Long-term risk assessment of subsurface carbon storage: analogues, workflows and quantification Open Access

The application of oil and gas subsurface and engineering expertise to carbon capture and storage (CCS) is a vital component of all the energy transition projects that plan to sequester CO₂ (Bowden and Rigg 2004; Underhill 2022).

This paper is concerned with the permanent geological storage of CO₂ and does not address the ‘capture’ or transport aspect of CCS; therefore, we will adopt the term subsurface CO₂ storage (SCS) when referring to subsurface assessment.

The objective of this paper is to analyse learnings from selected past and current injection projects, and to summarize key elements of risk assessment frameworks utilized in oil and gas projects, noting differences required for their application to SCS risk assessment. Oil and gas production projects require management of declining reservoir pressure over time, with steps taken to maintain pressure over a typical field life of 20–50 years. During the injection phase of SCS projects, reservoir pressure tends to increase, requiring management to avoid excessive pressure build-up, until injection stops, when pressures will dissipate over a period of hundreds of years. Therefore, a different approach is required, and this paper highlights existing risk assessment methodologies developed for SCS and proposes an extension to derive a quantitative risk profile that could be utilized within this framework.

Practical and nearly universal application of chance and volume assessment in hydrocarbon exploration has been developed over time by many practitioners – it is the intent of this paper to contribute to a similar effort by all concerned with safe and effective storage of CO₂. The methods to assess and calibrate the range of hydrocarbon volumes and chance of success (risk) have been established and refined over at least the past five decades, as exemplified by Newendorp (1975), Cozzolino (1977), Megill (1984), Rose (1987), MacKay (1995), Citron and Rose (2000), Rose (2001), Yielding et al. (2010), Roden et al. (2014), Milkov (2015) and Edmundson et al. (2021). A variety of schemes have been proposed for the evaluation of risk in SCS projects, including Bowden and Rigg (2004, 2005), Maul et al. (2007), IEAGHG (2009), Oldenburg et al. (2009), Stauffer et al. (2009), Gerstenberger et al. (2013), Metcalfe et al. (2013), Bourne et al. (2014), Espie and Woods (2014), Watson (2014), Pawar et al. (2015), Alcalde et al. (2018) Lackey et al. (2019), Bump et al. (2021), Vasylkivska et al. (2021) and Wei et al. (2022).

The objective of SCS projects is permanent storage of CO₂. This requires a framework to assess both uncertainties and risks over hundreds to thousands of years. The assessments will cover site identification and appraisal of the storage complex, injection of CO₂, pressure monitoring, and post-injection monitoring to ensure long-term retainment. This very long-term risk assessment requires additional skills and methodologies that are not generally required for oil and gas projects.

Therefore, we propose that a paradigm shift is required from the focus on geological and subsurface success that is common practice in hydrocarbon exploration and appraisal programmes, to a broader assessment of the chance of failure throughout the life of a project. Operators will be required to develop plans for all stages of an SCS project, and that will include monitoring and mitigating events that could lead to project shutdown.

A paradigm shift is also required when considering failures. Exploration and production (E&P) companies are able to plug and abandon dry holes and walk away. If a CO₂ injector is abandoned because the injectant ended up in the wrong formation, there could be a legal liability. If the CO₂ plume or pressure front moves beyond the licence area it could impact nearby subsurface operations. The CO₂ could also leak upwards into a shallow aquifer or to surface, resulting in contamination or injury.

Unplanned events affecting one CCS project have the potential to impact not only the operator and the project but also the perception of the global emerging industry of CCS. For example, Marshall (2022) recently provided a critical academic analysis of the Gorgon CCS project in Western Australia. Among Marshall's conclusions (Marshall 2022, p. 17) is the following statement:

There is no indication that the Gorgon project, even if it is fully successful, will reduce the emissions from the fossil fuels it excavates and sells, and given the problems it faced, it seems unlikely that storing a significant amount of emissions produced by burning would be possible.

Opinions such as these highlight the challenge facing the emerging CCS industry in gaining and maintaining the social licence to operate (SLO) throughout a project. A rigorous risk assessment is a key component of managing long-running and complex projects, and demonstrating rigorous due diligence.

Oil and gas geoscientists and reservoir engineers apply their skills in assessing the volumes and geological chance of success for prospects. However, during exploration, appraisal and development programmes, companies accept that there will be failures due to dry holes, non-commercial discoveries and wells that produce far less than predicted, which result in an economic burden on the company.

Drilling an increasing number of independent or negatively correlated prospects can create the portfolio effect that ensures that the value of discovered volumes exceeds the programme costs. A model portfolio presented in Figure 1 demonstrates that as the number of wells drilled increases, the chance of zero geological and zero commercial discoveries decreases. Conversely, the chance of achieving a net present value (NPV) greater than zero increases but is not certain (Fig. 1).

Examples of portfolio analysis, portfolio management and many of the underlying techniques have been exemplified by Walls (2004), Willigers (2009), Back and Kirk (2012), Willigers et al. (2013), Newendorp and Schuyler (2015), MacKay et al. (2016), Al-Adwani et al. (2017), Zhukov et al. (2018) and LaCosta and Milkov (2022). Commercial and in-house exploration portfolio software is widely used in E&P companies.

We contend that this ‘portfolio effect’, accepting a certain number of project failures, will not be acceptable to companies, regulators or the broader societal interests in geological carbon storage projects. Failure to inject the contracted CO₂ volumes or contain CO₂ in the target reservoir or storage zone could result in significant mitigation actions or even project shutdown. The potential cost related to being unable to inject the contracted amount of CO₂ is illustrated by the Gorgon project, where the operator has had to acquire and surrender carbon credits. These are estimated to have cost the Gorgon joint venture between US$100 million and US$184 million (Robertson and Mousavain 2022).

At the extreme, if CO₂ is not contained in the intended reservoir, there is the potential for environmental damage or loss of life. A complete assessment of the subsurface, infrastructure and project risks is a prerequisite for project design and for planning effective mitigation and corrective measures.

In this paper, storage volume is defined as the product of net pore volume and storage efficiency. Storage capacity, a term commonly used in SCS work, is the product of storage volume and CO₂ density. Storage capacity is not used in this paper in order to avoid confusion with the term ‘capacity in the storage resources management system’ (SRMS: SPE 2017), which refers to those quantities of CO₂ anticipated to be commercially stored.

Risk, uncertainty, chance, success and failure are familiar terms that are regularly used in the course of daily work in the oil and gas industry. However, there are significant differences between individuals regarding the definition and understanding of these terms, which makes it important to clarify them.

