Publications with abstracts

Now that Earth Modelling tools are used routinely by most petroleum companies, efforts are under way to adapt multidisciplinary data inversion techniques to better constrain these models by geological, geophysical and dynamic data. A new approach has emerged for building 3D geological models. This approach, known as Multi-Point Statistics (MPS), uses a radically new concept as compared to variogram-based approaches. The geologist first builds a “training image”, that is a representation (in 3D) of what he expects the geological architecture of the modelled field to look like. Then, MPS extracts the statistical patterns from this image and reproduces them when generating a 3D realisation of the reservoir, whilst matching well data and other constraints such as non-stationary trends. Stanford University has played a key role in the development of MPS. Increasinly, the integration of seismic data is performed using stochastic approaches. Thanks to a technique known as geostatistical inversion, a large number of high-frequency acoustic impedance realisations are generated which all match the 3D seismic data. From these realisations, uncertainty figures can be obtained about the spatial distribution of high and low acoustic impedance streaks in the reservoir. Universities such as Edinburgh, Heriot-Watt, Trondheim or Stanford play a key role in these developments. The use of geostatistical realisations is also spreading to the generation of models constrained by dynamic data. One of the new approaches naturally linked to multi realisation concept is the Ensemble Kalman Filter (EnKF). EnKF has already given quite promising results on a number of studies. It can be used to generate reservoir models matching production data with spatially continuous 3D fields like porosity and permeability constrained in each grid cell. We are working closely with Norwegian universities such as Trondheim, Bergen, Texas A&M or Oklahoma, who are priogressing very rapidly on this topic. The importance of rock physics is becoming stronger and stronger, as it provides the link between the geological, the seismic and the dynamic domains. With many fields reaching maturity and hence a critical stress state, we should also expect geomechanics to play a growing role in Earth Modelling applications. The benefits of such multidisciplinary integration can be measured in terms of more realistic models of the subsurface which are consistent with all available data and conceptual views in all domains, adding significant value to a large range of complex fields, including deeply buried high pressure and high temperature North Sea fields in the UK.

The relationship between kriging, conditional simulation and other inversion approaches is discussed. We recall that kriging, radial basis functions, splines are formally identical, and that kriging can also be seen as a regularisation approach. The relationship between conditional simulation, Bayesian inversion and Kalman filtering is also addressed. This is illustrated by practical examples of geostatistical seismic inversion and production data inversion using Ensemble Kalman Filtering. Then the issues associated with the joint integration of geological, seismic and dynamic data are discussed, and it is stressed that there is still a long away to go, especially as far as uncertainty quantification of the jointly inverted model is concerned. We conclude by more general considerations about the geostatistical and the Bayesian approach.

Faults affect lateral fluid flow by juxtaposing different sedimentary layers or due to the petrophysical properties of the fault zone materials themselves. To predict the behaviour of intra-reservoir faults, an estimate of fault permeability is required. As the petrophysical properties of large (?10m displacement) faults in the subsurface are rarely sampled, measurements must instead be taken from natural faults in core or outcrop, or from synthetic faults created by laboratory or numerical deformation experiments. The relative insensitivity of permeability to shear strains >1 (e.g. Crawford et al., 2002) suggests that the properties of small faults can be extrapolated to larger features, provided they are corrected to account for in situ stress. Furthermore, small fault data give an estimate of the inherent variability of fault permeability, albeit at the core plug scale, which can be used for uncertainty analysis. The aim of this study is therefore to define a probability density function (PDF) for estimating fault permeability at the reservoir grid scale. The data used in this study are from small faults in North Sea Middle Jurassic reservoirs (Fisher & Knipe, 2001; Sperrevik et al., 2002), supplemented with core and outcrop data from various basins (Gibson, 1998) and additional unpublished core data. Experimental fault data (Crawford et al., 2002) are also included, albeit modified to give an estimate for across-fault permeability. Through deformation-induced mixing (Fisher & Knipe, 2001), clay content is assumed to be the main control on fault permeability. Due to the difficulty of directly measuring fault rock clay content (denoted Vf), it is assumed to be equal to host rock clay content (Vh), except for clay smear samples which are assigned Vf = 60%. Fault permeability may show a negative relationship with maximum burial depth (Zmax) due to the effects of mechanical and thermal compaction, growth of diagenetic minerals and enhanced quartz precipitation (e.g. Fisher & Knipe, 2001). A multiple regression between log ko, Vf and Zmax for all samples with depth data available (n = 178, 0 < Vh ? 60%), gives r = 0.62 with a standard error of 1.56. It is shown that the residual values can be modelled as having a normal (Gaussian) distribution (i.e. lognormal for ko). The derived equation can be used to compute mean permeability from an estimated value of Vf, e.g. using Shale Gouge Ratio (SGR). The exact form of the equation is less important than the definition of an uncertainty range within which fault permeability values may occur. A 95% confidence interval encompasses most of the variation between various published equations. The log permeability (residual) distribution derived above is applicable at the core plug scale; the statistics must be upscaled before they can be applied in a reservoir grid. To do so requires knowledge of the spatial variability of fault permeability, which is highly speculative. We assume that fault permeability has a correlation length (variogram range) that is larger than a core plug (ca. 10cm) but smaller than a typical simulation grid block (50-100m). A method to upscale the permeability distribution is to perform unconstrained Sequential Gaussian Simulation (SGS) within a 2D grid equal in size to a reservoir grid block but discretised at the core plug scale. Each realisation of log ko is averaged using the arithmetic mean (i.e. geometric mean of ko); the standard deviation of all mean permeabilities defines the upscaled distribution of residuals. This PDF can then be used in stochastic simulations of fault permeability at the grid scale. An upscaled PDF allows random sampling of fault permeability values, enabling fault properties to be incorporated into uncertainty analysis workflows. Fault transmissibility multipliers for use in reservoir simulations can be computed by also sampling from an appropriate fault thickness (tf) PDF, with mean thickness estimated from fault throw. Transmissibility multipliers can then be included as uncertain parameters in assisted history matching applications.

