Sleipner 26 years: how well-established subsurface monitoring work processes have contributed to successful offshore CO₂ injection Open Access

Initially the CO₂ injection operations were licensed and performed under the Norwegian Petroleum Act. The CO₂ injection project was regarded as an important learning opportunity, and several research projects were conducted during the first 6–8 years of injection, such as SACS1 and SACS2 (‘Saline Aquifer CO₂ Storage Project’), CO₂STORE (‘On-land long term saline aquifer CO₂-storage’), and CO₂REMOVE (‘CO₂ Research–Monitoring–Verification’). Time-lapse seismic monitoring was just emerging as a viable technology for offshore oil and gas monitoring at the time (Lumley 2001) and both the SACS and CO₂REMOVE projects co-financed seismic monitoring programmes over the Sleipner CO₂ injection site. Similarly, seabed gravimetric monitoring was being developed in the early part of the new millennium (Eiken et al. 2008), with major improvements in measurement precision. Both seismic and gravimetric monitoring were applied for Sleipner CO₂ monitoring, with promising results. These early monitoring projects had strong research elements, with a focus on proof of concept.

In 2014, the EU directive for CO₂ storage was implemented as part of Norwegian law. It outlines regulations for operator responsibilities related to CO₂ monitoring, with a special focus on monitoring; both for CO₂ conformance and containment, and as an important factor in providing contingency measures. The EU directive is implemented both in the Norwegian Petroleum Act and in the Norwegian Pollution Act. CO₂ injection at Sleipner was then permitted (technically re-permitted) under the new regulations in 2016, based on three main components of the monitoring programme: time-lapse seismic monitoring (twice in the 2016 to 2021 time-period), monitoring of CO₂ pressure and temperature at the injection well head, and monitoring of the CO₂ volume stream using an orifice flow meter. Below we will focus mainly on results from the subsurface monitoring, with a special focus on seismic.

Since injection start, a total of 10 repeated 3D seismic surveys have been acquired (in 1999, 2001, 2002, 2004, 2006, 2008, 2010, 2013, 2016, and 2020), all with reference to a seismic baseline acquired in 1994 (Furre et al. 2017). Acquisition parameters for these are shown in Table 1. The frequency of the seismic repeats reflects an early need for learning (with frequent repeats in the early 2000s), and a realization that less frequent repeats would be needed as confidence in the storage safety increased and injected volumes decreased.

Over the nearly 30 years that have passed since the base line seismic survey was acquired, there have been many advancements both in seismic acquisition and processing technology. For these types of long-term seismic monitoring projects there will always be a trade-off between the need to maintain consistency for optimized repeatability, and the desire to incorporate new technologies which will potentially add value. Cost will need to be balanced with the benefit of optimizing acquisition, processing, and interpretation. In the early years of Sleipner CO₂ injection the seismic surveys were often combined with other acquisition purposes, which meant that they were usually optimized for deeper targets than the relatively shallow depth of Utsira Fm. (Furre and Eiken 2014).

Some of the repeated surveys only covered parts of the CO₂ plume or were shot deviated from the base line (NS) seismic shooting direction (e.g. 2002 and 2004). A larger repeat area than previous surveys was acquired in 2020, in response to the observation of CO₂ starting to migrate towards the west in 2016 (Fig. 1b). The extension towards the west was shot with EW lines to avoid the Sleipner Øst platform.

As new technology emerged, new broadband acquisition technologies like dual streamer (in 2010), slanted cable (in 2013), and multi-sensor (in 2020) were deployed, and optimized processing technologies were also tested (Furre and Eiken 2014; Furre et al. 2017; Wierzchowska et al. 2021). There has also been a continuous effort to reprocess new vintage datasets in a consistent manner, although it has not been a priority to reprocess all prior time-lapse surveys every time new developments have been incorporated. This means that the baseline data have been reprocessed and matched for each new repeat and that repeat vintages have different levels of amplitude matching.