Uncertainty refers to the range of possible outcomes, such as the range of values for reservoir thickness, injectivity or storage volume. Risk is the future threat of harm or loss and can be expressed in terms of negative outcomes including injury, damage, financial and reputational loss. Kloman (1992, p. 305) defined risk management as follows: ‘Risk management is a discipline for living with the possibility that future events may cause harm’.

For low-frequency events, a forecast that nothing will happen is likely to be correct most of the time, except when it is not. Ignoring such low-frequency events means that early indicators might be misinterpreted or dismissed, with no preparations or mitigations put in place. Gordon Woo, author of Calculating Catastrophe (Woo 2011, p. 197), includes the following observation about null forecasts:

Yet quality in a forecast is not about being correct most of the time. This is because for rare events, one can be correct most of the time with a simple null forecast – never saying an event will happen.

Risk matrices have been widely used to provide a semi-quantitative view of likelihood and consequence. Figure 2 presents an example risk matrix from a report relating to the Port of Rotterdam CO₂ transport hub and offshore storage (Porthos) project offshore Netherlands (Neele et al. 2019). The lowermost left cell, A-5, ‘Very low likelihood/very large amount of CO₂ migrates out of the reservoir and storage complex’, is designated as a medium risk level, coloured orange in this version. The ‘monitoring necessity’ and ‘risk reduction’ actions are set to as low as reasonably practicable (ALARP).

Thomas et al. (2014) illustrate the ‘flaws and dangers resulting from the use of risk matrices (RMs)’ (p. 56). Figure 3 illustrates part of the problem set they identified. A blowout is a remote event of catastrophic consequence and falls in the yellow ‘monitor, reduce if possible’ category, equivalent to ALARP (Fig. 3a). The problem is that the consequences of a blowout or failure in this part of the risk matrix can be orders of magnitude larger than smaller events (Fig. 3b). These authors conclude that, for this and the other reasons they detail, that risk matrices ‘should not be used for decisions of any consequence’ (p. 63).

Therefore, the recommended approach we develop in this paper is to evaluate the chance of an event separately from its consequences or impact.

There are two dimensions of risk and uncertainty that must be considered in SCS projects. Epistemic probabilities and uncertainties are those that can be determined and refined by increasing knowledge of the underlying physical properties of the system. In contrast, aleatory probabilities and uncertainties are inherently random. Although aleatory probabilities can be observed and estimated, they cannot be reduced by technical work (Woo 1999, p. 74). Earthquake prediction provides an example of aleatory probability. Aagaard et al. (2016), in a United States Geological Survey Fact Sheet on earthquake prediction in the San Francisco area, stated that there is a ‘72% probability of one or more M ≥ 6.7 earthquakes from 2014 to 2043’. For SCS projects, the team may forecast an annual frequency that seismically mapped faults, or inferred faults, may slip as injection proceeds. This will be an aleatory probability forecast.

The epistemic probabilities and uncertainties of SCS projects may be evaluated by using increasingly sophisticated reservoir and system data and models. These factors can be reduced by additional data gathering and improved analysis; for example, the range of the average porosity of a target reservoir can be narrowed by the careful analysis of additional wells. Frequency data from relevant analogues can inform the estimation of aleatory probabilities and uncertainties. Alcalde et al. (2018) provide extensive frequency data on many aspects of CO₂ storage in the supplementary dataset to their paper. IOGP (2019a) provides extensive well failure rates from representative drilling areas.

This duality means that while deterministic reservoir models are necessary, they are not sufficient to fully describe risk in SCS projects. Dubois (2010) shows that parameter variability in models can arise from both natural heterogeneity (aleatory) and from a lack of knowledge (epistemic). These two types of uncertainty should be considered separately and clearly documented. The combination of aleatory and epistemic uncertainty was applied to a model of CO₂ plume extension in geological storage by Bellenfant et al. (2009). They proposed weighting optimistic and pessimistic forecasts to describe the overall uncertainty to decision-makers.

Success is generally defined as accomplishing a favourable or desired outcome and is a familiar concept in the evaluation of hydrocarbon prospects and projects through the use of parameters such as ‘chance of discovery,’ or ‘chance of development’. In the context of SCS projects, failure is a project outcome resulting from foreseen or unforeseen event(s) that cause significant financial, reputational or other losses, and which may require project cessation before objectives are met.

We contend that in order to understand the probability of success for a carbon storage project, the question ‘how could the project fail?’ is more relevant than the opposite question, ‘how can the project be successful?’ A focus on successful analogues, the use of heuristics (rules of thumb) and our intuition makes us more vulnerable to biases that can skew our assessments. Biases occur in all human activities (Bratvold and Begg 2010), and the oil and gas industry is no exception (Welsh et al. 2005; Sykes et al. 2011). The cost of biased decisions can be significant (Welsh and Begg 2007).

Some of the more common biases observed include: anchoring (relying too heavily on a single analogue or dataset); confirmation (searching for and interpreting data that confirm our beliefs); overconfidence (overestimating the accuracy of one's analysis or interpretation); and motivation (actions driven by the desire for a particular outcome). As an example of their impact, Kahneman and Tversky (1982) showed that in complex systems that require a series of events to succeed, the anchoring bias can lead to an underestimation of the probability of failure. Even if the individual probability of failure for each event is low, the aggregated chance of failure may be significant. Focusing on potential failure modes and applying a risk assessment workflow suitable for evaluating long-term, low-frequency events can address some of these biases.

This also highlights the importance of evaluating injection projects that experienced significant unforeseen events. An understanding of what caused these events provides important lessons that should be applied when planning for future SCS projects.

An event is a material occurrence or change in a particular set of circumstances (ISO 2017), whereas a hazard is any source of potential damage, harm, or adverse health or other effects on something or someone (CCOHS 2022). Therefore, events, if they occur, may or may not be hazards. For example, a single micro-seismic event of less than 2 on the Richter scale is very unlikely to be a hazard. However, swarms of these small events can and do bring reputational and operational damage, and therefore can become real hazards to project continuation. A contemporary example in The Netherlands is described by Muntendam-Bos et al. (2022).

Subsurface characterization is critical for any SCS project, and, in particular, the need to quantify accurately the storage volume, injectivity and containment potential of the target reservoir, and the associated chance of failure. An accurate and comprehensive understanding of the project subsurface is a prerequisite to the assessment of project risk. Without this work, early indicators of later events, which could result in project failure, may be missed.