Fault properties estimated using lithoseismic Vclay are compared to those derived from a classical geomodelling approach with reservoir NTG. Throw and Shale Gouge Ratio (SGR) calculation at the seismic scale can provide a significant improvement in resolution and accuracy over the geomodelling approach. Fault properties included in numerical reservoir simulations give a qualitative match to observed production and time-lapse seismic effects in an offshore turbidite field. Errors introduced during upscaling of seismic properties into ‘stairstep’ faulted grids can be overcome using a virtual deformation of the grid geometry. This method could be extended to upscaling fault permeability estimated from seismic SGR.

The genetic stratigraphy concepts state that the distribution of sedimentary bodies is controlled both by the accommodation rate and by the sediment supply. The main rule introduced by this stratigraphic approach is known as the A/S ration concept, describing the different stratigraphic system tracts. For instance, transgressions are identified by an A/S ratio greater than 1 while regressions corresponds to a ratio lower than 1. Based on this concept, we propose a methodology for the modeling of the paleotopography, by accounting for stratigraphic records, well data and sedimentological rules. From the paleotopography, facies probabilities are then deduced using facies occurrence rules. As a corollary of the A/S ratio concept, the sediment volumetric partitioning rule states that the sediments are better preserved in the distal pole during regressions and in the proximal pole during transgressions. If transgressions and regressions are well identified on the chronostratigraphic correlations, the lateral gradient of the sediment thickness provides an indicator of the distality vector. Using DSI interpolation of membership functions constrained by trend vectors, we propose to compute a second probability field accounting for the distality. The advantage of this additional probability field is to better capture information about the facies distribution in case of confinement, like for instance in lagoon systems. By combining these facies probabilities, the suggested solution provides a more complete description of the sedimentary system and gives more realistic facies simulations. This methodology implemented in the Thalassa plug-in is illustrated on clastic and carbonate reservoirs.

Faults affect fluid flow in hydrocarbon reservoirs on multiple length and time scales. Sampling of subsurface information is limited to spatially extensive but low resolution seismic data and sparse, high resolution well penetrations. Large (=10m displacement) faults are usually avoided as a drilling hazard and very rarely cored, while typical seismic surveys are designed to image sub-horizontal reflectors rather than faults. This lack of knowledge must be filled by making inferences from indirect data (inverse modelling) and/or by employing some kind of conceptual model to predict fault properties from other information. In siliciclastic sediments, the type of fault rock developed depends mostly on host rock composition and, to a lesser extent, deformation conditions and temperature history. Fault permeability may be measured from small faults in core, faults in analogue outcrops or synthetic faults created in laboratory deformation experiments. Fault rock relative permeabilities may sometimes need to be modelled, depending on well/fault pattern, production mechanism and simulation parameter of interest. Observed pressure differences, 4D seismic data, well tests, analogues and history matching may all be used to calibrate fault property predictions, although a change of scale is often required between empirical models and dynamic data. A prerequisite of any method used to optimise properties is knowledge of the range of values within which a particular property (e.g. fault permeability) may occur. Within limits, fault permeability seems relatively insensitive to strain, suggesting that statistics measured from small faults can be extrapolated to similar larger features. In situ stress and sample support should also be taken into account. In addition to fault rock property uncertainties, model uncertainty occurs for ‘geometrical’ parameters (sedimentary architecture, structure geometry, fault geometry). Fault properties may make a significant contribution to overall uncertainty, although the choice of modelling approach is highly dependent on data quality and the problem under investigation. Further work is required:

• To better define representative ranges for fault property values (fault permeability, thickness).
• investigate flow sensitivity to spatial variations in fault permeability and thickness.
• integrate fault and sedimentary uncertainties for reservoir modelling.
• evaluate different upscaling/downscaling approaches for fault permeability.
• quantitatively calibrate fault property predictions from dynamic data.

The two-term approximation of plane wave reflection coefficients at an interface is commonly used for AVO analysis. However, this approximation works poorly for large angles of incidence, where the anistropic effect, that can be caused by the presence of fractures, becomes significant. Riede et al. (2005) present a method for AVO analysis which represents accurately the reflection coefficients in the near-mid and large angle domains. For the purpose of characterizing fractures, whose effect is significant at high angles of incidence, we extend Riede’s method for azimuthal variations and investigate its use for fracture density discrimination.

We develop a technique for azimuthal AVO inversion of fracture properties. The new methodology is based on a linear approximation to the AVO response presented by Riede et al. (2005) for a single azimuth. The method makes use of prior information from the rock properties above and below the interface, given by well log or core data. The approximation is data driven, found by calculating the principal components -through singular value decomposition- of the modelled AVO response of the prior information. We illustrate the method and test its robustness for synthetic 3D surface seismic CMP gathers under different noise levels.

As an alternative for the removal of seismic acquisition footprint noise, a threshold filter using Discrete Wavelet Transform was applied to both 3D and 4D difference datasets from the North Sea. A very simple wavelet family was used, yet the results were quite satisfactory. A good separation of the noise was achieved which was verified by the lack of organized signal in the residuals other than the acquisition footprint and random noise. These are still preliminary results and there is plenty of scope, not only for improvement, but also for applications to different problems.

In recent years, the quest for better predictability of reservoir behaviour has led to the need to construct more complex and realistic models, where reservoir engineering and geosciences disciplines interact. Time-lapse (4D) seismic has been applied in both quantitative and qualitative manner to better understand and predict the dynamic behaviour of reservoirs. For stress sensitive geological systems, hydrocarbon production induces changes in the effective stress field within and around the reservoir, which result in expected effects, including changes in seismic velocities and layer thickness. Reservoir characterization and monitoring will therefore require the integration of flow, geology, geomechanics and time-lapse seismic. In this study, we have developed an integrated workflow that combines results from 4D seismic inversion, reservoir simulation and geomechanical modelling in a shared earth model. It has been applied to a high pressure, high temperature field located in the North Sea. This is a structurally complex field which poses a big challenge for a shared earth model approach. The 4D seismic have been inverted into interval time-shifts with a maximum estimated value of 5 ms (about 1 sample). In parallel, a coupled reservoir-geomechanical model has been constructed from seismic, dynamic and geological information. The interval time-shifts from the seismic inversion are then used to fine tune this coupled model in order to preserve the consistency with the 4D seismic response. Assuming vertical ray paths, we used a direct relationship between timeshifts and vertical strain. This is characterised by a dimensionless parameter, R, which represents the ratio of timeshifts that result from changes in velocity to timeshifts resulting from displacements. The fine tuning procedure referred to above involves the estimation of values for R that, when applied to the geomechanical model, will reproduce the observed seismic timeshifts response.

In this work a mature pore scale network model for oil depressurisation has been used for the first time to simulate typical core scales, initiating a new phase in the use of such techniques for core analysis. Important results clearly demonstrate the fact that it is now possible to reproduce the physical scale and pressure dependent balance of forces acting along the entire height of a vertically-mounted laboratory core during a solution gas drive experiment — without the need for upscaling pore-to-core methodologies. Now it has become possible to reproduce the complexity of an evolving gravity/capillarity force balance and investigate its nonlinear impact upon bubble break-up and coalescence phenomena throughout the course of an experiment. Using the macroscale approach explained above, we investigate the effect of varying the underlying Bond number of a simulation and examine sensitivities to the rate of depletion (bubble densities), the fluid properties, system scale, and the petrophysical characteristics of the sample. We show that relative permeabilities can be predicted according to the particular flow regimes exhibited by gas (dispersed and/or continuous) and demonstrate how flow is largely determined by the size and density of gas clusters, whether originating from nucleation or from break-up of larger structures during migration. In conclusion we show the different ways in which gas saturation gradients can develop along the height of a core sample. The results are compared against available experimental data — specifically, in situ gas saturation profiles and production histories — from equivalent sized samples (10 centimetres in height). These comparisons are utilised to provide a physical description of the mechanisms taking place during the experiments.