Despite the diverse history of seismic repeat surveys, the CO₂ plume has always been easy to interpret, and the plume size has gradually increased over time (Fig. 3), with individual layers of up to 20 m in thickness (Boait et al. 2012; Furre and Eiken 2014; Furre et al. 2017). There has been no indication of any CO₂ escaping into the overburden (Furre and Eiken 2014) and maximum pressure buildup has been estimated to be below 1 MPa (Chadwick et al. 2012).

In addition to the extensive repeat 3D seismic survey program, several other seismic tests have been acquired at Sleipner, focusing on investigating shallow monitoring technologies. In 2006, several 2D seismic lines were acquired in a star-shaped pattern over the CO₂ plume, using typical site survey acquisition parameters, with a shallower shot and receiver depth (3 m) than for the 3D seismic surveys (typically 5–7 m towing depth). This resulted in much-improved imaging of the plume (figure 12 in Williams and Chadwick 2012). Similar results were obtained by optimizing the 2010 3D repeat, utilizing the benefits of dual streamer acquisition during processing to enhance resolution compared to the 4D processing (Furre and Eiken 2014).

In 2020 and 2021, several 2D seismic lines were acquired using XHR (extended high resolution) short cable streamers (75 m) to investigate whether the technology could provide sufficient quality for time-lapse seismic at these depths (Dehghan-Niri et al. 2022). The tests showed that it was possible to image the CO₂ plume well with improved resolution of some internal detail, however there were limitations related to 2D line migration.

Offshore gravimetric monitoring underwent a rapid development during the early 2000's, and presently detectability is down to a few $μ$Gal (Alnes et al. 2008; Eiken et al. 2008). Gravimetric data provide an independent measure of subsurface density changes, which complements the seismic data. Gravimetric monitoring is limited to relatively shallow depths and does not have the lateral or vertical resolution of time-lapse seismic. On the other hand, it can be used quantitatively for material balance estimates (while seismic is generally less suited for distinguishing between full and partial levels of CO₂ saturation). A benefit of gravimetric surveys is that they come with sub-cm precision measurements of seabed inflation (or subsidence), which is an indirect measure of pressure changes in the subsurface.

Benchmarks for gravimetric surveys were first installed at Sleipner in 2002, and gravimetric repeats were acquired in 2005, 2009, and 2013. The initial layout started with 30 seafloor gravity stations distributed in two 2D lines crossing the plume supplemented with a denser layout over the central part of the plume (Alnes et al. 2008). The layout was later extended with 13 new stations further covering the growing CO₂ plume. Joint inversion of gravimetry and seismic has provided a good estimate of where most of the CO₂ is located (Alnes et al. 2008; Alnes et al. 2011; Furre et al. 2017). There are no measurable indications of seabed inflation over the Sleipner CO₂ injection site, confirming that pressure build-up in the storage unit is minimal (and likely less than 1 MPa; Chadwick et al. 2012).

Due to poor internal reflectivity in the Utsira Fm., only three surfaces could be interpreted with confidence prior to injection start (Fig. 4a): Top Sand Wedge, Top Utsira Fm., and Base Utsira Fm. On the first repeated seismic survey in 1999, nine separate strong amplitude anomalies consisting of pairs of yellow and blue reflections (corresponding to the top (soft) and base (hard) response of layers with reduced acoustic impedance) were observed within the Utsira Fm (Fig. 4b). These were interpreted as CO₂ being trapped beneath various shales, namely at the top of the Sand Wedge, beneath the 5–9 m thick mudstone and beneath several thin mudstone layers within the Utsira Fm. In each case, the change in reflectivity is caused by a soft contrast between a CO₂ saturated sandstone and the brine saturated sandstones and mudstones (Arts et al. 2002).

Figure 5 shows detailed interpretations of each amplitude anomaly along with selected time-lapse RMS (root-mean-square) difference maps for 1999, 2008, and 2020 repeats, respectively. Table 2 shows dimensions of all the nine interpreted anomalies in 2020.

The CO₂ amplitude anomalies are labelled numerically from bottom to top (Fig. 4b–d), with anomalies 8 and 9 referring to the CO₂ trapped underneath the 5–9 m thick mudstone and in the Sand Wedge, respectively (Furre and Eiken 2014). In more recent years it has become increasingly challenging to separate reflections, particularly in the central part of the plume where deeper reflections have weakened, exemplified by the development of reflections 1, 2, 3 and 4 from 1999 (Fig. 4b) as compared with their expression in 2010 and 2020 (Fig. 4c, d).

