SPE 128417

Characterization and Simulation of a Complex Fractured Carbonate Field Offshore Tunisia Riadh Bejaoui, Ryadh Ben Salem, ETAP, Hamed Ayat, Prism Seismic, Imen Kooli, ETAP, Dave Balogh, Gary Robinson, Theodore Royer, Prism Seismic, Taoufik Boufares, ETAP, Ahmed Ouenes, Prism Seismic

Copyright 2010, Society of Petroleum Engineers This paper was prepared for presentation at the SPE North Africa Technical Conference and Exhibition held in Cairo, Egypt, 14–17 February 2010. This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents of the paper have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect any position of the Society of Petroleum Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written consent of the Society of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract This paper describes a workflow that fully utilizes the pre-stack and post-stack seismic attributes to derive reliable geologic and fracture models that are validated by multiple blind wells and reservoir simulation. The first step in the workflow is to run post-stack seismic processes, which includes post-stack inversion and spectral imaging. The second step consists of applying a pre-stack seismic process called elastic inversion which will lead to various key seismic properties that provide good discrimination between reservoir facies. The third step consists of using the various post-stack and pre-stack seismic cubes to derive 3D geologic and fracture models. The fourth step is to use the derived models in a reservoir simulator to verify the validity of the models. This workflow was applied to a complex fractured carbonate field in offshore Tunisia which produces oil from the El Gueria formation. A large number of post-stack and pre-stack seismic attributes were generated in time and then depth converted within a 3D geocellular grid. These seismic attributes were used as input in REFRACTTM, Prism Seismic fracture modeling software, to create geologic and fracture models. The resulting porosity and permeability models were put into EclipseTM reservoir simulator software. Individual well performances were matched at eighty percent of the wells, confirming the reliability and accuracy of the derived geologic and fracture models and the usefulness of the workflow. Introduction The simulation of fractured reservoirs remains a challenging task for most E&P companies. The main difficulty when simulating these reservoirs continues to be the characterization of the flow properties of the fractures at a reservoir simulation scale. Most of the current fractured reservoir models use transmissibility multipliers around the wells to improve the match of the well performances. Unfortunately, the use of local transmissibility multipliers reduces the predictive capabilities of the models since these changes are usually arbitrary and are not strongly supported by geologic and/or geophysical information. Since most of the data (logs and/or cores) that could provide an indication on the density of the fractures are available at a centimeter scale and the reservoir simulation gridblocks are tens of meters wide and few meters high, the estimation of reservoir simulation scale permeabilities poses a major problem for most users. This problem is divided into two sub-problems: 1) how to distribute the centimeter scale fracture density information available at the wells in the 3D reservoir volume, and 2) how to turn these 3D fracture distributions into fracture flow properties computed at a reservoir simulation scale. These two issues are addressed simultaneously using the robust and proven workflow available in Continuous Fracture Modeling (CFM). The solutions provided by the CFM approach are illustrated with a complex fractured carbonate field in offshore Tunisia where a detailed fracture density model was built in 3D and used to estimate reservoir scale permeabilities that match most of the well performances. A Complex Fractured Carbonate Field Offshore Tunisia The studied field is a NW-SE extending elongated faulted anticlinal structure (Fig. 1) that produces oil from the El Gueria limestone formation. The reservoir facies (nummulitic limestone) is about 70m thick in the central part of

2 128417

the field, but it rapidly deteriorates to poor reservoir quality rock along the north, northwest, and southeast flanks of the structure. The field was discovered in the early 1970s, and it produced under primary recovery until the mid 1970s, when water injection was initiated. Platforms are used to produce the oil from a large number of wells. The reservoir produces from the matrix, which has its permeability enhanced by fractures. The reservoir behaves as a single porosity system, where the storage is in the matrix and the permeability of the matrix is enhanced by fractures in some places. A 3D seismic survey was shot in the late 1990s. The 3D seismic consists of 678 inlines with a spacing of 25m and 2779 crosslines with a spacing of 12.5m. The processing of the seismic produced the usual post-stack seismic cube sampled at 2ms as well as three stacks: a near-angle stack (0˚-15˚ angles), a mid-angle stack (15˚-30˚) and a far-angle stack (25˚-40˚). Six wells had shear sonic logs, which is necessary for the elastic inversion that provides some of the key seismic attributes used in the described workflow. Workflow-Overview

