New methods boost 4D seismic role in reservoir management

Seismic receiver network placed at the seabed to get high repeatability in 4D seismic (Fig. 1; photo courtesy BP FARM.)

• Seismic source monitoring system to check the repeatability of the seismic source signal (Fig. 2 [29,656 bytes]).

Reservoir management today is a science of approximation when it comes to the rate and direction of fluid front movement. Optimal management requires up-to-date information throughout the entire reservoir volume. Access to the latest data on fluid distribution in a reservoir and knowledge of how the distribution is changing with time allow engineers to develop cost-effective strategies to get the most out of every field at the lowest possible risk.

In addition to static, or one-time, measurements, time-dependent measurements from various oil field disciplines help constrain, refine, and improve the accuracy of reservoir models. Time-lapse logging of fluid saturation through casing can show which zones are contributing to production and which are watering out or being bypassed.

Permanent downhole sensors provide continual observations of pressure, temperature, and other diagnostics of reservoir performance. These measurements supply crucial information about fluid behavior at the well location but fail in the vast interwell region.

Three-dimensional seismic measurements have routinely been relied on to provide interwell data. In the past, seismic surveys were mainly interpreted for structural features and stratigraphic variations within the reservoir, but they can also be sensitive to contrasts in fluid types. Applied in surveys separated by periods of production, time-lapse, or 4D-3D plus time-seismic images can map fluid changes in a producing reservoir.

Four-dimensional seismic has the potential to monitor hydrocarbon movement in reservoirs during production and could thereby supplement the predictions of reservoir parameters offered by the reservoir simulator. The changes in the seismic response can be attributed to changes in the reservoir due to production (i.e., fluid saturation changes in the reservoir pore space).

This article describes a set of new tools whereby data from the reservoir model are integrated with seismic data. The potential of such tools is demonstrated through a case study from a 4D survey over the Gullfaks field in the North Sea.

4D seismic monitoring

The basis for the 4D seismic method is quite simple. If a seismic 3D survey is repeated, the changes in the seismic response in the time lapse between the two surveys are attributed to changes in reservoir parameters (fluid saturation, pressure, and temperature).

The repeatability of the surveys is important. It can be demonstrated that the relative change in the seismic frequency response between surveys is:

The relative change in seismic response = The relative change in the seismic survey parameters + the relative change in reservoir reflectivity + noise.

That the relative changes in the reservoir reflectivity are of the order 0.5-5% repeatability implies that the relative changes in the seismic survey parameters have to be neglectable relative to this percentage range. Only then can the observed seismic response change be related to changes in the reservoir reflectivity.

There are various ways of fulfilling the repeatability requirement. Permanent seismic sensor arrays placed on the seabed, as implemented in the BP/Shell Foinaven FARM project is one example (Fig. 1 [47,819 bytes]). Another tool is seismic source monitoring systems, where any perturbations in the seismic source signal are recorded and subsequently corrected for to comply with the repeatability requirements (Fig. 2 [29,656 bytes]).

Once repeatability requirements have been met, the measured seismic response change or reservoir reflectivity change can be inverted to establish the desired changes in reservoir parameters such as fluid saturations, pressure, and temperature. This inversion step is delicate since different combinations of these parameters might lead to similar changes in the reservoir reflectivity. Hence the inversion is nonunique and has to be constrained with other types of measurements. These measurements are repeated well logs synchronized with the time lapse seismic surveys.

Having these repeated well logs as additional constraints facilitates inversion for the desired reservoir parameters in the interwell space using the 4D seismic. An expert system has been built for this purpose and might be "trained" to understand the relationship between reflectivity changes of the well logs and the changes in the desired reservoir parameters in these locations. Subsequently, this relationship can be applied to the full seismic 3D cubes and thereby populate the whole interwell space with the desired parameters.

Reservoir simulators predict the behavior of saturation, pressure, and temperature in the interwell space as a function of production time. The predictions are based on explicit prior knowledge of the 3D flow model and often history-matched to the well logs. However, this flow model is hampered with a high degree of uncertainty in the interwell space.

Since 4D seismic represents an indirect measurement of the flow model, the combination of 4D seismic and reservoir simulation offers an attractive way to reduce the uncertainty and thereby reduce the risk. In the following case this approach led to identification of bypassed oil and undrained reservoir compartments. Further, the sealing capacity of the fault network could be assessed and updated.

Gullfaks case study

In 1995-96, a time-lapse 3D seismic survey was shot over the northern part of the Gullfaks field in the North Sea.

