 μm and a voxel depth of 50 μm. The slices were further processed with FIJI (ImageJ) software to reconstruct the fracture in 3D. Figure 8 shows the image processing protocol for 3D fracture reconstruction.

Fig. 8
figure 8

Fracture 3d reconstruction protocol, from the m-Ct image sequence

Full size image

The local thickness tool (Dougherty and Kunzelmann 2007) was used to analyze the fracture aperture distribution. This involved fitting the largest possible sphere in each point of a 3D void space, as shown in Fig. 9.

Fig. 9
figure 9

Local thickness definition

Full size image

After the CO2 exposure, the fractured specimen underwent CT scanning once again. However, this time we had to use different equipment (Soredex Scandora 3D—pixel size 120 × 120 μm, voxel depth 120 μm). While the images produced had lower levels of detail in comparison to the earlier scans, they remained adequate for capturing the changes in the fracture’s geometry resulting from the acidic erosion. We followed the same protocol as before to convert a 2D stack of images into a 3D model. Finally, we compared the fracture aperture distribution before and after the CO2 exposure (as shown in Figs. 10, 11, 12, 13 and 14).

Fig. 10
figure 10

CT scan images taken before and after CO2 exposure (z is the z-axis point of an image, here: 0.5–5–20–25 mm)

Full size image
Fig. 11
figure 11

Fracture void space before and after CO2 flow (left self-weight closure before CO2 flow, right after CO2 flow and 13 MPa confinement)

Full size image
Fig. 12
figure 12

3D reconstruction of fractured sample a before CO2 flow, b after CO2 flow. Fracture initial aperture distribution and final flow conduits formation can be seen

Full size image
Fig. 13
figure 13

Fracture a before and b after, CO2–H2O flow

Full size image
Fig. 14
figure 14

Steps involved in processing the images to determine the void area in the fracture before (a) and after (b) exposure to CO2. It should be noted that the CT scan for (a) was performed while the fracture was under self-weight only, whereas the CT scan for (b) was taken after the fracture had been under 13 MPa confinement pressure without being dismantled before scanning

Full size image

4 Experimental Results and Discussion

4.1 Estimation of Fracture’s Initial Aperture

Results of the aperture closure tests showed significant fluctuations (Fig. 15). A similar phenomenon was observed in micro-indentation tests conducted on unaltered and CO2-altered Entrada sandstone samples from Utah, USA, by researchers from Texas (Sun et al. 2016). The fluctuations in their study were attributed to the heterogeneity of the grain scale. In our experiment, the fluctuations are likely due to the micro-mismatch of the fracture surfaces at the beginning of each test repetition, as the two halves needed to be reassembled each time. This mismatch may have resulted in different initial effective normal stresses for the same load (different contact areas) and considering the low elastic modulus of the water-saturated soft rock material (E = 2.3 GPa), its elastic deformation response could justify the fluctuation in closure. Although the curves have an identical form, their starting point may have been shifted due to the aforementioned reasons.

Fig. 15
figure 15

Fractures aperture evolution under normal stress

Full size image

Because of the fluctuation in the aperture closure results, caused by the micro-mismatch of the fracture surfaces, a significant number of test repetitions (37) were conducted to determine the average/mean value of the fracture closure at 5 MPa normal stress, which was found to be 350 μm (Fig. 15). This value was considered as the initial mechanical aperture of the fracture. The fracture exhibited significant closure until 1 MPa normal stress (300 μm), after which further increases in normal stress up to 5 MPa did not induce aperture closure at the same rate.

It is worth noting that fracture closure has been shown to be scale-dependent (in experimental studies on hard rocks such as granites), as reported by (Raven and Gale 1985; Yoshinaka and Yamabe 1986). In general, longer fractures or joints tend to exhibit more closure in aperture for the same level of normal stress, (Raven and Gale 1985), however, it should be noted that these previous studies utilized parallel plate equivalent flow models.

4.2 Rock Matrix Permeability

The permeability of the sandstone matrix was measured to be an average of 1e − 18 m2 (equivalent to 1e − 3 milliDarcys) (Fig. 16). This low permeability is consistent with fine-grained and tight sandstones that have a porosity of approximately 8% (Alam et al. 2014; Zisser and Nover 2009) and mesoporous rocks (Boada et al. 2001). The pore-throat diameter of our sandstone was measured to be 1 micron (Fig. 17).

