The rising interest in subsurface CO 2 storage makes new calls on reservoir modelling skills, most of which have been developed for hydrocarbon production scenarios. The question for practitioners is: to what extent can the familiar production tools be transferred to the world of storage? In this paper, areas requiring attention are highlighted and high-resolution models are used to compare the behaviour of simulators for production v. storage for two reservoir analogue examples. It is concluded that modelling for storage makes a significant call on multi-scale modelling, to a much greater extent than in production scenarios, and the simplification or omission of reservoir heterogeneities (sometimes tolerable in production scenarios) are much less tolerable when modelling storage. Key static model heterogeneities include the modelling of faults as 3D features, the inclusion of fine-scale reservoir permeability contrasts and the avoidance of net reservoir cut-offs. For dynamic models, use of equation of state is necessary for storage in depleted fields, and correct representation of hysteretic effects of plume migration are a requirement for modelling in aquifers (always) and depleted fields (usually). Modelling for storage, especially for saline aquifers, sets the challenge of modelling volumes previously considered to be at exploration scale, but with an effective resolution more typical of production scales.
Reservoir architecture is key in determining reservoir performance and hydrocarbon productivity but varies greatly in deep-water clastic reservoir systems. However, the interpretation of turbidite architectures can be a daunting exercise for non-specialists or those without experience in deep-water settings. Parameters that can be used to make interpretations are copious, vary in usefulness from one system to another, vary in scale and can take a lifetime to master. In this paper we introduce the simple concept of the relative confinement matrix and explain how it can be used by non-specialist geologists to interpret depositional architectures and hence inform all-important forecasts of performance and productivity. We show how any available parameter can fit into this matrix and be used to derive interpretations in a relatively quick manner. Our central concept is the relationship between the size of a turbidity current and the size of the conduit through which it flows and hence the degree of relative confinement to which a flow is subjected, which can in turn be related to depositional style and resulting reservoir architecture. Predictive interpretations can be made by observing patterns at the core and log scale, relating these to relative confinement and thus to larger, seismic-scale architectures. Variations in relative confinement are expressed through lateral bed continuity, vertical connectivity, amalgamation ratio, net:gross, distribution of hemi-pelagics, distribution and uniformity of facies associations, bed thickness frequency distribution, bioturbation diversity and intensity, distribution of sedimentary structures, mineralogical content, variability & textural maturity, grain size and grain size variability. Notionally the expression of the interaction of flow size and conduit size is described for individual turbidity current deposits, but this can also be translated to the bed-set scale, or larger, in genetically similar units. The careful analysis of the parameters described above allows the prediction of depositional architectures through the understanding of relative confinement for use in exploration and development. The architectural predictions in turn provide the basis for understanding permeability length-scales, kv/kh ratios and hence production performance through these heterogeneous reservoirs.
We investigate how efficiently oil can be recovered from a carbonate rock during surfactant based enhanced oil recovery (EOR) at the core-scale, particularly when chemical processes change wettability, and analyse how geological heterogeneities, observed at the next larger scale (centimetre to decimetre) impacts the effectiveness of surfactant-based EOR at the inter-well scale. To quantify how heterogeneity across scales impacts surfactant flooding, we combine laboratory experiments with simulation studies at the core- and inter-well scale. We first analysed a series of surfactant imbibition experiments at different surfactant concentrations (from 0 to 3 wt. %) using reservoir cores from the Wakamuk field, a carbonate reservoir in Indonesia. We then built a 3D simulation model of the laboratory experiment and matched the experimental data to identify the key physical mechanisms (e.g., reduction in interfacial tension (IFT) and wettability alteration) that lead to increased oil recovery. Next, we parametrised the surfactant models using assisted history-matching methods to calibrate the relative permeability and capillary pressure curves as a function of surfactant concentration. These models were then deployed in high-resolution simulations at the inter-well scale. These simulations captured the small-scale geological heterogeneities that are typical for a carbonate reservoir system, e.g., the Shuaiba formation in the Middle East, but are not resolved in field-scale models. Our core-scale simulations demonstrate a change from co- to counter-current flow in the laboratory experiments and indicate that the resulting increase in oil recovery is due to a combination of IFT reduction, wettability alteration from oil- to water-wet, and capillary pressure restoration; these processes need to be captured adequately at the inter-well scale model. The increase in surfactant concentration above the critical micelle concentration (CMC) (i.e., from 1 to 3 wt. %) triggered the capillary pressure restoration and dominated recovery at the early-time. The changes in relative permeability and capillary curves during the surfactant floods were best modelled using a concentration-based interpolation. There is uncertainty when calibrating surfactant models using laboratory experiments. A key question hence is if geological heterogeneity at the inter-well scale masks these uncertainties. Results from our high-resolution simulations show that large-scale heterogeneity impacts recovery predictions, but it is the coarsening of the grid, not the upscaling of permeability, that dominates the error in field-scale recovery predictions during surfactant based EOR. Indeed, the error arising from numerical dispersion during grid coarsening can be as large as the error arising when selecting an inaccurately configured surfactant model due to the lack of quality experimental data. Hence appropriate grid refinement, possibly using adaptive grid refinement, needs to be considered when setting up a surfactant based EOR simulation, along with the appropriate configuration of the surfactant model itself.