The first step in the workflow is to interpret the horizons and faults. Using the key horizons and all of the faults, a structural framework is built in the time domain. The second step consists of generating post-stack seismic attributes through the use of post-stack seismic inversion and spectral imaging. The third step consists of using pre-stack elastic inversion to generate the basic pre-stack attributes, Vp, Vs, density as well as the subsequent elastic properties such as Rigidity (Mu) and Incompressibility (Lambda). The various 3D seismic cubes are depth converted and used to derive 3D geologic and fracture models which are validated with a large number of blind wells. The fourth step is to use the derived models in a reservoir simulator to verify the validity of the models and to design reservoir management strategies to optimize recovery. The details of this workflow are described in the next sections. Geophysical Modeling The use of seismic data in reservoir modeling involves structural interpretation, seismic-well ties, high-resolution post-stack and pre-stack inversions, and spectral imaging. The key to successful use of seismic attributes in reservoir modeling is to generate attributes that are related to rock and fluid properties. When using the post-stack seismic, the two most useful processes that can be applied to the seismic data are high-resolution post-stack inversion and spectral imaging.

Spectral imaging is a process that allows the examination of the seismic data in specific frequency ranges. The resulting seismic attributes are frequency dependent or represents statistical properties that describe the behavior of the seismic power spectrum at different frequencies. Many of the seismic attributes derived from spectral imaging reflect variations in fluid and rock properties but they remain rarely used in reservoir modeling unlike the seismic attributes derived from post-stack seismic inversion. The post-stack seismic inversion produces acoustic impedance (the product of velocity and density). The high-resolution post-stack impedance is the result of a combination of the log and seismic data. In this project, many wells had sonic logs and could be used in the inversion workflow. The inversion process starts by tieing the well data with the seismic, and using the resulting time-domain impedance log data, along with a structural framework created from the seismic horizons and faults, to populate the impedance values within the a priori model. The inversion process will update this initial model, so synthetic traces generated from the model best match the seismic traces. Like most of the carbonate fields, there is a strong correlation between the impedance and the porosity, which could be used when modeling the porosity. However, the availaibility of pre-stack data and shear logs could be used in elastic inversion that could provide more than just an acoustic impedance derived from the post-stack inversion.

When working with pre-stack seismic data, many processes could be applied to derive very useful seismic attributes. The most common of these processes are the Amplitude Versus Offset (AVO), Amplitude Versus Azimuth (AVAZ), and Elastic Inversion. Although AVO and AVAZ are very useful pre-stack processes, their use poses many problems since the anomalies observed could be related to a multitude of factors that are not easy to distinguish. On the other hand, Elastic Inversion provides the means to derive rock elastic properties that could play a major role in reservoir characterization. Unlike post-stack seismic inversion that requires only compressional sonic logs at the wells, elastic inversion1 also requires shear sonic logs. Using both compressional and shear sonic logs, well ties can be done and the various

[128417] 3

elastic inversion algorithms are able to provide three key seismic attributes: compressional velocity (Vp), shear velocity (Vs), and density (Rho). These seismic attributes by themelves are very useful since the plotting of acoustic impedance (Zp, which is the product of Vp and Rho) versus shear impedance (Zs, which is the product of Vs and Rho) shows in many situations, the capabilities for discriminating lithologies and fluids2. However, the separation between lithologies and fluids is better reflected through the use of the Rigidity (Mu) and Incompressibility (Lambda). Elastic inversion also provides two additional attributes2, MuRho, which is the square of the shear impedance (Zs2) and Lambda Rho ( which is equal to Zp2 – 2 Zs2). Both of these attributes have shown an improved resolving power that allows a better discrimination of lithologies and fluids. Other techniques3 allow the derivation of new attributes that combines the elastic properties in different ways and provide a better separation of lithologies and fluids. For example, in the studied field when plotting the gamma ray log (GR) versus the shear velocity (Vs), one can see very distinct regions showing different rock types (Fig. 2). Although all of these pre-stack seismic attributes were derived by applying simple cut-off values, one can directly extract from them the 3D distribution of key reservoir facies; their usefulness goes beyond the ability to distinguish between a reservoir facies filled with hydrocarbon and another one that has no pay. Since all of these seismic attributes are continuous properties available in the 3D reservoir model, they can be used quantitatively in reservoir characterization to build geologic and fracture models. Given the ability of these pre-stack attributes to distinguish between key reservoir facies, they represent some of the best drivers that could be put into the CFM workflow and the seismically-constrained sequential geologic modeling.

