The Wilmington oil field (California, USA) is located under the City of Long Beach and adjoining harbor. When this field was developed in the late 1960s from artificial offshore islands, more than 20 water injection wells were drilled and completed in the aquifer. The objective was to create a peripheral waterflood using an injected water volume exceeding 100,000 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 Field. This led to an increase in the oil production rate and overpressuring of the reservoir (see the Figure below). Wells were flowing with a surface pressure of 200 to 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 aquifer waterflood injectors for about six months. This allowed the pressure to decline in the aquifer before resuming injection at a reduced rate of 30,000 barrels per day. Additionally, the Seal Beach operator installed pressure gauges in idle wells to monitor the reservoir pressure and help ensure that no future well blowouts would occur. This case study is a good example of how injected fluids can cross barriers that were previously considered sealing.
The Huntsman Field (Nebraska, USA) produced 28 billion cubic feet 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 Huntsman, which implied some type of pressure connection.
West Engelland 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 eventually were attributed to tilting of the gas-water contact, allowing gas to flow from Huntsman to West Engelland.
Not surprisingly, this led to a lawsuit and an out-of-court settlement in 1985 to compensate the Huntsman operator for the storage gas that was siphoned to West Engelland. The figure below shows the relationship between the two fields and demonstrates how we can mistakenly conclude that separate geological structures are hydraulically isolated.
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 eight examples is presented, beginning with this post. The examples include methane, water, and CO2 injection projects.
This first example focuses on the Castor Project in offshore eastern Spain. The project involved a depleted oil field (Amposta) 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 billion cubic meters of natural gas, sufficient to meet 25% of Spain’s storage requirements.
Studies concluded that the reservoir characteristics, especially the distribution of porosity and the sealing capability of the caprock, were sufficient. However, this work did not consider the possibility of induced seismicity. Amposta Field had been produced under a natural water drive and there had been no previous injection in the field.
In 2013, just three days after gas injection began, a series of earthquakes occurred along a deeper, previously unidentified fault. This culminated in three earthquakes with a moment magnitude > 4. Injection was halted, and the site was permanently closed six years later at an estimated total cost to Spanish citizens of up to 4.7 billion Euros. A subsequent analysis showed that assessing fault stability prior to gas injection would have identified the risk of induced seismicity.
A cross-section (a) and map (b) depicting the location of injection is shown below. On the left, the yellow shading indicates the target reservoir. On the right, the black square is the injection well location, the circles are earthquake locations, and the black circles are the three largest earthquakes (from Foulger et al, 2018).
Reference: Foulger, G.R., Wilson, M.P., Gluyas, J.G., Julian, B.R. and Davies, R.J. 2018. Global review of humaninduced earthquakes. Earth-Science Reviews, 178, 438-514.
Fluids have been injected in underground reservoirs for temporary or permanent storage for more than a century. The first natural gas storage operations were undertaken in 1915 at a gas field in Canada. By the 1920s, the injection of water to increase oil recovery had been successfully demonstrated and the widespread use of injection wells to dispose of produced water began in the 1930s. 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 historic review of 228 reported events and failures worldwide by 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 40 or 50 years from now when carbon storage projects will be reaching the end of their injection lives.
A small but significant fraction of injection project events can be attributed to geological factors. These include, but are not limited to:
The figure below depicts many of these geological factors and the movement of CO2 into unintended portions of the subsurface.
Reference: Evans, D.J. 2009. A review of underground fuel storage events and putting risk into perspective with other areas of the energy supply chain. Geological Society, London, Special Publications, 313, 173-216.
Subsurface characterization is critical for any SCS project. In particular, we need to accurately quantify 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 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. 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.
Consider, for example, the use of waterflooding for secondary recovery. During the first few decades of implementation, many valuable lessons were learned including 1) recovery factors are highly variable and dependent on rock and fluid properties as well as development strategies and operational practices, 2) an absence of analogs and understanding led to multiple missteps such as implementing peripheral injection in fields that required in-zone injection, 3) attempts to create simple empirical relationships between reservoir parameters and recovery efficiencies failed to produce statistically valid correlations, and 4) models only became truly useful for designing and managing waterfloods after they were calibrated to multiple long-term projects.
In the next set of posts, we describe several examples of injection project events, illustrating what happened, the factors responsible, and if these events could have been foreseen. We then propose a quantitative risk assessment framework suitable for evaluating long-term, low-frequency events which can inform monitoring, mitigation, and remediation plans to help ensure a successful project outcome.
In a previous post, we defined chance as the probability of a given event. In contrast, the Storage Resources Management System (SRMS) defines chance as the probability of gain or success. But what about the probability of loss or failure? In our view, the question “how could my project fail?” is actually more relevant than the question, “how could my project succeed?”
A focus on successful analogues reinforces the use of heuristics (rules of thumb) and nurtures our intuition. This distorts our objectivity and makes us more vulnerable to biases which can skew our assessments. Some of the more common biases we have 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 motivational bias (actions driven by a desire for personal gain).
As an example of their impact, Kahneman and Tversky (1982) showed that in complex systems that require a series of events to succeed, biases such as anchoring can lead to the underestimation of the probability of failure. Even if the individual chance of failure for each event is low, and the impact of bias is small, the aggregated impact on the chance of failure may be significant. Focusing on potential failure modes and applying a risk assessment workflow that mitigates bias is therefore important for evaluating long-term, low-frequency, high-impact events.
Focusing on the chance of failure 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. But it’s important to remember that events, if they occur, may or may not be hazards. For example, a single seismic event of less than a magnitude of 2 on the Richter scale is very unlikely to be a hazard. However, swarms of these small events can and do become real hazards to project continuation.