To help to ensure all potential risks and uncertainties are assessed, past projects should be carefully studied, particularly those that experienced unexpected events. While large-scale SCS is relatively new, the process of storing injected fluids underground is not. Therefore, the history of these operations provides insights into the types and frequency of unexpected events and failure incidents. Analysis of these past projects can illuminate the difference between aleatory and epistemic risks and uncertainties, and provide valuable lessons for future projects.

This paper presents several examples of injection project events, illustrating what happened, the factors responsible and whether these events could have been foreseen. After all, ‘if it happened, you must admit it was possible’ (Megill 1984, p. 146). We then propose a quantitative risk assessment framework suitable for evaluating long-term, low-frequency events that can inform monitoring, mitigation and remediation plans to help to ensure a successful project outcome.

Fluids have been injected into underground reservoirs for temporary or permanent storage (permanent storage is also referred to as sequestration) for more than a century. The first natural gas storage operations were undertaken in 1915 at a gas field in Canada (Evans 2009). By the 1920s, the underground injection of water to increase oil recovery rates had been successfully demonstrated in Pennsylvania and New York (Umpleby 1925). The widespread use of injection wells to dispose of produced water began in the 1930s (US EPA 2022). The history of these operations provides insights into the types and frequency of unexpected events and failure incidents.

Overall, underground fuel storage projects have an impressive safety track record but there is also a catalogue of failures involving financial or property loss, closure and, in a few cases, casualties or evacuations. A review of 228 reported events and failures worldwide (Evans 2009) showed that most failures are associated with leaking wellbores or surface facilities. In many cases, these projects are using wells that are decades old, providing a sense of what could happen if appropriate mitigation actions are not taken, 40 or 50 years from now when carbon storage projects will be reaching the end of their injection lives.

The most spectacular example of a leakage failure was in 2015 in Aliso Canyon near Los Angeles, where a natural gas storage well blew out for 100 days emitting 100 kt of methane and other substances into the air. This forced thousands of people from their homes, and in 2021 Southern California Gas agreed to pay US$1.8 billion to individuals and businesses to settle their claims (Penn 2021).

Geological factors are responsible for a small but significant fraction of injection project events. These include but are not limited to: (1) induced seismicity; (2) injected fluids spilling out of the intended container; (3) migration of injected fluids along fractures or across faults previously considered sealing; (4) the misidentification of a storage reservoir; (5) a lower than expected seal capacity of the cap rock; (6) migration of fluids away from the injector in an unanticipated direction; (7) an overestimation of the storage volume; (8) an underestimation of the reservoir heterogeneity, which may actually have increased the storage volumes, as in the Sleipner project discussed below; and (9) pressure maintenance issues. Figure 4 schematically summarizes these and other factors. Examples are discussed below.

In order to learn from past projects and to illustrate the negative and positive impacts of unexpected events caused by subsurface fluid injection, a series of examples are presented below. The examples include methane, water and CO₂ injection projects.

The Castor project (offshore eastern Spain) involved a depleted oilfield (Amposta Field) selected for temporary natural gas storage. The reservoir consists of fractured and karstified dolomitic limestone. The plan was for the reservoir to store 1.3 Bcm (billion cubic metres) of natural gas, sufficient to meet 25% of Spain's storage requirements (Foulger et al. 2018). This was based on studies that concluded the reservoir characteristics, especially the distribution of porosity, and the sealing capability of the cap rock were sufficient for this project (Batchelor et al. 2007). This work did not consider the possibility of induced seismicity, even though there had been no previous injection in the field because the Amposta Field had been produced under a natural water drive.

In 2013, just 3 days after gas injection began, a series of earthquakes occurred along a deeper, previously unidentified fault (Fig. 5). This culminated in three earthquakes with a magnitude M_w > 4 (Cesca et al. 2021). Injection was halted, and the site was permanently closed 6 years later at an estimated total cost to Spanish citizens of up to €4.7 billion (The Corner 2017). A subsequent analysis showed that assessing fault stability prior to gas injection would have identified the risk of induced seismicity (Vilarrasa et al. 2021).

The Huntsman Field (Nebraska, USA) produced 28 Bcf (billion cubic ft) of gas prior to conversion to gas storage in 1963. In 1968, the West Engelland gas field was discovered adjacent to the Huntsman Field. It had a much lower reservoir pressure than the initial pressure at the Huntsman Field, which implied some type of pressure connection (Fig. 6). The West Engelland Field was then developed, and by the early 1980s had produced five times more gas than the original gas-in-place. Meanwhile the Huntsman gas storage field was experiencing some significant unexplained reductions in reservoir pressure, which were eventually attributed to tilting of the gas–water contact, allowing gas to move from the Huntsman Field to the West Engelland Field. Not surprisingly, this led to a lawsuit and an out-of-court settlement in 1985 to compensate the operator for the storage gas that was siphoned from the Huntsman Field (Tek et al. 1986).

The Wilmington oilfield (California, USA) is partly located offshore. When this field was developed in the late 1960s from islands constructed in Long Beach Harbor, more than 20 injection wells were drilled into the aquifer to create a peripheral waterflood (Fig. 7). The injected volume exceeded 100 000 bpd (barrels per day). This water was not expected to move downdip because of an interpreted sealing fault in the syncline between the Wilmington Field and the adjacent Seal Beach Field. However, the fault turned out to be non-sealing and the water moved through the syncline, creating a waterflood at Seal Beach. This led to an increase in the oil production rate while considerably increasing the reservoir pressure. Wells were flowing with a surface pressure of 200–350 psi adjacent to an area where houses had been built above abandoned wells. In response, the Seal Beach operator contacted the Wilmington Field operator who proceeded to shut-in all the peripheral waterflood injectors for about 6 months. This allowed the pressure to decline in the aquifer before resuming injection at a reduced rate of 30 000 bpd. In addition, the Seal Beach operator installed pressure gauges in idle wells to monitor the reservoir pressure and to help ensure that no future blowouts would occur (Allen 1977).