Reservoir modelling tools can be invaluable for integrating knowledge and for supporting strategic oil field decisions. The pertinent issue is the capability of the modelling toolbox to achieve the required support: does modelling generate insights into the characterization of the subsurface, does it increase or decrease our working efficiency and does it help or hinder us in decision-making? In this respect, we see two directions emerging in reservoir modelling and simulation. One surrounds software technology development and a move towards a grid-independent world. This is a current research issue but some of the components required to complete a new workflow are already in place and tools for certain specific applications may not be far away. The other involves a change in approach to model design. This involves a move away from big, detailed 'life-cycle' models to more nimble workflows involving multi-models (either multi-scale or multi-concept) which may or not include full-field modelling exercises. A distinction between 'resource models' and 'decision models' helps crystallize this, is a positive step towards achieving 'fit-for-purpose' models, and is a change of model design strategy which can be achieved immediately.
Summary Reservoir modelling studies are now widespread and are often built into a formal gated process used for decision-making, at least as an option. Once ubiquitous, it is easy for the models to simply become tools for verification of a decision that has partially (sometimes wholly) been made – ‘modelling for comfort’. This is particularly the case in mature fields, when the presence of an inherited model already anchors the view of the field, and the volume of production data discourages the practitioner from exploring uncertainties with multiple models. It is proposed that reservoir modelling offers most value when used to create some discomfort – a stress-test for decision-making that can identify upsides and secure against loss. This requires an awareness of the biases at work in model design and a conscious choice to move away from the default of a single, detailed, full-field model. This ideally means moving away from base-case led modelling altogether and typically involves multi-scale model design and multiple-models for uncertainty handling, based either on stochastic modelling or multi-deterministic, scenario-based approaches.
Reservoir modelling studies are widespread and are often built into a formal gated process used for decision-making. Now ubiquitous, it is easy for the models to simply become tools for verification of a decision that has partially or wholly been made - 'modelling for comfort'. This is particularly the case in mature fields, when the presence of an inherited model already anchors the view of the field, and the volume of production data discourages the practitioner from exploring uncertainties with multiple models. It is proposed that reservoir modelling offers most value when used to create some discomfort - a stress test for decision-making that can identify upsides and secure against loss. This requires an awareness of the biases at work in model design and a conscious choice to move away from the default of a single, detailed, full-field model. This ideally means moving away from base-case-led modelling altogether, and typically involves multi-scale model design and multiple models for uncertainty handling, based either on stochastic modelling or multi-deterministic, scenario-based approaches.
Seismic data play a prominent part in the quantification of the subsurface. Improved imaging and calibration give us a better starting point for interpretation and uncertainty analysis. However, aside from the technical aspects of evaluating seismic data, there are human factors that play a role in the way we use and analyze the data, and these tend to work against attempts to quantify realistic uncertainty ranges. We used a case study to reveal some common pitfalls and assumptions that can compromise our ability to produce sufficiently wide uncertainty ranges in our evaluations. The example highlighted three human factors that affected the decision-making process: anchoring, availability, and overconfidence. Interpreters should avoid anchoring on a base case and focus on alternative possibilities. They should be wary of judging which methodology is best only by the ease with which it comes to mind. Technical specialists should guard against overconfidence in their data, interpretation, and ability to describe the full uncertainty space. We suggested alternative methods that allow us to restore that uncertainty range using a multideterministic approach incorporating multiple data sets, interpretations, and methodologies.