Reference: Kahneman, D. and Tversky, A. 1982. Judgement under uncertainty: Heuristics and biases. In: Kahneman, D., Slovic, P. and Tversky, A. (eds) Judgement under uncertainty: Heuristics and biases. Cambridge University Press, New York, 3-20.
There are two dimensions of risk and uncertainty to consider in SCS projects. Epistemic probabilities and uncertainties are those that can be determined and refined by increasing knowledge of the physical properties of the system. In contrast, aleatory probabilities and uncertainties are inherently random and cannot be reduced by technical work.
Earthquake prediction provides an example of aleatory probability. For the San Francisco area, Aagaard et al. (2016) state that there is a “72% percent probability of one or more M >= 6.7 earthquakes from 2014 to 2043”. Similarly, for SCS projects, a team may forecast an annual frequency that faults may slip as injection proceeds. This would be an aleatory probability forecast. Frequency data from relevant analogues can inform the estimation of aleatory probabilities and uncertainties for SCS projects.
Epistemic probabilities and uncertainties for SCS projects can be evaluated using system data and models. Uncertainty can be reduced by additional data gathering and improved analyses. For example, the range of the average porosity for a target reservoir can be narrowed by the careful analysis of additional well logs.
This duality means that reservoir models will have parameter variability that arises from both natural heterogeneity (aleatory) and from a lack of knowledge (epistemic). These two types of uncertainty should be considered separately and clearly documented. For example, the combination of aleatory and epistemic uncertainty was applied to a model of CO2 plume extension by Bellenfantet al. (2009). They proposed weighting optimistic and pessimistic forecasts to describe the overall uncertainty to decision-makers.
Reference: Aagaard, B.T., Blair, J.L., Boatwright, J., Garcia, S.H., Harris, R.A., Michael, A.J., Schwartz, D.P., DiLeo,
J.S. 2016. Earthquake Outlook for the San Francisco Bay Region 2014–2043. USGS.
Reference: Bellenfant, G., Guyonnet, D., Dubois, D. and Bouc, O. 2009. Uncertainty theories applied to the analysis of CO2 plume extension during geological storage. Energy Procedia, 1, 2447-2454.
Risk can also be defined as the product of two factors: the likelihood of an event, and its consequences.
Risk = Likelihood x Consequences
However, if the likelihood (chance) of occurrence is assessed as very low, the risk of a high consequence event will be obscured. There may, in fact, be no distinctive separation between low likelihood / high consequence events and high likelihood / low consequence events. This can be seen in Risk Matrices which have been widely used to provide a semi-quantitative view of likelihood and consequence.
The figure below presents an example risk matrix from a report relating to the Porthos CO2 Storage Project in Offshore Netherlands (Neele et al. 2019). The lowermost left cell, A-5, is defined as having a very low chance of a very large amount of CO2 migrating out of the reservoir. This cell is assigned a medium risk level (colored orange).
Similarly, the uppermost right cell, E1, is assigned a medium risk level. For this cell, there is a very high chance that a negligible amount of CO2 will migrate out of the reservoir. Does it seem reasonable that both of these scenarios pose a medium risk to the project? Even though it’s a low chance, the impact of a large leak will be much more damaging to the project than an almost certain leak of negligible consequence.
Note in the Figure that the corresponding ‘Monitoring Necessity’ and ‘Risk Reduction’ actions for these two scenarios are set to ALARP (As Low As Reasonably Practicable). This is also problematic—it’s much more important to collect data that could herald the approach of a low chance, high impact event than it is to detect high chance, low impact events.
Given the ambiguity of the Risk Matrices, we recommend evaluating the chance of an event separately from its consequences or impact, and will develop this concept in future posts.
Reference: Neele, F., Wildenborg, T., Geel, K., Loeve, D., Peters, L., Kahrobaei, S., Candela, T., Koenen, M., Hopmans, P., van der Valk, K., Orlic, B., Vandeweijer, V. 2019. TNO 2019 R11635 CO2 storage feasibility in the P18-2 depleted gas field.
Risk, uncertainty, chance, and related terms are regularly used in the course of daily work in the oil and gas industry. However, there are significant differences between individuals and companies regarding the definition and understanding of these terms, which makes it important to clarify them:
For low frequency events, a forecast that nothing will happen is likely to be correct most of the time. Except, of course, when it’s not. And if the impact is massive, we might refer to it as a Black Swan event (per Nassim Taleb’s book). Black Swan logic argues that what you don’t know is far more relevant than what you do know.
In fact, many Black Swans can be caused and exacerbated by being unexpected. We ignore or misinterpret early indications of such events and make no preparations to stop or mitigate them. Taleb stresses that time and time again we concentrate on things we know and fail to consider what we don’t know. We need to keep our apertures wide and recognize, as we study the landscape of project failures, that “if it happened, you must admit it’s possible”.
During exploration and appraisal, oil and gas companies accept that there will be failures due to dry holes and non-commercial discoveries. For this reason, they plan to drill enough independent prospects to ensure that the value associated with the commercial discoveries exceeds the program cost (including dry holes). This outcome is referred to as the portfolio effect.
The figure below illustrates the portfolio effect. It shows 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.
We contend that the portfolio effect will not be acceptable to companies, regulators, or the broader societal interests in SCS 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 millions of dollars in carbon credits.
To mitigate against this, a complete assessment of subsurface uncertainties and risks combined with estimates of the chance of success and failure are needed. These are discussed in subsequent postings beginning with the definitions for these terms.