In the Tordis Field in the Norwegian sector of the North Sea, oily water was injected for disposal into a shallow aquifer thought to be the Utsira sands, a well-known, good-quality reservoir. However, a miscorrelation incorrectly identified the intended storage reservoir (the Utsira sands were later found not to be present in the area of the Tordis Field due to a pinchout). Instead, the water was injected into a sand lens within the overlying Nordland Group (Fig. 8). Because this lens had a limited volume, the reservoir pressure increased rapidly during injection and the seal was breached, resulting in fluid escape and the leakage of oily water upwards to the seafloor. This created a crater 30–40 m across and 7 m deep on the seabed (Eidvin 2009; Eidvin and Øverland 2012). In the wake of this, because the Utsira sands are the CO₂ storage reservoir at the Sleipner carbon storage project, environmental organizations began questioning the security of storing CO₂ in the Utsira sands, which demonstrates how a failure of one project can implicate others (Bjureby et al. 2009).

The In Salah carbon storage project (Algeria) provides an example of unexpected events leading to the early termination of a project. At In Salah, the injection of CO₂ took place in three wells located downdip of the partly depleted Krechba gas field (Fig. 9). Injection in these wells: (1) generated an uplift of up to 2.5 cm at the surface as of 2010 (shown in Fig. 9 as red-coloured areas surrounding the injector wells); (2) created fractures that might compromise the top seal; and (3) formed a flow path between the KB-502 injector well and the KB-5 observation well (White et al. 2014).

The surface deformation shown in Figure 9 was used as a means to estimate the extent of the CO₂ plume. It had been anticipated that the gas would migrate southwestwards into the updip structural closure. However, the CO₂ fluid migrated northwards through a saddle and towards another closure, moving beyond the block boundary, which was identified as the main project risk (Dodds et al. 2011). Multiple revisions to the risk assessment were undertaken and the decision was made to halt injection after 7 years, based on concerns regarding top-seal integrity. Between 2004 and 2011, a total of 3.8 Mt of CO₂ was injected (Ringrose et al. 2013; White et al. 2014), which was only 22% of the original injection target of 17 Mt (Haddadji 2006). This project highlights the importance of an effective monitoring programme to ensure early identification of significant events and rapid implementation of mitigation actions.

The Snøhvit Field, located in the Norwegian Barents Sea, is an example of a SCS project with built-in optionality. It contains two reservoirs into which CO₂ can be injected. The field produces gas condensate with 5–8% CO₂ from the Middle Jurassic Stø Formation. The CO₂ is separated and injected back into the subsurface. Initially, this CO₂ was injected into the Tubåen Formation, which is an underlying sandstone aquifer in the F Segment of the field (Fig. 10). However, pressures in the Tubåen Formation increased more rapidly than planned due to the presence of subseismic faults and lenticular sandbodies. This led to the early termination of injection into the Tubåen Formation (Hansen et al. 2013; Kaufmann and Skurtveit 2018).

Early risk assessment had anticipated this scenario and, as a result of advance planning, injection was switched to the water-bearing Stø Formation in the same fault block. Injection pressure declined and stabilized (Pawar et al. 2015). As of 2019, 6.5 Mt had been injected: 1.1 Mt into the Tubåen Formation and the balance into the Stø Formation (Ringrose and Sæther 2020).

The Sleipner project, with an injection history of more than 25 years and a total volume of over 20 Mt of CO₂ injected, is a prime example of a successful SCS project. However, even this project experienced unexpected events (e.g. Chadwick et al. 2004). Pre-injection analysis of reservoir heterogeneity from well data identified multiple 1 m-thick shales within a 300 m-thick reservoir sequence (Zweigel et al. 2004). As injection proceeded, the shales acted as baffles, limiting the upward movement of CO₂ from the injection point to the top of the reservoir. Instead of accumulating below the top seal as expected, a substantial portion of the injected CO₂ became trapped beneath the baffles and migrated laterally (Fig. 11). This is benefiting the project, however, as it results in a larger storage volume and a lower buoyancy pressure against the cap rock (Cowton et al. 2018; Ringrose 2018).

The Gorgon SCS project on Australia's Northwest Shelf began injection in July 2020 and is one of the largest active projects, with a nominal injection capacity of 4 Mt a⁻¹. A plant on Barrow Island receives feed gas from the Gorgon and Io/Jansz fields and separates the CO₂ from natural gas. The CO₂ is then injected into sandstones of the Lower Dupuy Formation at a depth of c. 2500 m (Fig. 12). In order to reduce pressure build-up in the storage reservoir, brine is extracted from the Lower Dupuy Formation and reinjected into shallower sands of the Barrow Formation (Trupp et al. 2013). Unfortunately, sand influx into the brine reinjection wells resulted in a shortfall in injected water volumes and, hence, in the volumes of brine extracted from the storage reservoir. This, in turn, increased the reservoir pressure in the Lower Dupuy Formation, limiting CO₂ injectivity and prompting the regulator to impose an injection rate limit to mitigate additional risks. Consequently, Gorgon SCS project has not been able to meet its contractual obligations. The project had to acquire and surrender carbon credits to offset its target shortfall of 5.23 Mt of CO₂ at an estimated cost in excess of US$100 million (Robertson and Mousavain 2022).

Whilst some of the events mentioned above had been anticipated, some were not. Building on these and other experiences, several observations can be made.

Any significant event occurring during an injection project should have been foreseen as a possibility. A thorough understanding of potential analogues, the allocation of sufficient time and tools to characterize projects, and the application of techniques to mitigate bias can help to ensure this happens.

Moreover, the chance of the event occurring can be quantified through additional data gathering and analysis, and the cost of impactful events can be reduced by pilot projects prior to committing to a full-scale project. This requires a strategy that does not unreasonably accelerate development and minimize expenditures.

Understanding the subsurface is fundamental for SCS projects, just as it is for hydrocarbon exploration. We begin with homogeneous and isotropic models, and then refine them to approximate the state of nature. In the absence of dynamic injection data, it is challenging to capture the true heterogeneity of the reservoir and its impact on CO₂ plume movement (Sleipner project), especially if natural fractures are involved (In Salah project). This has implications for how much fluid can be stored without pressuring-up the reservoir (Snøhvit project) or exceeding the fracture gradient and breaching containment (Tordis project). The thickness of a permeable reservoir (kh, where k is the permeability and h is the formation thickness) is key for achieving injectivity, which can be diminished by operational problems (Gorgon project) or halted by induced seismicity if geomechanical properties are unknown (Castor project). Once the CO₂ is injected, understanding the plume size, geometry and migration direction becomes critical. Injected fluids may impact offset accumulations (Wilmington and Huntsman projects) and pose significant liability, especially in areas where the pore space is privately owned.