A vertical corridor of amplitude dimming was observed in the middle of the stack of strong reflections already on the first repeat in 1999 (orange arrow in Fig. 4b). This observation has been referred to as a ‘CO₂ chimney’ in the literature and is interpreted as a vertical flow conduit where the whole sand column is filled with high CO₂ saturation (Arts et al. 2002). In map view, it appears as a circular feature, with a <130-m wide dimming area within an otherwise strong reflection (Fig. 5). It is also slightly wider at the base and narrower towards the top. The CO₂ chimney is still observable on the 2020 repeat survey, but its character has changed over time (orange arrows in Fig. 4b–d).

Most of the amplitude anomalies initially developed in an elongated manner in the SSW–NNE direction, e.g. anomalies 6, 5, and 4 (Fig. 5d–f), whereas a few were more confined around the injection location, e.g. anomalies 1, 3 (Fig. 5g, i). There are several observations of anomalies developing offset from the main anomalies or offset from the main chimney, e.g. anomaly 9 (Fig. 5a – southern anomaly in 2008 and western anomaly in 2020), anomaly 8 (Fig. 5b – southern anomaly in 2008 and 2020, and northern anomaly in 2020), anomaly 7 (Fig. 5c – southern anomaly in 2008), anomaly 6 (Fig. 5d – all vintages), and anomaly 4 (Fig. 5f – northern anomaly in 1999). Some of these anomalies have over time merged with their respective main anomaly whereas others have remained separate. Most anomalies also initially grew faster towards north than towards south, but this tendency appears to have evened out over time. A notable exception is amplitude anomaly 7, which seems to have developed only towards south (Fig. 5c). Over time most anomalies have continued developing in a similar manner to which they started; however, a few have developed more complex patterns, e.g. anomaly 8, which appears to develop in several separate layers (Fig. 5b) or anomaly 2, which started out confined near the injection point, but which now has an elongated character (Fig. 5h). Possible interpretations of these observations are discussed below.

In the overburden we also see soft amplitude responses, looking like the typical CO₂ signature described above. These were observed already prior to injection (white arrow in Fig. 4a) and are believed to be caused by shallow gas accumulations which are abundant in the Sleipner area (Nicoll 2011). In Figure 2a, a crossline through the injection point from the 2020 seismic survey shows these overburden reflections on a larger scale.

Amplitude maps can be used to monitor the overburden for signs of CO₂ migration into the caprock. Figure 6 displays maps of RMS-amplitudes extracted from an interval of the caprock (500–820 ms) above the Utsira storage unit (refer Fig. 2a for location). The maps show amplitudes from 1994 (pre-injection) and 2020. The shallow gas is seen as high amplitudes on the map (typically with values in excess of 7000–8000), and the pattern appears unchanged in this time interval, apart from some acquisition-related noise.

Figure 7 shows the RMS amplitudes of time-lapse differences for two selected repeat times. There is some noise in the maps due to differences in the seismic acquisition geometries, with the lowest noise level in the 2008–20 difference (Fig. 7a) since those two surveys were acquired with more similar acquisition parameters than the 2020–1994 difference (Fig. 7b). However, in both maps the amplitude level is much lower than what would be expected if CO₂ had migrated into the overburden, and there are no signs of extra reflectivity above the CO₂ plume compared to the surroundings.

During the early repeated surveys, all reflections were easily discernible from each other (Fig. 4b). Over time, however, several interpretation challenges have emerged. Most of the deeper anomalies are overlain by one or several shallower anomalies. With increasing volumes of CO₂ in the plume, time delays have developed, making it increasingly difficult to interpret deeper reflections confidently. In addition to time delays, the overlying CO₂ will also have the effect of attenuating seismic waves and generating peg-leg multiples (Boait et al. 2012; Furre and Eiken 2014). Finally, as CO₂ layer thicknesses increase, reflections will start interfering with each other (see modelling below). All these factors contribute to complicate interpretations in areas where several reflections are stacked.