Fig. 16
figure 16

Rock matrix water-permeability results under different confinement pressure

Full size image
Fig. 17
figure 17

Pore-throat size distribution measured by a mercury injection porosimeter (MIP)

Full size image

4.3 Acidic Flow Exposure (CO2–H2O Solution Flow)

In order to maintain the 30-day experimental duration, the piston accumulator had to be refilled three times (Fig. 18). However, during the solution preparation process, which involved closing all flow valves, the flow was temporarily halted (Fig. 19). This pause in flow and the permanent confining pressure of 13 MPa may have caused some partial closure of the fracture aperture. Additionally, the presence of gouge particles in the fracture may have obstructed aperture closure. These factors may explain the short anomalies or spikes in the pressure graph (Fig. 18). We suspect pressure built-up before flow resumed, possibly due to temporary obstructions from the gouge materials. However, during the second and third refilling/preparation of the new solution, the pressure build-up was lower and of shorter duration.

Fig. 18
figure 18

Differential pressure of fractures inlet–outlet. In red ovals, pressure rise at flow restart after piston accumulator refill

Full size image
Fig. 19
figure 19

Water-only outflow. The reduction of water flowrate marked in red ovals is due to two-phase flow (when sc-CO2 is in excess, co-injected with CO2 saturated water solution through fracture at the end of pistons travel)

Full size image

Overall, the inlet pressure decreased from 270 to 150 kPa at the end of the experiment (average values), resulting in a rise of water transmissivity by almost a factor of two. This increase in transmissivity was due to the formation of flow channels through the fracture (dissolute areas), as revealed by CT-scan comparison (Figs. 11 and 12). The reactive flow dissolved the fracture’s surfaces, resulting in a larger available aperture-void space for flow to take place (Figs. 10 and 14), which eventually led to the rise of the fracture’s permeability and the subsequent pressure drop for a steady flow rate of CO2–H2O solution.

4.4 Fracture’s Water Transmissivity Alteration

Once the reactive flow had ceased and favorable flow paths had been established in the confined fracture, a subsequent test of water-only flow transmissivity demonstrated a substantial increase. The flow rate through the altered fracture rose by nearly tenfold (i.e., an order of magnitude) compared to the initial conditions (Pinlet: 40 kPa, Poutlet: 0 kPa), increasing from 0.0002 to 0.0025 ml/hr, under an effective normal pressure of 5 MPa, which is similar to the high in situ stresses (13 MPa) Fig. 20).

Fig. 20
figure 20

Evolution of water flow rate (Q) inside sandstone fracture, under same pressure gradient (ΔP), before and after acidic flow, for normal stress up to 13 MPa

Full size image

The increase in normalized transmissivity, observed only during the final stage of the experiment (last five days-Fig. 21), is believed to be a result of two-phase flow. This interpretation is supported by a recent study (Ott and Oedai 2015), which investigated the formation of wormholes in carbonates. The study found that when a pure acidic solution (CO2 saturated brine) was injected into carbonate rocks, wormhole formation prevailed, resulting in a narrow evolving dissolution front along the flow stream. However, when CO2 was co-injected with brine in semi-formed wormholes (which had not formed all the way to the other end of the cylindrical specimen) under the same conditions, it tended to result in compact dissolution rather than following the narrow dissolution front of the wormhole.

Fig. 21
figure 21

CO2–H2O solution transmissivity change throughout the experiment; change was compared to transmissivity occurring on day 5 of the experiment where the flow was equilibrated

Full size image

In comparison to our experiment, it is possible that narrow primary flow channels were formed, which during the presence of two-phase flow were dissolved further in a more uniform manner (compact dissolution), resulting in a significant increase in void space (larger conduits) (Figs. 22 and 23).

Fig. 22
figure 22

Cumulative distribution of void space area (fracture cross-section area) per ct-scan slice

Full size image
Fig. 23
figure 23

Distribution of fracture’s cross-section area before and after exposure to CO2–H2O solution. Arrows indicate the formation of new dissolute areas (preferential flow conduits/ channels). Should be noted that fracture was ct-scanned under self-weight interlock before exposure to CO2, while after exposure it was ct-scanned without being reassembled (opened and left interlock under self-weight) with potential permanent deformation due to 30-day 13 MPa normal stress