These are not easy hurdles to overcome, and a concerted effort is needed to address them. Some of the techniques advocated are framing sessions, a staged approach, peer reviews, external audits and performance lookbacks. These promote a rigorous assessment of the uncertainties, which reveals the key risks, including those that are not immediately apparent. An understanding of what unexpected events could occur, and what their impacts could be, is crucial. Failure cases should be constructed and shared with the regulator and stakeholders. The use of probabilistic techniques is highly recommended in this process for quantifying project framing, analysis and risking.

While it is true that regulations in many countries are comprehensive in terms of what is required for permitting, in the end site-specific geological and project characteristics must guide appraisal and decisions regarding implementation. This means that the burden is on the operator to quantify and mitigate risks with data gathering, analysis, analogues, modelling, demonstration projects and monitoring. Failure to do so not only jeopardizes the commercial viability of the project but can also imperil the wider social licence if the impact of unforeseen events negatively affects the public.

Prospect risk assessment of subsurface oil and gas projects focuses on quantifying the characteristics of potential hydrocarbon accumulations and the chance that they will be discovered. For example, either the reservoir was deposited at the well location or not; the seal has effective capacity for the predicted column or not. Consistent risk evaluation of these elements is critical to management of a portfolio of exploration opportunities.

While initial subsurface risk assessment for SCS site characterization is like that of oil and gas prospect evaluations, a significant part of the full life-cycle risk assessment requires evaluation of the likelihood of future events occurring, or not, and their potential impacts. Risk assessment utilizing forecasts of dynamic data is required for both oil and gas production and SCS projects. Production models are continuously updated by history matching to actual well performance and, if available, 4D seismic reservoir monitoring. Initial SCS reservoir models, which may have limited data and no CO₂ injection results to use for calibration, are unlikely to provide an accurate view of reservoir behaviour, especially beyond the time frame for typical production operations. Accordingly, SCS reservoir models constructed before injection operations begin are a useful and necessary guide but are insufficient for a full risk assessment.

There are multiple established risk assessment techniques that have been adapted from other industries and applied to SCS projects. Many risk assessment methodologies, both qualitative and quantitative, have been published and demonstrated via case studies, and it is not the intention of this paper to provide a detailed review of these methodologies. Condor et al. (2011), Jewell and Senior (2012), Pawar et al. (2015) and Li and Liu (2016) provide thorough comparative analyses of several risk assessment methods. More recently, other authors have published methodologies and guidance including the International Standards Organization (ISO 2017), Alcalde et al. (2018) and SRMS (SPE 2017, 2022). Owing to the complexity, long time frames and uniqueness of each SCS project, no single existing published risk assessment technique will be universally suitable. Therefore, we suggest defining a robust framework that acknowledges learnings from past projects, allows for the utilization of the best combination of existing risk assessment methods for each project and which enables an understanding of project risk to evolve as the project progresses.

This paper proposes such a framework that can be used to determine the risk profile of a project through time, including an alternative method for quantification of a project risk profile and determining the period (or periods) of Peak Risk when the system is close to maximum stress and the risk of an event occurring is highest.

The general topic of risk management is well established, and Aven (2016) provides an overview of perspectives and approaches. The steps in a risk assessment commonly include setting the context, identifying and evaluating the risks, planning and implementing risk management strategies, and reviewing and reporting. Building on this established process, we incorporate the following key aspects in the proposed quantitative Peak Risk workflow:

1. project framing;

2. definition of a risking scale;

3. comprehensive event and hazard identification;

4. quantification of the frequency of long-term events; and

5. review and refinement of risks.

There are important lessons to be learned and applied from the management of oil and gas assets as well as past injection projects. Large amounts of capital have been wasted by not understanding what investments are required to realize the expected (average) technical requirements, and poor decision-making on when to continue or exit a project (Jenkins and McLane 2019).

Thorough framing of the project at the outset following a staged approach and establishing success criteria for each stage gate will assist in a disciplined approach to managing large and complex SCS projects.

A staged approach can address those aleatory risks that may impact an SCS project through incremental exposure of risk and capital as the project evolves. Careful consideration of the value of information (VOI) is required. A working definition of VOI is that the information obtained only has value if it can affect a decision. Some of the injection projects discussed above, such as the Castor project, would have benefited from a staged approach. If, prior to committing to development, an injectivity test would have been carried out (since no injectivity data were available for the Amposta Field), the induced seismicity issue might have come to light in time to halt the project at an early stage, thus avoiding the loss of capital spent on facilities and infrastructure. Clearly, such appraisal information would have added value. A front-end risk assessment should have identified there was no injectivity data available, prompting a discussion about how to address this data gap.

In general terms, a staged approach can be summarized in a decision tree (Fig. 13). For simplicity, we will assume four stages to aid discussion in this paper: Identification – to select a site and to characterize the storage complex (also called the screening or pre-injection stage); Appraisal (or site evaluation) – to evaluate the injectivity parameters; Injection – which tests the true storage effectiveness; and Monitoring – to ensure retainment.

In the site Identification stage, the volume estimate of the storage complex is key. The threshold is achieved if the volume is deemed sufficient to safely store the required amount of CO₂. This may be achieved without drilling a well if appropriate analogue or pass-through well information exists. The Appraisal stage drills at least one well to determine whether CO₂ can be injected at a sufficient rate to be commercially attractive and store the required CO₂. Core, wireline log and seismic data should be analysed to understand reservoir heterogeneity, especially lateral extent, vertical variations and compartmentalization. It may also be necessary to produce these wells to quantify brine production rates if wells are needed to reduce the reservoir pressure (as demonstrated by the Gorgon project). In the Injection stage, additional injector wells (and possibly production wells) are drilled to meet the total project storage requirements. In this stage, confidence in containment is crucial. The injected CO₂ must stay where intended without any significant upward leakage or migration outside the areal limit of the project. Injection also must not exceed the cap-rock closure stress or induce significant seismicity.

It is worth noting that, for commercial projects, there will always be a desire to shorten or even eliminate the time and money spent in the Identification (storage complex) and Appraisal (injectivity) stages. This may be possible if there is plenty of good quality subsurface data and analogues but this decision cannot be taken lightly. The purpose of the staged approach is to incrementally expose increasing amounts of capital as each threshold is met. A failure to reach the defined threshold pauses the project, and no additional capital is spent until (or unless) better results are anticipated. This helps project executors to be good stewards of shareholder investment. Prematurely removing these thresholds increases the risk of squandering large sums on failed projects.