Note that all interpretations have until now been performed in TWT (two-way-time) because uncertainties in saturation and layer thicknesses means that there is a large uncertainty on intra-plume velocities. Despite working to develop a full waveform inversion of the 2010 data (Mispel et al. 2019), we do not presently have a detailed enough velocity model to perform proper depth conversion within the CO₂ plume for all vintages.

Simple seismic modelling was conducted to illustrate typical CO₂ responses (Fig. 8). The modelling was based on exploration well 15/9-13, located approximately 900 m from the injection point. This well was chosen because it is believed to represent the properties of the injection site better than the highly deviated injection well 15/9-A-16 (which at Top Utsira level is located 1300 m laterally offset from the injection perforation interval). Well 15/9-13 has several thin mudstone intervals which were used as basis for the modelling. Note that all these intervals cannot be confidently correlated to the seismic observations in the plume, so this modelling should be regarded as a ‘type well’ modelling rather than an accurate representation of the plume. Well 15/9-13 was drilled in 1984 and data log quality is generally poor. Even though both density and compressional sonic logs were acquired in the well, large sections of the Utsira Fm. had to be replaced with synthetic logs based on resistivity and gamma ray relations.

Nine markers were identified in the well, based on high VSH (volume shale) values. The two shallowest markers correspond to the top of the Sand Wedge and Utsira Fm, respectively. Fluid substitution from brine to CO₂ was conducted using Gassmann relations (Gassmann 1951) in the sand underneath each mudstone layer, varying the thickness of the CO₂ substituted zone between 1, 2, 5, 10, and 15 m (Fig. 8). In addition, one scenario was modelled where the sandstones both in the Sand Wedge and Utsira Fm. above injection point were filled with CO₂. This was done to investigate both thin layer effects and the transition to more continuous CO₂ columns.

Endpoint saturations were set at 90%, based on core measurements for Utsira Fm. (Chadwick et al. 2004), and a homogeneous fluid distribution was assumed, generated using harmonic averages for mixing brine and CO₂ (Wood 1955). Fluid properties were calculated using FLAG software. This software calculates elastic fluid properties based on input parameters such as brine salinity, pore pressure, and temperature (here 70 000 ppm, 10 MPa and 30°C, respectively). The synthetic seismic modelling was a simple 1D convolution using a Ricker 30 Hz wavelet representative of the frequency content at Utsira depth (note that such a simple convolution does not model time-shifts caused by velocity push-down). Due to uncertainties related to log quality, thin layer distribution, fluid properties, and CO₂ saturations and thickness, this modelling should only be regarded as a qualitative indicator. The model nevertheless gives a good impression of expected behaviour.

The modelling shows that thin (1–5 m thick) CO₂ filled sandstone layers set up a stack of separate strong reflections with increasingly stronger amplitudes with thickness (Fig. 8, panels 4–6). The top reservoir reflection comes in slightly high due to tuning effects for these thin layers. As layer thicknesses increase, the amplitudes increase too, up to a limit near the tuning thickness which is approximately 8 m for this modelling exercise. For these thicknesses layers also start interfering with each other, setting up complex seismic responses (Fig. 8, panels 7–8). When CO₂ fills the whole succession, only the top and base of the CO₂ filled column sets up strong reflections (Fig. 8, panel 9). In the next section we will use this modelling to support the interpretation of the seismic response that was presented in the previous section.

The main chimney (orange arrow in Fig. 4b–d) has an obvious reflection dimming character, like the modelled response shown in Figure 8 (panel 9). There is also a break in the reflection continuity at Top Utsira Fm, and a strong amplitude response at top of the Sand Wedge, most notably observed on the 2010 and 2020 repeats (Fig. 4c, d). We interpret this as being due to all sandstone layers being partly or fully saturated with CO₂, leading to minimal acoustic contrast within the chimney. As this dimming area is set in an environment of stacked, highly contrasting reflections, it is prominent both on vertical seismic cross sections (Fig. 4b–d) and on map view (dimmer amplitudes or even total absence of reflections within the chimney in Fig. 5, especially notable for Layers 8, 7, and 5).