Full size image

The increase in transmissivity is attributed to the formation of flow conduits that resulted from the dissolution of grains’ carbonate cement (calcite) and calcite grains on the fracture’s surface by the acidic flow. Two main flow paths were established on the fracture’s sides (Fig. 24a and b), where surface dissolution was most intense, leading to an increase in local available, cross-sectional area for fluid flow. CT-scan results showed that the initial void space cross-section measured up to 1.5 mm2, with most openings less than 0.5 mm2. However, after the reactive flow, openings measured up to 13 mm2 (Figs. 22 and 23). It should be noted that the initial measurements were on a self-weight interlocked fracture, while the second measurement was taken after the sample had undergone significant normal stress (13 MPa total, 5 MPa effective, for 30 days, with potential permanent deformation) which was not separated before CT-scanning. Nonetheless, the increase in void space due to mineral dissolution/erosion is undeniable. The area of the fracture’s surface, where the flow paths were maintained, was calculated to be 28.9% of the total area (Fig. 24).

Fig. 24
figure 24

a Normal projection of flow path thickness, b binarization, c outline selection of flow path borders for area calculation, d sample after CO2 flow

Full size image

As stated earlier in the introduction, it is crucial for numerical simulations that employ DFN/DFM for mechanical response and DPDP models for hydraulic simulations to consider the fact that changes in the mechanical and hydraulic properties of fractures occur on a small fraction of the available fractured surfaces.

5 Discussion

The permeability of a rock fracture is influenced by a range of static and dynamic factors, such as the mechanical aperture of the fracture, the roughness of its surfaces, and the contact areas within the fracture. These factors can undergo changes due to the stress field exerted on the fracture walls, making it a complex and interconnected phenomenon.

In our experiment, we utilized only a single fractured sample, and consequently, we did not evaluate the influence of fracture roughness on flow. However, it is noteworthy that a recent study (Wang et al. 2023b) on two-phase flow within rough fractures, although it involved Na gas (not CO2), observed that roughness does indeed have an impact. This impact was more conspicuous on water relative permeability than on gas relative permeability. It’s important to mention that as surface roughness increases and flowrates become higher, the flow becomes more tortuous, and the fracture’s conductivity diminishes (Rong et al. 2021). Fractures characterized by increased roughness typically exhibit a reduced hydraulic-to-mechanical aperture ratio as dilation and flow rates increase, because of rising presence of eddies (Zhang et al. 2021, Dimadis et al. 2014).

Moreover, it’s crucial to note that roughness, by itself, does not exert direct control over the flow. Instead, it influences aperture, and it’s the spatial distribution of aperture that governs flow channeling (Lee and Babadagli 2021). The impact of roughness becomes more pronounced under confinement conditions, resulting in smaller aperture values, and dissipates when fracture dilation occurs, such as during shear displacement (Li et al. 2023). Furthermore, in the context of two-phase flow, the relative permeability of fluids and the saturation of the residual phase (Zhang et al. 2021) are contingent on the wettability of the fracture surface (Babadagli et al. 2015).

It’s worth noting that laboratory-induced fractures, when compared to natural ones, typically exhibit lower permeability (Gale 1982). Therefore, it’s essential to exercise caution when attempting to generalize our results to different types of geomaterials and fracture categories, which may range from smoother to coarser. Nevertheless, we believe that our findings provide indicative insights.

Figures 22 and 23, indicate that material dissolution lead in widening of local aperture which consequently concentrate the reactive flow. This allows for further acidic attack on the surrounding surface resulting on widening of the formed flow paths, leaving rest of fracture surface almost untouched.

As depicted in Fig. 19, during the initial 5 days of the experiment, the flow rate was not 48 ml/day as intended. Instead, it was lower due to an issue with the syringe pump’s operation. While the flow rate was maintained at the intended slow rate during this period, we cannot be certain whether this had an impact on our results, particularly in terms of potentially reducing dissolution rates. However, we can confidently assert that it did not have a detrimental effect on the transport of solids or the potential for clogging.

Figures 22 and 23 illustrate that the dissolution of materials resulted in rise of local aperture, leading to the concentration of reactive flow. This facilitated further acidic interaction with the adjacent surfaces, causing an enlargement of the established flow pathways while leaving the remaining fracture surface largely unaffected.