Although experimental work for solution gas drive processes is routinely carried out and interpreted for the purpose of defining critical gas saturations and relative permeability data, developing a thorough understanding of the results to facilitate confident application to the field is a hard task. Unfortunately, existing macroscopic models are unable to reproduce or take into account several different features of solution gas drive experiments. For example: (i) the impact of outlet boundary conditions upon the formation of saturation gradients, and (ii) transitions between different flow regimes (disconnected immobile gas, disconnected mobile gas, continuous gas flow) that are characterized by different pore-scale dynamics. These features can be considered using pore-scale modeling techniques. However, such microscopic applications, even in the few cases where they are refined enough to include viscous or gravitational forces, have perhaps their greatest limitation in the number of pores that can be simulated. This consequently makes it virtually impossible to simulate flow regimes that are not capillary dominated at an appropriate physical length scale. In this work a pore scale network modeling approach is presented which is capable of reproducing gravitational effects during oil depressurization through simulation in samples of macro-scale height. The models used comprise several hundred thousand pore elements. The primary objective is to fully simulate the physical height of routinely used laboratory samples, in this way reproducing the real scale and pressure dependent balance of forces: this allows us to show how the ratio of gravity to capillarity changes in relation to the rates of depletion (bubble densities), the rock, the fluid properties and the scale of the sample, and how this contributes to affect relative permeabilities and critical gas saturations. In particular it is shown that relative permeabilities can be predicted according to the particular flow regime operating during a given experiment (dispersed and/or continuous). Moreover we show that flow is largely determined by the size and density of gas clusters, whether originating from nucleation or from break-up of larger structures during migration. Furthermore, the ways in which different saturation gradients can develop over the length of a sample and the effects of the outlet boundary conditions upon such gradients are explained. The results are compared to available experimental data.

The ability to accurately characterise fault zone properties is of major importance with many of the world’s remaining hydrocarbon reserves being found in structurally complex reservoirs. The inclusion of accurate fault properties in both geological and simulation models has become a key part of the reservoir management strategy, but how can this be achieved? In an attempt to answer this question, the Total Geoscience Research Centre (GRC) devised a research programme. The result of the research programme was a workflow loop consisting of: •Detailed modelling of fine-scale fault rock properties •Generation of upscaled properties for reservoir simulation •History matching of the simulation model and feedback into the fine-scale geomodel Having developed in-house tools to address parts of the workflow, the GRC chose to sponsor specialised contracted research projects to provide additional solutions. A number of Joint Industry Projects were selected, mostly brokered through the ITF Structurally Complex Reservoirs Programme. The results produced by the ITF-based projects have provided a valuable addition to our knowledge and filled some of the gaps that existed in our workflow loop. However, a number of points require further investigation, principally relating to the accurate representation of fault zone properties within dynamic simulation grids. We feel that the future of fault characterisation research lies in developing: •Appropriate techniques to bridge the gap between fine and up-scaled fault properties, •Pertinent geomechanical tools, •A new generation of meshes and multi-phase fault-rock flow properties. A major research effort will also be required to develop characterisation techniques for faulted carbonate sequences.