Another vertical chimney with seemingly shorter extent is also observed in 2010 (yellow arrow in Fig. 4c, d). It appears as a small break in layer 5 and seems to stop in layer 7. Because this chimney is so short, it is difficult to interpret it with confidence; however, the fact that it is still present in 2020 strengthens the assumption that this is another vertical chimney. In map view it appears as a circular low amplitude anomaly in Layer 5 (Fig. 5e – black arrows in 2008 and 2020).

There can also be other, more subtle chimneys in the dataset, where CO₂ is likely migrating more gradually vertically into shallower layers. If these features are not set in a background of alternating high and low amplitudes, then it is very difficult to observe them on cross sections (as shown by the dim signature in the modelling for the thicker layers in Fig. 8); however, they can be identified by anomalies which start to develop in locations offset from the main anomalies. Such offset anomalies could either be caused by lateral migration below the detection limit or they could be caused by feeder pathways from deeper layers. Each case needs to be considered individually to determine the most probable migration mechanism.

The first observation of an anomaly clearly developing offset from the main chimney was in layer 6 (Fig. 5d – 1999). Given the far distance from the main chimney, and the early observation after injection start, we are confident that this anomaly is linked to CO₂ being charged by a separate chimney. Note how this layer lies above a pooling of CO₂ near the NE edge of layer 5, which in turn lies above a separate anomaly in layer 4 (black arrows in Fig. 5e and f – 1999). This area is further underlain by layers 3, 2, and 1. We believe that layers 6 and 4 are fed by their respective underlying layers in this location, and it is possible that a longer chimney going all the way from layer 1 to layer 6 is feeding all the intermediate layers (dotted black arrows in Fig. 9), potentially in combination with some lateral CO₂ migration within each layer.

Figure 10 shows the analysis of an anomaly south of layer 8 which was first observed in 2006 (Fig. 5b – 2008 and 2020). This anomaly was underlain by both layers 5 and 7 at the time. We also interpret this anomaly to be a result of CO₂ migration from deeper layers, as indicated by the dotted black arrows (Fig. 10). Given that the Top Utsira Fm. surface has a clear depression between the two anomalies and because the two anomalies have so far not merged, we are confident that this is another example of vertical migration.

The final example of vertical CO₂ migration is the western anomaly in the Sand Wedge, first observed for layer 9 in 2020 (Fig. 5a – 2020). This anomaly is clearly related to CO₂ migration in underlying layers 5, 6, and 8, as illustrated in Figures 11 and 12. Layer 5 started developing towards the west already in 2013 (not shown) and the expansion continued in 2016 and 2020 (Fig. 11b, c). Layers 6 and 8 appear to either be fed from below and/or develop laterally (Fig. 12a). Finally in the 2020 dataset, the small anomaly located in the Sand Wedge is observed to lie above the layer 8 anomaly which indicates that the CO₂ has migrated obliquely from Top Utsira Fm (Fig. 12b). This behaviour is quite different from any of the previously described observations and therefore could be a combination of vertical and lateral migration. This anomaly notably occurs in an area of disturbed amplitudes on the base line seismic associated with strong amplitudes in the overburden (Fig. 11a). This unusual migration pathway may be caused by creation of sandstone-to-sandstone connections driven by sedimentary processes like soft sediment remobilization or erosion followed by sandstone amalgamation. The latter would allow fluid migration and bypass through a predominantly layered stratigraphic framework.

Mainly on the basis of seismic and biostratigraphic data (very limited core have been acquired) several interpretations have been proposed for the depositional environment of Utsira Fm., typically evoking either a shallow or deep marine origin, with sediments sourced from East Shetland platform (Gregersen et al. 1997; Galloway 2002; Rundberg and Eidvin 2005; Gregersen and Johannessen 2007; Arts et al. 2008; Williams and Chadwick 2021). Our observations and results of selected, previous studies (e.g. Gregersen et al. 1997; Piasecki et al. 2002; Williams and Chadwick 2021) suggest that deposits of Utsira Fm. in the vicinity of the Sleipner storage site represent a confined submarine fan deposited in an open marine, relatively deep-water setting, based on the following observations: (1) the distance from Sleipner injection area to the contemporary shelf (>50 km), (2) the stratigraphic location at the base of steeply SE to east dipping basin margin reflectors, (3) the Utsira Fm. being overlain by 100 m high prograding clinoforms, (4) the onlapping of the Utsira Fm. towards the basin margins, (5) the predominantly open marine biofacies, (6) the seismic geomorphology suggesting some roughly NS trending channels, and (7) the mounded and onlapping nature of seismic reflections corresponding with blocky wireline-log patterns.