More than 50% of the estimated recoverable reserves in Gullfaks have been produced since the start-up in December 1986. One important objective of the 4D study performed on the Gullfaks field was to identify potentially undrained reservoir compartments after 9 years of production (1986-95). In addition, changes in fluid saturation as a consequence of production were to be estimated by integrating a set of different data types (4D seismic, well logs, and reservoir models).

The top reservoir unit in Gullfaks is the Tarbert formation, which consists of 60-70 m of Bathonian transgressive sands deposited in a tide-fluvial dominated delta environment. These clean sands show an average porosity of 34%.

In the design of the 1995 survey, emphasis was put on being able to repeat the 1985 survey as close as possible. The recording instrumentation had changed much in 10 years time. However, on the most significant survey parameters source and receiver configuration could be repeated without sacrificing the capacity improvements gained by the multisource/multistreamer developments (Fig. 3 [75,986 bytes]).

Reflectivity, reservoir

There exist alternative ways to establish how the changes in reservoir flow parameters will modify seismic reflectivity.

Fluid substitution modeling predicts seismic reflectivity of the reservoir rock for various saturation levels of oil, gas, and water in the pore space. Seismic reflectivity is a band-limited signal; the most relevant signal components for reservoir parameter inversion are named seismic attributes. Fluid substitution modeling might reveal the attribute set that best characterizes the saturation changes in the pore fluids.

If a suitable set of well logs exists, the relationship between the reservoir parameters and the seismic attributes in the well locations can be established. Fig. 4a [139,015 bytes] is a crossplot of seismic attributes and the hydrocarbon saturation for the Gullfaks case for a limited number of wells.

This relationship can subsequently be fed to an expert system purposely developed for 4D seismic analysis. The expert system will then apply the relationship to all seismic data in the interwell space. Fig. 4b is an example of an output from the expert system.

The principle outline of the expert system is shown in Fig. 5 [52,491 bytes]. The interface technique used is referred to as geostatistical classification. The classification could be done without prior knowledge of the field through unsupervised classification methods or by combining the seismic attributes with prior knowledge (for example well log information) in supervised classification.

The basic principle behind the supervised classification is the assignment of an attribute response to a specific reservoir parameter. The prior information (called training data) is used to "calibrate" the classification: that is, to estimate the classification function that maps the points into the different parameter classes. The result of the classification will be maps or volumes showing the distribution of the reservoir parameters in space and the confidence of the classified result.

Fig. 5 shows the data flow of the seismic analysis system: Seismic attributes are generated from 3D seismic volumes, prior information is obtained from well logs or seismic modeling, and finally classified maps or volumes are produced showing the possible fluid distributions. These fluid distribution maps might be used to condition the reservoir model-building.

The 4D seismic fluid-mapping of the reservoir interval (Tarbert formation) on Gullfaks was done with this seismic classification system; the resulting fluid distribution maps are shown in Fig. 6 [145,661 bytes]. The classification was done through use of an equivalent set of attributes for the baseline survey (1985) and the time-lapse survey (1995).

Expert system

One crucial seismic attribute for the expert system is acoustic impedance. A new method based on the best feasible approximation (BFA) has been developed to invert from seismic measurements to acoustic impedance cubes.

The method uses error estimates from kriging of well logs to constrain the inversion. The resulting impedance cube is the best feasible solution honoring the seismic measurements, the well logs, and the error estimates (or accuracy requirements).

This method has attractive characteristics for 4D seismic. From the baseline survey the initial acoustic impedance cube could be estimated prior to production. On the time-lapse cube the initial impedance can be used as the prior knowledge, and, combined with adequate accuracy requirements, the impedance changes as a result of hydrocarbon production can be established:

Impedance (time lapse) = impedance (base line) + impedance change (production).

In the Gullfaks case the baseline survey (1985) and the time lapse survey (1995) were inverted and the impedance change computed. Under the assumption that no pressure change took place during the lapse, the impedance change could be attributed to saturation changes. In the Gullfaks case the changes in impedance from oil-filled reservoir to water-filled was of the order 9% as measured in the observation wells.

Combining data

The saturation distributions from the 4D seismic measurement offer an alternative to reservoir simulation. As noted earlier, the reservoir-simulated distributions rely on an accurate knowledge of the flow properties of the reservoir. This accurate knowledge is available only in the wells and hence 4D seismic results might be combined with the reservoir simulations to reduce the uncertainties in the saturation distributions.

A new methodology for a composite analysis has been developed and named Statistical Reservoir Volume Analysis. The basis for the method is to compute the total change in saturation composition per reservoir compartment, accounting for the influx of water (Gullfaks has water injection for pressure maintenance) and the production of hydrocarbons in the time period between the surveys.