Even in the early stages of an SCS project it is important to gain an appreciation of critical future risks that may occur during the Injection and Monitoring stages. Therefore, defining a process for building up a risk assessment in a progressive manner during the project framing phase may provide a practical solution to ensure that the risk assessment is fit for purpose and can evolve with the project. Implementing a robust assurance process during the project framing phase, which includes regular reviews and incorporates independent reviewers, can help to avoid certain biases and ensure objective assessment of the progress and risks to a project. These initial pre-injection risk assessments will undoubtedly need to be revised as performance data are gathered. Formal comprehensive risk reviews at selected time intervals are needed as part of the overall project plan.

This paper advocates that a quantitative risk assessment is required for SCS projects in order to fully appreciate their complexity and potential impacts, and to implement effective measurement, monitoring and verification (MMV) strategies over long time frames.

An SCS project must include an assessment of both current and future risks (Fig. 14). Current risks refer to those subsurface elements that can be assessed now, such as the pore volume of the reservoir and the presence of a seal. Future risks refer to elements that can only be fully assessed after injection begins, such as the connected pore volume and the effectiveness of the seal over the expanding plume.

Subsurface characterization of potential CO₂ storage sites in the Identification and Appraisal stages requires the evaluation of current risks. This is similar to the risking of oil and gas prospects, and typically uses a probability scale of 0–100% (Fig. 14). Many SCS site-selection studies have used risking strategies that were developed in the oil and gas industry, including Common Risk Segment (CRS) mapping and geological chance of success (e.g. Bump et al. 2021). These techniques are most useful in the characterization of the storage complex before continuous injection commences.

While the focus during the Identification and Appraisal stages of a project is on storage complex characterization and the evaluation of injectivity potential, an appreciation of future risks must also be developed during these early stages. This requires an estimate of the frequency of future events that could occur at some point during the Injection and Monitoring stages. The frequencies of future events are assessed as the probability of the event occurring in any given year. The probability values in this context are far lower than those used in oil and gas risk assessments; hence, an expanded log scale is more appropriate (Fig. 14).

For some events it is possible to estimate the frequency of future events from empirical data. As an example, Alcalde et al. (2018) and IOGP (2019a, b, 2022) have published data on rates and annual frequencies of events that may impact an SCS project. Other analogue frequency data that could be utilized as a guide to the base rate probabilities of events in SCS projects include seismicity data, natural disaster events, industry and government databases, and detailed analogue project reviews.

For events where insufficient empirical or analogue data are available, an estimate of likelihood must be made based on expert judgement and experience. Individual events are likely to have very low annual frequency rates; therefore, utilizing an expanded log scale is necessary. Because these low probabilities are outside of our usual experience, it is challenging to assign low probabilities to events using an individual's expert judgement alone.

Since this is a very subjective exercise, a risking scale supported by descriptive risk terms should be agreed during the project framing phase and used consistently by the project team (Fig. 14). When combined with a structured process such as the Delphi Method (Grime and Wright 2016) or use of an Expert Panel as recommended in the RISQUE methodology (Bowden and Rigg 2004), an internally consistent estimation of likelihood can be derived.

The expanded log scale presented in Figure 14 demonstrates that the annual frequencies calculated for a single event are often very low. However, when the potential for events is evaluated over very long time frames, the unmitigated risk of an event occurring increases during the life of the project (Fig. 15). The table in Figure 15, built from a series of Monte Carlo simulations, illustrates the effect of a series of constant but low-frequency events. In SCS projects, the frequency of events will vary across the long time frames involved; hence, a process to estimate these time-varying frequencies follows.

The features, events and processes (FEP) methodology is an established risk assessment practice that aids in identifying all potential elements of a system that could become hazards. This methodology is used in the nuclear energy industry for the risk assessment of radioactive waste disposal (NEA 2000; Maul et al. 2007) and is also recognized as a valid methodology in building an initial risk register for geological CO₂ storage projects (Cawley et al. 2005; Paulley et al. 2011). In the context of SCS projects, a large part of the FEP methodology involves identification of all potential escape routes and failure modes without judging initially how significant each one might be (Cawley et al. 2005). Complete risk assessment of CCS projects not only includes the subsurface technical risks but also needs to address above-ground risks such as public perception, and regulatory and commercial risks.

A comprehensive listing of FEPs in a risk register at the beginning of a project is critical to ensuring a thorough risk assessment that is suitable for informing robust MMV plans. This process can also be helpful to avoid biases that might impact interpretations and decision-making. A thorough risk register document also forms a record of decision-making and project knowledge that should be considered a living document throughout the project as it evolves.

A comprehensive database specific to geological CO₂ storage projects was published by Quintessa (Savage et al. 2004; Quintessa 2020) where more than 200 generic FEPs and 20 impacts were characterized for subsurface and above-ground risks. All of the FEPs included in this database will not be relevant to every project but by starting with a comprehensive database and supplementing it with analyses of past projects, an initial risk register can be built that ensures important failure modes are not overlooked or prematurely discounted because of cognitive biases.

Of the many FEPs that could impact an SCS project, not all will result in hazards or project failure. However, unexpected events may provide an early warning that initial model assumptions need to be revisited. Interdependencies between hazards and the potential for cascading effects that could increase the likelihood or severity of impacts also need to be identified and considered (ISO 2017). Therefore, building a comprehensive view of all possible FEPs is an important early step in the risk assessment process prior to determining the frequency with which that event may occur.

As discussed earlier, the ‘consequence’ term of the risk equation is a mixture of aleatory and epistemic uncertainties, while the ‘likelihood’ term is an aleatory chance factor. We recommend a two-step approach in which project teams first evaluate the epistemic uncertainty in the project factors and then, once they have tried to assess the full range of uncertainty, proceed to assess the chance that the assessment is correct and that unexpected events might occur. Both sets of data and judgements should be stored separately and summarized at critical stage gates in the project so that decision-makers can be fully informed.

Linking the FEPs to impact is important in order to remain focused on ensuring monitoring and mitigation activities are appropriate to identify any possible deviation from the plan. Understanding project risks and their potential impacts should be used to inform monitoring plans to ensure that they are appropriate to identify any loss of containment from the intended storage container; with mitigation and remediation plans developed to deal with such a contingency should it occur. Bourne et al. (2014) demonstrated a methodology for linking the SCS risk assessment with MMV design to ensure a fit-for-purpose plan that is specific to the project in order to manage the identified risks.