In the Sleipner area an isochron map of the Sand Wedge provides evidence of a major channel system leading towards the south and feeding a smaller distributary channel network forming a channelized lobe (Fig. 13). Also, CO₂ migration observations from anomaly 9 (Fig. 5a – 2008) indicate two NS directed channel-forms, suggesting that the system imaged within the Sand Wedge consists of several discrete geo-bodies with distinct depositional geometries.

A similar pattern of plume migration can be observed in layer 6, 4 and 2 outlining elongated and slightly sinuous/banded CO₂ amplitudes (Fig. 5d, f and h). This suggests that channel-geometries are likely not exclusive to the Sand Wedge, but that a large part of the Utsira Fm. is channelized, and the organization of reservoir elements is rather complex due to the likely presence of several levels of channel hierarchy (cf. Williams and Chadwick 2021).

Depositional interpretations of layers 8, 7, 5, 3, and 1 (refer Fig. 5b, c, e, g and i) are more ambiguous due to the irregular nature of the CO₂ reflections. This is likely due to CO₂ reaching spatially isolated architectural elements resulting from multiple episodes of cut-and-fill deposition and the sub-cropping of older channel-fills beneath younger channels. Alternatively, CO₂ could be filling channel-attached sheet-like elements. The sand-to-sand contacts are likely due to erosion and amalgamation between channel sands or driven by sediment remobilization and vertical sand injection.

We interpret these CO₂ layers as developing in a fill-spill scenario, in a three-stage process (Fig. 14):

• Delivery of CO₂ from deeper layers directly via vertical corridors and/or through lateral migration.

• Trapping/back-filling underneath mudstone baffles.

• Further escaping of CO₂ to shallower layers within the Utsira storage complex through breaks within the mudstones or around the mudstone fringes.

The observations on the development of the CO₂ plume and detailed seismic mapping of the reflections have highlighted depositional geometries that suggest an intricate interplay between channel erosion and sediment deposition which resulted in preservation of remnant sandstone geo-bodies and mudstone baffles that controlled trapping and CO₂ migration patterns in all layers. CO₂ migration maps (Fig. 5) indicate a likely highly channelized setting, which together with the isochron map of the Sand Wedge (Fig. 13) and known context suggest a turbiditic origin of the Utsira Fm. at Sleipner injection area. However, the final description and interpretation of the shape of the CO₂ amplitudes needs to account also for the overall structural trap alignment, which still has some uncertainties.

An important question related to CO₂ monitoring is to quantify how large a volume might go undetected. This is related both to conformance and containment monitoring but is mostly relevant for estimating possible emissions in the case of leakage. The detectability of CO₂ depends on a mixture of geological parameters (depth, porosity, saturation, layer thickness) and monitorability (technical solution, repeatability). For saline aquifers the smallest detectable volume is related to the contrast between brine and CO₂ saturated pore fluid; a contrast that decreases with depth and increases with increasing porosity and CO₂ saturation. Layer thickness constitutes a complex factor in this equation, as the contrast in seismic reflectivity between a CO₂ filled layer and the surrounding brine-filled medium is at maximum near seismic tuning, which depends on seismic frequency.

Seismic repeatability and detection confidence depend on the technical solution selected for time-lapse monitoring. Source and receiver position repeatability, source signature, geological complexity, and environmental noise all influence the seismic image. Notably coherent noise (e.g. related to suboptimal multiple removal or inadequate migration) might be misinterpreted as a time-lapse seismic signal. Both 2D and 3D seismic images could be used to detect CO₂, however having 3D data usually provides higher confidence in observed anomalies. This is particularly true if the goal is to detect potential minor leakages into the overburden.