The ability to accurately characterise fault zone properties is of major importance as operating companies seek to develop reservoirs that are increasingly structurally complex. Ideally, fault properties should be predictive, verifiable and consistent with all available data. However they must not be ‘tuned’ beyond the realms of geological realism just to achieve a production history match. A number of commercial software tools aim to provide answers to these problems, but it is our experience that there are often significant differences in the results from different tools, even for “standard” fault properties such as throw and SGR. This is related to a number of key issues: i) Are the data of the correct type and density for the case under consideration? ii) Is the fault zone adequately represented, or is refinement needed? iii) Is the level of detail sufficient to capture a representative volume for each fault? iv) Is throw computed accurately? v) Do factors other than lithology affect the fault properties? Seismic attribute data can increasingly be used in the prediction of fault zone properties. The accuracy of any such prediction relies on the availability of sufficiently high quality seismic data, in which case it can help to reduce uncertainty associated with populating fault zones in reservoir models. Fault zone permeability has received significant attention in recent years producing several different algorithms. These aim to predict the fault zone permeability from various factors, typically including an Shale Gouge Ratio (SGR) type component. Care is required, however, when applying these algorithms to reservoirs not considered during their formulation, particularly carbonate systems or very-deeply buried reservoirs, as the results may be meaningless. Typically fault properties are predicted at the scale of the earth model, but the results may then have to be upscaled for use in a reservoir simulator. A potentially more important issue is the representation of faults in simulation grids. In certain complex cases, non-traditional gridding methods may be required to accurately represent the fault geometry. Fault characterisation has become routine over the past few years with many tools available that aim to predict fault zone properties. However, caution is required, particularly with complex reservoirs. The importance of honouring all available static and dynamic data should not be forgotten.

An environment and culture has been developed within the UK that enables both the oil industry and academic researchers to work easily together. Research providers in the UK are generally well organised into JIP or consortia in which a number of oil companies sponsor and advise on the same project. Research facilitators, such as ITF and ITI, play a key role in bring together Universities and companies that have common research interests. The weak link in this chain however, is the transfer of knowledge from one part to the other. Identifiable drawbacks come in the form of the usefulness and quality of the deliverables provided. On the other-hand, even if the results are efficient, industry partners do not always allocate sufficient time and staff to promote the added value of this research. Staff working in an operational environment often solve problems on a day-today basis. They try to solve their current and recurrent problems using “simple” and immediate solutions. They have a need for solutions that are both accurate but yet quick to apply. It is into this environment that we have to transfer the knowledge generated by the research JIP and consortia. Is this a role for the universities or for each company? Total opened the Geoscience Research Centre (GRC), 15 years ago, in the UK to support corporate R&D through a direct link with UK-based Universities. The GRC has four main objectives, i) operate research projects in geoscience and engineering, ii) co-operate with UK-based universities and innovative companies on research projects, iii) provide technical support to the UK subsidiary and iv) offer opportunities to young geoscientists and engineers. The GRC must play a role as the conveyor belt transferring needs and knowledge between the customer (industry) and supplier (Universities and SMEs). We also have to ensure that research will meet both our current and future needs. Ensuring that Universities provide efficient results in an easily manageable format. Recently the increased use of web-based deliverables has improved the ease of knowledge transfer. This is not enough however, we have to ensure that the results will look accessible to the non-specialist and guide them to the information they are after, also that any software that might result from a project can be easily included into our reservoir modelling workflows (programme plug-ins provide a solution to this point). Here is a summary of one oil company’s point of view. As research partners, universities probably have a similar list of comments from their side. We need to continue to increase our collaboration to ensure a win-win situation. Though continued collaboration between the petroleum industry and Universities our ability to transfer knowledge will improve and become smoother. We look forward to working in the future with Universities-based in the UK and around the world as a means to increase our knowledge and ability to produce increasingly more difficult and costly hydrocarbon reserves.

We present results of a joint PP- and PS-wave AVO inversion from a multicomponent dataset acquired offshore West Africa. We compare these results with the output of the single, P-wave only, inversion. In both realistic synthetic examples and real data the joint inversion produces a better estimation of the shear impedance reflectivity and of the pseudo Poisson’s ratio reflectivity compared with the single inversion.

We analyse a subset of the converted-wave data from the Valhall LoFS surveys 1 and 3 in order to investigate whether any time-lapse signal can be seen on the converted waves. Measurements of shear-wave splitting indicate a change in the magnitude of azimuthal anisotropy in the overburden between the surveys. After PSTM we observe a significant time-lapse anomaly at the reservoir level at the location of a production well. The anomaly is only found for azimuth sectors around the average polarization direction of the slow shear wave in the reservoir interval.

A Petro-Elastic model is often tuned to a given dataset using a trial-and-error fitting approach. A new approach is presented, where this ‘manual inversion’ is re-formulated as a global and almost automatic optimization problem. The parameterization in terms of saturations, pressure, and temperature has been simplified and embedded as an objective function within a simulated annealing algorithm. The implementation is object-oriented, allowing for easy modification, and improvement, as well as an easy selection of inversion parameters. It has been successfully tested on well data from West Africa and the North Sea. In both cases, parameters controlling the pressure sensitivity (frame compressibility) of sand and shale were inverted. The global optimization also provided valuable information regarding uncertainties on the model predictions and eventually on the data itself.