Sequential Geologic Modeling The considered carbonate reservoir exhibits strong lateral and vertical heterogeneities; consequently, building accurate 3D models of any petrophysical property is a very challenging exercise. The post-stack and pre-stack 3D seismic attribute volumes are used to provide data in the inter-well regions to generate more reliable petrophysical models. Given the availability of a large number of seismic attributes, the geologic modeling can be done in two different ways. One simple way is to use a geostatistical method, such as collocated co-kriging, to incorporate one soft dataset. Unfortunately, such methods cannot use all the seismic attributes simultaneously. Given that many seismic attributes contain valuable geologic information and could be very beneficial to include as a constraint when estimating geologic attributes, the use of a neural network approach is more appropriate. In this project, a sequential geologic modeling approach developed by Prism Seismic is used to estimate various geologic models. The sequential modeling approach uses multiple seismic attributes as a building block to estimate dependent petrophysical properties such as facies, gamma ray, porosity, water saturation, and permeability in a sequential manner. The sequential method first identifies the reservoir property that has the potential to be the primary property controlling most of the other subsequent reservoir properties. In this study, the facies was chosen as the primary reservoir property. Thus, the first geologic model built using a neural network approach will be a facies model constrained by multiple seismic attributes. The input at the wells is the facies logs where the rock was divided into four (4) facies. The constraints are the multiple seismic attributes available in the geocellular grid. The neural network will attempt to find a relationship between the facies at the wells and the seismic attributes also available at the wells. Once this relationship is found, it is applied to every cell of the grid to create a 3D facies model (Fig. 3). In the considered project, only one third of the wells are used to build the facies model and the other two thirds of the wells are used as blind wells for validating the derived facies model. The ability to use very few wells in the modeling and keep most of the wells as blind wells is the result of the use of key pre-stack seismic attributes that provides a direct indication of the facies. The facies model that provides the best prediction at two thirds of the blind wells will be considered as the best facies model. This facies model can subsequently be used to estimate a secondary reservoir property. The secondary reservoir property in this study is gamma ray, which is available at many wells. In estimating the gamma ray, one can use the newly created facies model in addition to the available seismic attributes. As a result, the gamma ray model (Fig. 4) is constrained by seismic attributes and the newly derived facies model. This process could similarly estimate all the subsequent reservoir properties, including the matrix porosity (Fig. 5) and permeability (Fig. 6). However, given the presence of fractures in this reservoir, the modeling of permeability requires some additional considerations since the matrix porosity and permeability are enhanced by the fractures.

4 128417

Continuous Fracture Modeling (CFM)