The time-lapse seismic monitoring at Sleipner has demonstrated how a relatively shallow storage site can retain CO₂ in the subsurface. At 800–1000 m injection depth, CO₂ injection into the Utsira Fm. at Sleipner is at the limit of what is considered practical for injection purposes, because CO₂ at shallower depths would transform into gaseous phase and take up a significantly larger volume due to the lower pressures and temperatures at these depths (Span and Wagner 1996). At Sleipner, the high porosity reservoir combined with the acoustic properties of the CO₂ form a strong contrast to the properties of the in situ aquifer, promoting a detailed seismic mapping of the CO₂. We can therefore regard Sleipner as a proxy for the overburden for most other (deeper) sites.

Layer 9 is believed to be the last layer to be charged with CO₂ because the injection happened at a depth around 150 m deeper than the Sand Wedge_. Several attempts have been made at estimating the CO₂ layer thickness at layer 9 in 1999, and it is generally agreed that detectability is 1 m, with thickness increasing up to 3 m towards the middle of the layer (Chadwick et al. 2004; Kiær 2015). We have not been able to find documented examples of smallest detectable layer thicknesses from the literature on CO₂ injection at other sites; however, estimates of detectable volumes are more easily found (Table 3). This motivated us to try to estimate the lowest detectable volume observed at Sleipner.

One of the most notable results from the time-lapse monitoring of the CO₂ plume at Sleipner has been to demonstrate how thin mudstones have contributed to spreading the CO₂ in a reservoir with excellent porosity, permeability and net-to-gross (Chadwick et al. 2004; Boait et al. 2012, refer also to Fig. 4b–d, and all the mapped layers in Fig. 5). When the CO₂ is distributed beneath separate thin mudstones in such a manner, it has the potential to facilitate increased CO₂ dissolution because a larger portion of the CO₂ would be exposed to brine, as shown in laboratory experiments comparing homogeneous and heterogeneous trapping media (Trevisan et al. 2017). This has direct implications for optimizing injection strategy towards injecting at deeper targets, increasing storage capacity beyond simple structural traps.

There is significant uncertainty with respect to actual CO₂ dissolution over time. Estimations from natural analogues range from 10–50% being trapped on geological timescales and with rates decreasing over time (Leslie et al. 2022). Recent simulations for Sleipner based on input from gravity monitoring indicate 10–15% being trapped on an intermediate time scale (Ringrose et al. 2021). The CO₂ not trapped within structural traps or trapped by dissolution or capillary forces is then expected to migrate towards the top of the structure due to gravitational forces (Lindeberg et al. 2000). The time-lapse seismic data at Sleipner seem to confirm these predictions, with a clear tendency that an increasing fraction of CO₂ has been structurally trapped in layers 8 and 9 over time (Fig. 16).

Globally, there has recently been an increased focus on implementing CCS, both onshore and offshore. In Europe several new CO₂ injection projects are in the pipeline, with the Norwegian Longship project leading the development. Longship is expected to start injecting CO₂ in 2024, as the world's first large scale full value chain project. The UK, Denmark, the Netherlands, and Norway have all recently announced offshore CO₂ storage licenses and several nations have initiated financing of new CO₂ storage sites, for research or full-scale implementation. Reliable and cost effective MMV plans (Measurement, Monitoring, Verification) are an important factor in the licence to operate for these projects.

Worldwide, most CO₂ storage projects (including CO₂ injection as part of enhanced oil recovery projects) are located onshore, whereas these new European projects are mostly offshore, in the North Sea, Barents Sea or Irish Sea. Each CO₂ storage site is unique, both with respect to location (onshore, offshore, depth), and storage capacity and containment risks (which are both dependent on geological conditions). However, some general insights can still be drawn.