The Valhall Field in the central graben area of the North Sea is a chalk reservoir field. The reservoir units have high porosities that vary between 35% and 50%, but the matrix permeability only reaches 10 mD at maximum. Particularly at the crest of the anticlinal structure the flow is significantly enhanced by the presence of natural fractures. Due to the geomechanical properties of the chalk the reservoir rock is compacting considerably during production. This in turn leads to subsidence of the seafloor that currently amounts to 25 cm per year. The geomechanical effects have a strong impact on the anisotropy both in the overburden and the reservoir, as well as on observed time-lapse changes from seismic data (Olofsson et al., 2003; Herwanger and Horne, 2005; Hatchell et al., 2005; Barkved and Kristiansen, 2005). In 2003 the ‘Life of Field Seismic’ project (LoFS) began at Valhall with the installation of 120 km of OBCs over an area of 45 km2. Multi-component seismic surveys are acquired every 3 to 6 months. The shot locations form a 50 x 50 m grid; the data available for this study had been decimated to a receiver grid spacing of 300 x 300 m. Time-lapse analysis has so far been concentrated on the P-wave data of the LoFS surveys. However, the P-wave image is obscured by a gas cloud in the central part of the field. In this study we therefore investigate whether time-lapse changes can also be observed in the converted-wave data. Furthermore, we use the converted waves to analyse azimuthal anisotropy both in the overburden and the reservoir. We focus our study on a 4 x 4 km grid of ACCP locations covering the southern flank of the field where strong time-lapse anomalies had been observed on the P-wave data. We analyse data from the first and the third LoFS surveys, which are seven months apart.

Any quantitative workflow, designed to constrain reservoir models to 3D/4D seismic data, must rely on petro-elastic modelling (or PEM), which relates fluid and rock properties to elastic ones. Various scales must be accounted for: laboratory cores and well logs, geological and seismic grids, fluid flow simulator models. The petro-elastic model is generally a fine-scale model (“pem”), defined and calibrated for each specific case against core and logs data. Aiming a 4D history matching workflow at the flow model scale, we then need to validate the use of the logs-scale calibrated “pem” at a larger scale, vertically and laterally. In this paper we proposed a methodology to define an upscaled “PEM” (new set of relationships valid at reservoir-scale), by tuning a fine-scale existing “pem”, adjusting the most sensitive and relevant parameters, by an optimisation procedure.Some previous studies already addressed downscaling problems (from reservoir to geological/seismic scale), but no previous work has proposed any solution for an upscaled PEM. The main results of this study, using real field data, are the following: upscaling is necessary, depending on the degree of static and dynamic heterogeneity; the optimisation procedure is successful in calibrating a fine-scale “pem” to get a reservoir-scale “PEM”; the procedure is sensitive to the Backus averaging parameters, which must be defined carefully; this workflow is performed at wells in this study, but could be extended to reservoir scale, when a fine-scale geological model is available.

Solution gas drive is the mechanism by which bubbles nucleate and grow from dissolved gas present in oil when reservoir pressure is lowered below the bubble point — gasphase expansion subsequently drives oil to the wellbore. A 3-phase pore network simulator has been developed to account for the fundamental steps of such a depressurisation process, where the nucleation of bubbles is modelled as a function of the petrophysical parameters and the physico-chemical properties of the oil. Each bubble grows by solute diffusion and expansion, until it coalesces with other bubbles creating a continuous gas phase that can eventually migrate and be produced. The predictive capability of such a tool, in terms of oil recovery profiles, critical gas saturations and relative permeabilities, is enhanced by the fact that each network can be anchored to real core samples (using mercury injection capillary data). Here, the complex dynamics specific to heavy, light and critical oil depletion are explored. The bubble nucleation process, the impact of different fluid properties upon bubble nucleation and growth and the dynamics of depletion at different rates are also studied in great detail. The simulations show that bubble densities, supersaturation histories and gas evolution strongly depend upon the particular fluid under investigation, as a result of the complex interplay between different interfacial tensions, gas/oil diffusivities and dissolved gas-oil-ratios characterising the fluid. Interfacial tensions were found to play the most dominant role in determining bubble nucleation density and subsequent gas evolution. Sensitivities at different depletion rates finally confirm the relationship between nucleation behaviour and supersaturation that motivates higher bubble densities as depletion rates increase.