The reservoir-simulated results are history-matched, and the input and output fluid flux has been included in the modeled fluid distributions. This enables the hydrocarbon volume to be thresholded for the specific saturation level that according to the crossplot in Fig. 4a gives a seismic attribute response detectable by the expert system. Together with the top reservoir interface and the fault planes, the thresholded interface can be thought of as a closed geometric "container" for hydrocarbons above a certain saturation level. The hydrocarbon volume in such a "container" might be expressed:

Hydrocarbon volume = average hydrocarbon saturation x average porosity x height of "container" x area of "container."

If the same procedure is applied to the 4D seismic results, the saturation and the average porosity are assumed to be identical in the two situations, and the "container" is now derived from the 4D.

If the hydrocarbon volumes from the two "containers" have been "material-balanced" as explained above:

Height of "container" x area of "container" = height of "container" x area of "container"

implying that the volume of hydrocarbons in a given compartment as computed from the 4D seismic and the reservoir simulation have to be equal. However, the geometric shape of the "container," as defined by height x area, might be different.

Statistical Reservoir Volume Analysis allows the user to assess the match between 4D results and the simulation results in a statistical framework. The probability of the match between the hydrocarbon distribution as seen by the 4D and the geometric "container" as predicted by the simulation is computed with a cost function per reservoir compartment.

Fig. 7 [54,998 bytes] shows results from the analysis.

Fault-sealing analysis

How flow units of the reservoir are connected represents crucial information for optimal depletion of hydrocarbons. The sealing capacity of the fault network is a typical problem in this context.

For the Gullfaks case study a new method for fault-sealing analysis was tested. Basically, the saturation changes during production as inverted from the 4D seismic are combined with the information of the fault network and checked if the changes have taken place across the faults. The output for such an analysis is displayed in Fig. 8 [125,273 bytes].

In the context of the statistical reservoir volume analysis, the fault-sealing analysis might be looked at as a situation where there is a mismatch between changes of the shapes of hydrocarbon "containers" derived from the 4D seismic and from the reservoir simulator. If the 4D shows a saturation change across a fault that has been modeled as sealing, it will be color-coded as nonsealing, and the reservoir flow model might be updated and the simulator rerun.

Well results

Based on the time-lapse seismic results, a long horizontal well (C-36) was planned by Statoil to reach four reservoir compartments showing poor drainage after 9 years of production. The four compartments indicated by arrows in Fig. 6a are separated by major faults (vertical black zones).

Compartment 2 was predicted to be partly drained, based on the results from the flow simulations, while the 4D seismic analysis predicted it to be more or less undrained. The new well C-36 confirmed the results from the 4D analysis.

At present well C-36 has reached compartments 1, 2, and 3, confirming the changes in saturation predicted from 4D seismic classification. Technical problems and saturation indicators in block 4 both have combined to stop the advance of this well.

An abandoned producer (well B-13) in the southern part of the field was side-tracked to reach areas where the

4D seismic predicted no drainage. This well now produces about 1,000 cu m/day.

Acknowledgment

The Gullfaks 4D study was carried out in cooperation with Statoil. The authors acknowledge the contribution from Lars M. Pedersen, Lars Kristian Stroenen, and Martin Landroe. We would also like to thank the partners in the Gullfaks field, Saga Petroleum ASA and Norsk Hydro, for permission to publish this work.

The work of this project has been partly funded by the European Union's Thermie Research program (OG 117/ 94 UK).

The Authors

Helene Veire is a research scientist for Geco-Prakla, which she joined in 1995 after receiving a master's degree in geostatistics from Norwegian Technology University. She currently works on 4D projects.

Satyavan B. Reymond is a project leader at Schlumberger Geco-Prakla, doing research on new seismic facies characterization tools (3D and time-lapse seismic). He has a PhD in earth sciences on 3D sequence-stratigraphy from the University of Lausanne (Switzerland).

Claude Signer is currently a project team leader for Schlumberger Geco-Prakla in Stavanger. He works on developing seismic attributes analysis systems. Signer is a graduate of University of Geneva with a PhD in Geology.

Per Ola Tenneboe is a computer research scientist with Schlumberger Oilfield Services. He holds an MSc in optical electronics and a BEng (Hons) in computer and electronic systems from Strathclyde University.

Lars Soenneland is new product development manager for Geco-Prakla. He holds a PhD in applied mathematics from the University of Ber. He worked in geophysical applications until 1989 and during 1990-96 held several technical management applications, playing major roles in development of 3D seismic, the Charisma seismic interpretation software system, and seismic reservoir characterization. He currently manages 4D seismic and multicomponent analysis projects.

Copyright 1998 Oil & Gas Journal. All Rights Reserved.