The Sleipner CO₂ injection site has been very useful for testing and maturing technologies for offshore CO₂ monitoring. Subsurface CO₂ injection monitoring requires sufficient flexibility to change and extend monitoring targets, because injection strategies are likely to change over time, both as a response to subsurface knowledge maturation and fluctuating CO₂ deliveries. The extension of the repeat seismic survey over Sleipner acquired in 2020 is an example of such flexibility. In addition, subsurface CO₂ injection monitoring also requires a combination of relatively high resolution imaging in the shallow section as well as adequate resolution for deeper targets. Short cable technologies (such as the XHR tests conducted over Sleipner in 2020–21) might provide a cost-efficient monitoring option for monitoring the shallow subsurface.

One important learning is that the baseline seismic survey should be designed to include a minimum area covering all predicted CO₂ plume migration routes. It should over time also be possible to extend this baseline to accommodate potential unexpected changes in the injection program or as a response to non-conformance or non-containment.

Different injection sites might require different monitoring programs. Streamer seismic might be better suited for monitoring as compared with permanent seabed seismic sensors for the case of large-scale projects, or projects where the final CO₂ plume is expected to migrate large distances. This is because compared to permanent seismic it is relatively easy and cost-efficient to modify the monitoring program for streamer seismic surveys. In other locations permanent seismic monitoring methods might be better suited, as they have advantages in high repeatability, offering potentially higher resolution and quicker turn-around time. For the Sleipner site the streamer seismic acquisition programme has proved sufficient; however, for new CO₂ injection sites an optimal monitoring program should be designed, depending on geological conditions, injection volumes and local risk assessments, and taking into account new technology developments such as fibre optic sensing.

The ongoing CO₂ Sleipner injection project has clearly demonstrated that offshore CO₂ storage is feasible and monitorable. The successful repeat marine-streamer seismic monitoring programme at Sleipner has given confidence that seismic is a well-suited subsurface monitoring technology both for conformance and containment assurance. Gravimetric monitoring, although providing an independent measure of subsurface density changes, is not a sufficient independent monitoring tool due to the associated depth and resolution limitations.

A total of 10 seismic repeat surveys have consistently shown that there are no indications of CO₂ migrating into the overburden over the Sleipner site. Layer thicknesses down to 1 m can be detected from time-lapse seismic, and volumes down to 10.5 kt of CO₂ were detectable even on the first seismic repeat survey. This is in accordance with insights on detected volumes at other CO₂ injection sites and provides confidence that we would be able to detect such layer thicknesses and corresponding volumes if the CO₂ started migrating into the overburden.

Seismic time-lapse monitoring at Sleipner has revealed a complex migration pattern even in the seemingly homogenous Utsira Fm. Sandstone. This is due to a complex set of reservoir geometries caused by the original sandstone depositional architecture, with modifications by channel erosion, and preservation of mudstone drapes as well as structural trap alignment. These thin internal clay-rich barriers act to distribute the CO₂ within the reservoir (by structural trapping), which then promotes the effects of residual trapping and CO₂ dissolution into the in situ brine.

L046 licence partners Equinor Energy, Vår Energi, LOTOS Norge and KUFPEC Norway are acknowledged for allowing us to publish these data. SACS and CO₂REMOVE consortia contributed to financing some of the early acquisition and processing.

The owners are also acknowledged for making all time-lapse surveys up to and including 2010 available to the public through the CO₂DataShare (https://CO2datashare.org/), managed by SINTEF.

Njål Solberg Greiner and Muna Hassan Mohamoud Haid are acknowledged for assisting with seismic interpretations. We are grateful for the comments of two anonymous reviewers and to Philip Ringrose for proofreading the manuscript.

A-KF: conceptualization (lead), formal analysis (lead), investigation (lead), methodology (lead), visualization (lead), writing – original draft (lead), writing – review & editing (lead); MJW: conceptualization (equal), visualization (supporting), writing – original draft (supporting), writing – review & editing (supporting); HA: visualization (supporting), writing – original draft (supporting), writing – review & editing (supporting); ASMP: conceptualization (supporting), writing – original draft (supporting), writing – review & editing (supporting)

This work was funded by the Equinor (Not Applicable).

All authors are employed by Equinor ASA.

The seismic datasets used in this study acquired in or before 2010 are available from the CO₂ Datashare Portal, operated by Sintef: https://co2datashare.org/.