Literature data show a wide scatter in critical gas saturation when near-critical oils are depleted. Some controls have clearly been identified, such as the depletion rate, but cannot explain such a large scatter. In fact, the influence of several other parameters is still unclear. This paper sets out to study the effects of rock characteristics, core height and presence of initial water saturation, and the influence of the corresponding capillary to gravity ratios on critical gas saturation for solution gas-drive. The combination of three different approaches has been used to decipher the variability: 1. Firstly, pore network simulations have been performed to assess the influence of rock properties and Swi and design the experiments 2. Secondly, to check the validity of the pore network model predictions, several depletion experiments were performed with the same fluid (C1/C10) at 38°C and 400 bars initial pressure. During the depletion at 4 b/d depletion rate, fluids were recovered and measured at high pressure. In-situ saturation monitoring ensured a good control of saturation gradients within the cores. • The depletion behaviour of a 750 mD homogeneous sandstone is compared with a 4 mD rock. • The depletion performance of a dry Berea sample is compared with a 26 % initial water saturation sample • The depletion behaviour of a high permeability, one metre long Berea core is compared with a 10 cm tall sample 3. Thirdly, numerical simulations with both pore network and ECLIPSE simulators were carried out and compared with experimental data. It is concluded that: • Both the rock characteristics and the amount of initial water saturation have a very significant effect on critical gas saturation and experimental observations are in agreement with predictions of pore network models • Whilst there appears to be little evidence of the effect of core height on critical gas saturation, in-situ saturation profiles versus time and relative permeabilities are significantly affected. These issues are well described by the ECLIPSE simulations.

Although experimental work for solution gas drive processes is routinely carried out and interpreted for the purpose of defining critical gas saturations and relative permeability data, reaching a unique and complete understanding of the results for a confident application to the field is a hard task. Existing macroscopic models cannot reproduce or account for several different features of the experiments: the formation of saturation gradients and the relation with the outlet boundary conditions are still unclear subjects as well as what governs the transition between the stages of disconnected immobile gas phase, disconnected mobile or continuous gas flow. On the other hand, pore scale network models, in the few cases where they are refined enough to include viscous or gravitational forces, have perhaps their greatest limitation in the number of pores that can be simulated, which makes it difficult or impossible to simulate real life effects of non capillarity. In this work a pore scale network modelling approach is presented which is capable of reproducing gravitational effects during oil depressurisation through simulation in samples of macro-scale height, anchored to the real rock and including several hundreds thousands pore elements. The primary objective is to simulate at the same scale routinely used lab samples, in this way reproducing the real scale and pressure dependent balance of forces: this permits to show how the ratio gravity to capillarity changes in relation to the rates of depletion (bubble densities), the rock properties, the fluid properties and the scale of the sample, and how this contributes to affect relative permeabilities and critical gas saturations. In particular it is shown that relative permeabilities can be predicted at each phase flow stage (dispersed and/or continuous) and how flow is largely determined by the size and density of gas clusters, whether originating from nucleation or from break-up of larger structures during migration. Furthermore the modalities by which different saturation gradients can develop along the sample and the relation to the outlet boundary conditions are explained. The results are compared to available experimental data.

The current industry practice for hydrate prevention is injecting hydrate inhibitors at the upstream end of pipelines based on the calculated/measured hydrate phase boundary, water cut, worst pressure and temperature conditions, and the amount of inhibitor lost to non-aqueous phases. In general, systematic ways of controlling and monitoring along the pipeline and/or downstream to examine the degree of inhibition are very limited. It is also known that hydrate formation would result in some changes in the aqueous phase. Therefore, it should be possible to detect initial hydrate formation by detecting these changes. The focus of our work is to develop a warning system against initial hydrate formation either by detecting minute hydrate particles and/or any changes in the water structure. The aim is to give the operator adequate time to initiate remedial steps prior to massive hydrate formation/build up which could result in pipeline blockage. In this paper, we present a new approach for hydrate monitoring and warning system using physical property measurements such as dielectric properties and ultrasonic wave signal. The results demonstrate that dielectric properties at microwave frequencies has potential to be used as a downstream analysis and online analysis for detecting the initial hydrate formation and/or presence of hydrate particles and/or change in water structure due to hydrate formation. The ultrasonic signals (e.g., FFT and amplitude) could detect the presence of minute hydrate crystals and even nuclei, while there is no sign of hydrate formation from monitoring system pressure. The results are very encouraging and could potentially change the industrial approach to gas hydrate control strategy. This work is part of a comprehensive project which covers monitoring the degree of inhibition against hydrate formation to further improve the safety of offshore/deepwater operations, as well as optimizing inhibitor injection rate to minimize their environmental impact and improve the economical aspects of the development. Some of the methods developed in this work are equally applicable to reducing the risks associated with wax and/or salt depositions, further improving the flow assurance aspects of deepwater developments.