As previously mentioned, we deliberately maintained the normal load on the fracture surface below 13 MPa to avoid any potential damage to asperities. This precaution was taken to preserve the surface in an unaltered state for the repeated phases of testing. However, we acknowledge that in reservoir depths, normal stress may exceed this threshold. Despite this, we hold the belief that there will not be a significant deviation in fracture transmissivity. Our confidence in this belief is grounded in the findings presented in Fig. 21, which align with the observations of numerous studies. These studies, including the work of Bandis, (Bandis et al. 1983), indicate that the closure of fractures tends to approach an asymptote. This phenomenon also holds true for fluid flow rates, as demonstrated in the research of Raven (Raven and Gale 1985).

Regarding rough natural fractures, it’s important to note that the closure of a fracture, which involves the reduction of mechanical aperture, enhances channeling flow. While this may lead to a reduction in transmissivity, it does not result in a complete cessation of flow, as evidenced by studies like the one conducted by Zou (Zou and Cvetkovic 2020).

Furthermore, when considering reservoir depths for CO2 storage, the estimated maximum overburden stress could reach approximately 20–25 MPa (assuming a depth of 1 km and a density range of 2–2.5 g/cm3). Even in cases where a fracture experiences such normal loads and exhibits ductile behavior, involving the plastic deformation of asperities, there may still be presence of hydraulic aperture (Lavrov 2017). It’s essential to understand that overburden stress does not always equate to the normal load acting on the fracture, as fractures can have orientations other than horizontal. Consequently, overburden stress can induce fracture displacement or dilation, leading to increased hydraulic apertures, (A. Baghbanan and Jing 2008), therefore chosen maximum normal stress (13 MPa) of this study may be representative for geologic reservoirs.

In our specific case, the induced fracture was of limited length size (3.8 cm). In natural settings, fracture sizes can vary considerably, often extending to greater lengths. For fractures with greater lengths, closure under the same normal load may indeed be more pronounced (Giwelli et al. 2009). However, it’s worth noting that in lengthy fractures, shear displacement can also occur, which may contribute to increased permeability due to dilation, (Wang et al. 2023a).

6 Summary and Conclusions

This study aimed to investigate the effects of CO2–H2O flow on an artificially induced fractured cylindrical specimen composed of calcite-rich sandstone, which was extracted from a potential CO2 geologic sequestration reservoir located in North-West Greece. For a duration of 30 days the specimen was subjected to 13 MPa confining pressure and temperature of 33 °C, to simulate conditions that are likely to be encountered near an injection well.

The specimen was subjected to a CO2–H2O flow at a rate of 2 ml/h, under a confining pressure of 13 MPa and a temperature of 33° C. The outlet pressure was maintained at a constant 7.5 MPa to keep CO2 in supercritical conditions, while inlet pressure and water outlet were closely monitored. After a period of 30 days, the overall conductivity of the fracture under normal stress had increased by an order of magnitude.

The initial entry pressure required to open the closed fracture for fluid flow to occur under the 13 MPa confinement was 1.5 MPa. However, once the flow was established, the pressure gradient ΔP ((inlet–outlet pressure)) decreased to less than half its initial value (from ΔPstart: 270 kPa to ΔPfinal:150 kPa) over the same 30-day period.

CT-scan analysis conducted before and after the experiment demonstrated the formation of two primary flow conduits or flow paths, where calcite cement dissolution resulted in increased local aperture. These two flow paths were the only areas where fracture flow was maintained, and their projection area represented nearly 30% of the fracture’s plane.

These findings suggest that accurate modelling of hydraulic behaviour is critical for CO2 sequestration in fractured reservoirs with low matrix permeability and significant Ca content. Even well-matched fracture sets, under considerable normal stress, can be invaded by acidic CO2-water solution that will form inside the reservoir. This is particularly relevant at sites where WAG stimulation method is used, as fresh CO2-water interfaces can lead to significant acidic attack in a short amount of time. The formation of dissolution-driven localized flow conduits within the fracture network can also occur rapidly.

The presence of supercritical CO2 and acidic CO2-water solution inside a fracture can lead to the formation of compact dissolution patterns on the fracture surface, causing dissolved conduits to widen rapidly and further increasing fracture transmissivity.

The utilization of fluid over the entire fracture surface may be limited, which can promote localized dissolution and erosion of the fracture walls. Following exposure to CO2 conditions, the increase in water transmissivity (in the case of WAG technique implementation) indicates a ten-fold rise, allowing for similar predictions about CO2 injectivity.

The findings of this study could be beneficial for modelling large-scale fractured reservoirs with different components.

Subsequent studies will concentrate on the mechanical reaction of fractured reservoirs that have been modified by the presence of CO2-water acidic flow.