Quantitative risk assessment is necessary to gain a full appreciation of the potential impacts of low-frequency events and the changing risk profile over time. SCS projects must consider a large number of events with variable risk profiles throughout the project life cycle. Uncertainty also exists in many elements of these complex projects at each project stage, which needs to be captured and appropriately communicated. Tools such as Monte Carlo simulation, aggregation and confidence curves have been used successfully in oil and gas projects for many years. When adapted for much longer time frames, these techniques can support SCS project risk assessment.

An alternative methodology to running the multiple Monte Carlo models in Figure 15 is to utilize the Poisson formula (equations 2 and 3) to calculate the probability of the number of events during specified time intervals. The basis of the Poisson distribution is that the rate of the events is constant over the specified time interval, and that each event is independent. The results from the Poisson formulae tie with the Monte Carlo simulation results in Figure 15 when each constant rate is applied across the full 1000 year period.

However, in SCS projects the general model is that risk, the combination of probability multiplied by consequence, is not constant but increases during the injection period, and declines after the injection is completed. The Poisson formula can be applied to the event frequencies for each discreet time interval in order to estimate the chance of zero, one or more, or exactly N events in that time interval.

Dodds et al. (2011) showed the results of a proprietary computer assessment and Espie and Woods (2014) used reservoir engineering principles to determine the behaviour and extent of a CO₂ plume over time. These authors describe the concept of ‘Peak Risk’ for an SCS project. A key conclusion is that:

The process of immobilizing the plume through capillary trapping starts as soon as migration starts post-injection. Since the risk of leakage is a function of the volume of mobile CO₂ and the probability of faults or fractures, this implies that risk will decrease steadily to a low level in the post-closure period

(Espie and Woods 2014, p. 5460).

This paper builds on previous qualitative descriptions to present a workflow to develop a quantitative approach to the determination of Peak Risk. When combined with the range of possible consequences, a quantitative view of project risk can be developed. Tools such as those described above (staged approach, quantitative risking scales, real-world frequency data, FEP analysis and Poisson equation) can be applied to build a quantitative assessment to compare between different sites during the site Identification stage, and then further developed as detail is added when a project progresses through successive project stage gates.

As noted above, the frequency of events – and, hence, the overall risk profile – will change over the life of the project depending on the operations underway during each stage. This general model leads to the concept of ‘Peak Risk’ for a project (Fig. 17).

In the Identification stage, the geological setting is assessed and high-risk locations are rejected. Those storage complexes that are considered to have good potential for safe storage and which meet other initial commercial and technical criteria are progressed to the Appraisal stage, where the reservoir parameters are refined and injectivity potential confirmed, often through drilling and testing of an appraisal well. Project risk would generally manifest itself as a failure to locate the required reservoir quality, or a failure to meet the required injection rate during testing. The impact of a failure in these first two stages would be a loss of time and money limited to the companies involved in the effort.

During the Injection stage, all of the FEPs identified, as well as those that were missed, will start to be stress tested. This is when the whole storage system from transport to injection, pressure management and migration will be put under maximum stress and the greatest number of hazards are active at the same time, testing the overall storage effectiveness. Project risk would then be expected to decline during the post-injection Monitoring stage as the plume migrates, then stabilizes in the near to mid-term, and reaches geological stability to achieve long-term retainment.

The timing and duration of this Peak Risk period depends on each specific project, so the shape of the curve will vary. Because this period from starting injection to the plume becoming geologically stable can be very long (hundreds to thousands of years), a quantitative risk assessment can assist in better understanding the overall risk profile for each specific project.

Dodds et al. (2011) presented a quantitative assessment for the In Salah project using an internal BP methodology. Injection began in 2004, and data available by 2008 predicted a Peak Risk period between 10 and 20 years after the start of injection. The project was closed after 7 years of injection (Ringrose et al. 2013). Various qualitative descriptions of risk as a function of time have been published by Benson and Surles (2006) and Espie and Woods (2014). Bourne et al. (2014) presented a risk-based framework for MMV of the Quest CCS project. Their approach was to focus on the quality of the technology needed for effective and actionable conformance and containment monitoring. They used a slightly different set of stages – Pre-Injection, Injection, Closure and Post-Closure – as the framework for the technology assessment. In the Quest project Post-Closure refers to the handover of the site back to the Canadian authority, known in Canada as the Crown. A risk metric was presented that is based on the chance that one consequence arises due to the initiation of at least one threat and the subsequent failure of all the intervening safeguards. Apart from the stage gate description, there was no quantitative discussion of the changing risks over time. The specific geology and lightly drilled approved sequestration lease indicate that risks are indeed likely to decrease over time in the specific Quest project case.

Alcalde et al. (2018) provided substantial quantitative details on a wide variety of possible leakage routes, together with the computer code that produced their results. A key driver of their model is the area of the CO₂ plume. This variable is developed from an area/mass ratio distribution that was developed from the documented properties of gas fields in the Southern North Sea. Their conclusion, that well-regulated storage in areas with moderate well density retains essentially all of the injected CO₂, is predicated on this area/mass ratio distribution. Their model assumes a storage efficiency of 100%. Lower storage efficiencies, of the order of 5–10%, would have a significant impact on their model and conclusions. The range of storage efficiencies also has a significant impact on the theoretical storage capacity of the Rough Field (de Jonge-Anderson et al. 2022). The point here is that presenting the critical variables in models is highly recommended as this transparency enables detailed scrutiny.

The Peak Risk workflow is based on assessing a particular opportunity to develop an estimate of Peak Risk; in other words, assuming that a storage complex has been identified for detailed appraisal. The data in the simplified example below was built based on published information of a potential offshore storage site, assuming that the Identification stage was completed:

1. Decide on the durations of the time frames to be used, which are generally tied to the agreed stages of the project – here these are Appraisal, Injection and Monitoring.

2. Define the quantitative event frequency scales and build a set of financial impacts characterized on the scale; for example, Incidental, Minor, Moderate, Major Severe and Catastrophic. As the financial impacts are inherently uncertain, we recommend using a log-normal distribution for each category, with the distribution defined by a P₉₀ and P₁₀ value. Here we are using the E&P industry standard in which the P₉₀ is the low value and the P₁₀ the high value. The table used in this example workflow is shown as Figure 18.

3. Define and refine the list of FEPs that are relevant to the particular project.

4. Proceed to evaluate, as an expert team, each FEP systematically, including the impact if the FEP should fail, and the annual chance of failure in each of the stages. An example of a simple user interface to record these discussions is presented in Figure 19. As each FEP is evaluated, the results are stored in a separate worksheet and the cumulative frequency for all FEPs are calculated and aggregated.