If the 2D kriging interpolator is written as a function k(x,y) of the coordinates (x,y), its expression proves equivalent to that of radial basis functions. This helps clarify the relationships between kriging and many other mapping techniques, especially splines. These relationships are discussed in the paper and many practical examples are given. Kriging with a constant trend and a De Wijs variogram (?(h)=Logh) is equivalent to 2D interpolation with harmonic splines. Thanks to its harmonic property, Logh is also associated with the familiar power spectrum of fractals. Kriging with a linear trend and a generalized covariance equal to h2Logh is equivalent to 2D interpolation with biharmonic splines. Thanks to its biharmonic property, h2Logh is also associated with the familiar power spectrum of fractals. Smoothing Splines calculate a function minimizing the sum of an energy functional – or regularization term – plus a distance to the data. This is equivalent to the computation of kriging with a nugget effect. In 2D, splines consist of calculating a function minimizing an energy functional, related to the stretching or a bending energy of a plate. The choice of this energy functional is equivalent to fixing the degree of the trend function and the covariance model for kriging. In other words, fixing the energy – or regularization - term of splines is equivalent to fixing the a priori model for kriging. There is a fundamental inverse relationship between the spline functional and the covariance. This fundamental relationship also applies in the frame of discrete Bayesian statistics. The consequence of this relationship on the spectral density is straightforward. In many of his writings Claerbout discusses the “geoestimation” problem. He defines the Prediction Error Filter (PEF) as the filter such that its output tends to a white spectrum. This means that the PEF is identical to that associated with the roughening spline operator derived from the inverse of the covariance. Kriging can be formalized in the frame of energy-based estimation techniques such as splines. As a result, the regularization term commonly used in inversion methods can be regarded as an expression of the prior knowledge about the phenomenon. This is due to the link between the inverse of the covariance function and the roughening filter implicit in the quadratic regularization term.

We present an analysis of C-wave PSTM images obtained on three 2D multicomponent (4C) lines acquired offshore West Africa. We applied an anisotropic PSTM sequence developed by the Edinburgh Anisotropy Project (EAP), characterised by a complex parameterisation and robust parameter estimation with a relatively fast workflow. We also analyse the overall contribution of converted waves in this area. The geological setting is not ideal for S-wave propagation due to the presence of low-velocity unconsolidated shales, which behave like mud. Nonetheless C-waves give improved images in the deeper part of the section and clear definition of the main faults, which leads us to believe that joint interpretation of P- and C-wave sections can help improve interpreter confidence. These lines were also processed (in parallel) by two contractors. There are large differences in the imaging results, which highlights a diversity of approaches to C-wave processing and the high sensitivity of C-waves to the parameters used.

Pressure transient analysis is a well established reservoir evaluation method. By analyzing pressure and pressure derivative curves from build-up and drawdown tests, it is possible to identify reservoir characteristic parameters and heterogeneities. However, much of the pressure data recorded during a well test may be dominated by wellbore effects that can mask reservoir characteristics and lead to erroneous well test interpretations. This is particularly true when the well production rate is controlled at surface and more than one phase is flowing. These effects, which are transient in nature, include phase change, flow reversal, and re-entry of the heavier phase into the producing zone.

This paper presents the results of experiments carried out at Imperial College to investigate the effects of phase redistribution and phase re-injection on pressure build-up data. Singlephase and two-phase flow tests were conducted with air and water. An experimental system was designed, which includes a reservoir connected, through a resistance, to the base of a simulated well. The "reservoir" is recreated by a pressurised vessel, while the "well" is simulated by a vertical pipe with an active height of 10.3m and a diameter of 31.8mm. The "well" was flowed at controlled rates to mimic those encountered in gas condensate reservoirs. After steady-state conditions had been attained, the "well" was shut in at the topand the associated transient phenomena monitored via distributed measurements of pressure, temperature, liquid hold-up and wall shear stress. Pressure build-up data were interpreted using established well test analysis techniques.The experiments provide a qualitative and quantitative understanding of the effects of gas rates, liquid rates and rising gas bubbles on wellbore phase redistribution and re-injection.

The results yield an insight into the corresponding impact on well test transient pressurebehaviour.