These steps are brought together in a recommended workflow to develop an estimate of the Peak Risk (Fig. 20).

Figure 21 presents the cumulative frequency data for a total of 146 FEPs that were evaluated in this example. The data are displayed in terms of the six categories of assessed impact and by the relevant time intervals. The number of FEPs in each category is included in the diagram.

Figure 22 presents a simple example in which the average frequency for specific time intervals was built using the data in Figure 21. As the ability to predict the timing of the occurrence of future events is highly uncertain, a smoothed function could be used to describe the annual likelihood of events. The area beneath each of these curves is identical.

Figure 23 presents the chance of zero, one or more, and exactly one, two or three events using the smoothed likelihood data in Figure 22 as the input into the Poisson equation. Using a threshold of 50% probability of an event as an informal measure, the project will be subject to an increased frequency of events between years 20 and 90.

A Monte Carlo simulation of the likelihood of occurrence and the P₉₀–P₁₀ range of possible impacts was used to derive the range of risk, expressed in annual millions of dollars, for the modelled example (Fig. 24). Further analysis of the model can identify those FEPs that are key drivers during the period of Peak Risk. In turn, those data can assist in the design and implementation of a robust and adaptive MMV programme.

The key workflow elements discussed above, from project framing to quantifying Peak Risk, are summarized in Figure 20.

Modelling forms an essential part of the workflow across all stages of an SCS project and informs the risk assessment of multiple hazards. However, many published pre-injection model-based predictions show either minimal leakage or no failure over very long timescales (Oldenburg et al. 2009; Alcalde et al. 2018; Karvounis and Blunt 2021) (Fig. 25). At Sleipner, the initial reservoir models were updated several times as the scenario of minor intermediate seals was considered and eventually confirmed by 3D time-lapse surveys (Zweigel et al. 2004; Ringrose 2018). Subsurface models are necessarily a simplification of the complex geological system; therefore, stress testing of models is a key due diligence step.

Steps should be taken to mitigate common biases related to models. Best practice recommends that multiple subsurface models should be generated; however, due to the time and computing power needed for some reservoir models, there can be a tendency to generate a ‘best case’ model without fully exploring the impact of parameter uncertainty on the risks. Including real-world failure frequencies in the models, in addition to subsurface data, will help to identify and better assess the risk of failure mechanisms. In addition, an ongoing evaluation of subsurface models in conjunction with the monitoring of injection and plume migration can inform updates to the models and identify which of these is the closest representation of reality. It is possible that models which show no significant events occurring over many thousands of years contain cognitive errors, and detailed scrutiny of the underlying assumptions is recommended.

A recurrent theme in past injection projects is that subsurface models did not capture key reservoir heterogeneities, creating the impression of a low-risk, viable project. Regulatory requirements for subsurface models may, for this reason, not allay the risk of project failure unless key uncertainties are addressed. Recently, independent critical evaluations of storage complexes have been undertaken and published. Several critical issues were identified in the HyNet project in Liverpool Bay, in the UK, including the shallow top of the main reservoir, a waste zone above the main reservoir, faults that may reach the seabed and concerns about seal integrity (Chedburn et al. 2022). Independent analyses such as these should form part of a comprehensive risk assessment for a project. In general terms, a regular schedule of reviews should be defined during the project framing stage. The paper referenced above supports our recommendation that technical reviews should include independent reviewers to aid in addressing the cognitive biases that may impact a project team deeply involved in a project and who want to see it implemented.

Examples of the types of questions that should be asked by assurance teams and decision-makers to facilitate discussion and ensure all aspects have been considered could include:

• What is the source of the numbers that justify the recommendation? Are they consistent with project performance history? If the numbers are better than performance history, what justifies this? Are they anchored to appropriate benchmarks, analogues and/or models?

• Are credible alternatives included along with the recommendation? If not, request additional options that, for example, require more or less capital expenditure.

• Does the recommendation assume that an approach that is successful in one area will be just as successful in another? If so, comparable examples are needed to eliminate false inferences.

• Could the recommendation be overly influenced by an analogy to a rare but memorable success? If so, additional analogues and an analysis of their similarity to the current situation are needed.

• Could the base case be overly optimistic? Build a case taking an outside view (pose challenges that an investor would).

• Is the worst case bad enough? Conduct a pre-mortem assuming the project has failed. What caused this?

• Is the recommendation overly cautious? May need to realign incentives to promote prudent risk-taking.

• Is there an over-attachment to a history of past decisions? People ascribe more value to projects that they have owned.

• Were there dissenting opinions leading up to the recommendation? Were these explored adequately and resolved?

• If we delay a decision on this project for 1 year, what data would be gathered in the interim and what impact could this have?

CCS is a low-margin industry; therefore, if it is going to be successfully commercialized at scale, the costs will have to be reduced relative to what companies spend to characterize portfolios of oil and gas prospects that are expected to generate millions in revenue. This may mean that less data will be available for subsurface characterization, requiring subsurface professionals to apply their industry experience and use relevant analogues to fill the gaps. But it will also require more efficient and selective project development.

We suggest that by establishing a workflow incorporating the key elements presented in Figure 20, project teams and decision-makers can develop a rigorous risk assessment process that is flexible enough to evolve with a project as it matures, while helping to ensure critical events and hazards are identified ahead of time.

Companies must be very clear as to what can go wrong in a SCS project, the impacts that can result, and how to mitigate risks to obtain and maintain their social licence to operate. This licence cannot be taken for granted and much work is required to gain and maintain trust. As a nascent industry, there is currently a lack of understanding by the general public of what SCS is, and it is already burdened by poor publicity through strong association with the oil and gas industry. Therefore, tolerance for failure will be low. As the Gorgon project (discussed above) has already demonstrated, problems and failures in one project have the potential to negatively impact other projects in the eyes of some stakeholders. A robust and thorough quantitative risk assessment workflow for the full life cycle of a project is critical to demonstrate due diligence.

The authors wish to thank all reviewers for their constructive and useful comments and suggestions.

CJ: conceptualization (equal), investigation (equal), writing – original draft (equal), writing – review & editing (equal); PP: conceptualization (equal), investigation (equal), writing – original draft (equal), writing – review & editing (equal); PC: conceptualization (equal), investigation (equal), writing – original draft (equal), writing – review & editing (equal); RC: conceptualization (equal), investigation (equal), writing – original draft (equal), writing – review & editing (equal).

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

The datasets generated during and/or analysed during the current study are not publicly available due to commercial sensitivity but are available from the corresponding author on reasonable request.