The CFM approach4-7 is a fracture modeling technique that has been proven with blind wells, actual drilling and reservoir simulation results given its ability to predict fracture density at undrilled locations and do so with the required vertical resolution of 2-3 meters. This unique characteristic of the CFM approach comes from its ability to truly integrate geophysical, geologic and engineering information in building the fracture models. The extensive use of key seismic attributes in the CFM approach guarantees the availability of inter-well information, which is critical in predicting the fractures. While most of the current efforts in using seismic data in fracture modeling are focusing on the use of azimuthal anisotropy (derived from a wide-azimuth or multi-component survey), the CFM approach provides the unique opportunity to use any seismic attribute, including, if available, those derived from azimuthal anisotropy8. The CFM approach does not attempt to extract fracture information from the seismic data itself, but rather uses the seismic data to estimate attributes that have geologic meaning and represent fracture drivers that control where fracturing occurs. It is common knowledge that lithology, structure, proximity to faults, and other geological factors control the location and intensity of fracturing in the reservoir. Most of these fracture drivers can be identified not just from wellbore data, but also from seismic data. Typical geological drivers include facies types, porosity, and distance to faults. For all these geologic drivers, specific seismic workflows lead to accurate seismic attributes that could provide a very good description of these geologic drivers, which are by their nature available in the 3D reservoir volume. For example, volumetric curvature9 not only allows the imaging of the faults of different scales in great detail, but also provides various key structural drivers to the CFM approach that are far superior to anything available in the past from simple surface curvature methods. It is also very common, especially in carbonate reservoirs, to estimate the impedance derived from post-stack seismic inversion since it is usually well correlated with porosity, which is a common fracture driver. Other post-stack processes, such as spectral imaging, provide a plethora of seismic attributes that play a major role in any fracture modeling effort. The use of various post-stack seismic attributes has been extensively used to solve a wide variety of practical problems encountered in complex fractured reservoirs10-14. In the considered field, the derived post-stack and pre-stack seismic attributes, along with the previously derived geologic models, will be used as drivers to estimate the distribution of fracture frequency that will be used to derive the reservoir simulation scale permeability. The estimation of reservoir scale permeability is accomplished in the CFM model in two steps using the REFRACTTM software. The first step involves the use of the fracture frequency computed at the wells and the geologic, geophysical and geomechanical drivers available in the 3D reservoir volume. The second step is to use the derived fracture frequency models and well test data to calibrate the models so that they have the correct magnitude corresponding to the reservoir scale permeability. In the first step, the CFM model uses the fracture frequency available at the wells and all the geophysical, geologic drivers derived in the geophysical modeling and the sequential geologic modeling. In the case of this project, the estimation of the fracture frequency model used many pre-stack and post-stack seismic attributes, along with all the derived geologic models including facies, gamma ray, porosity and permeability. In addition to the geophysical and geologic drivers, the estimation of the fracture frequency uses geomechanical drivers such as slopes and curvatures. All of these drivers are available in 3D as geocellular models. Many fracture frequency realizations were generated, and blind wells were used to select the best ones (Fig. 7). The fracture frequency models were also validated by examining the resulting fracture directions at different layers and comparing these directions (Fig. 8) to the known fracture directions available from the wells. Since the CFM approach does not use fracture directions as input, the ability of the derived CFM models to honor the known fracture directions adds validity and confidence to the derived fracture models. Since the geocellular grids are made of cells that are 100m x 100m in the X and Y directions, and the reservoir scale permeability must account for the effects of fractures, the fracture frequency models require a calibration to the well test data to capture the large-scale effects due to both the fractures and the grid block size. The calibration approach available in REFRACTTM uses the well test data to calibrate the fracture frequency at the wells and transform them into reservoir scale permeability. These calibrated reservoir scale permeability logs have now captured the large scale effects, and are used as input to a new REFRACTTM project. The drivers that will be used in this new project are the previously generated models of fracture frequency. The CFM approach is used once more to estimate the calibrated permeability in 3D using the calibrated permeability at the wells and the generated multiple realizations of fracture frequencies. The resulting reservoir scale permeability models (Fig. 9).

[128417] 5

are used as input in a reservoir simulator, and the model that best matches the well performances will be used for further modeling. Model Validation in a Reservoir Simulator The final matrix porosity model estimated from the sequential geologic modeling and the resulting best reservoir scale permeability derived in the CFM approach was used in EclipseTM in a single porosity/single permeability system. The PVT and flow functions were added to the simulation input and thirty-five years of production history was simulated. In the thirty-five years of history, the reservoir has undergone both primary production and water injection. The validation of a model is achieved when a majority of the individual well performances are matched without the need for any history matching. Out of the 43 wells, 34 wells (or 80%) of them have their pressure and water-cut matched within reasonable tolerances (Fig. 10) without any history matching. These matches include the difficult-to-match water-cut that was able to capture the water breakthrough. Four of the wells have an average match and five wells have a poor match. The good match of 80% of the wells confirms the ability of the derived model to account for the complex fracture network.

Conclusions This study produced the following conclusions.

1. The use of post-stack and pre-stack seismic attributes derived from spectral imaging, post-stack inversion and elastic inversion could provide the necessary seismic attributes needed to accurately characterize a complex fractured carbonate reservoir.

2. Seismically constrained geologic models, such as facies, and matrix porosity could be accurately estimated by using multiple seismic attributes.

3. The validation with many blind wells shows that continuous fracture models generated in REFRACT using a fracture frequency at the wells and a multitude of geologic, geophysical and geomechanical drivers could provide an accurate description of the fracture network.

4. The fracture frequency models were successfully calibrated with well test data to provide the reservoir permeability needed for the dynamic model.

5. The dynamic model was able to watch the individual well performances at 80% of the wells without any history matching.

Acknowledgments The authors would like to thank the management of ETAP for permission to publish this work. The authors would also like to thank Hatem ben Kilani and Mahmoud Abed for their valuable technical contributions to the project. REFRACT is a Trademark of Prism Seismic. Eclipse is a Trademark of Schlumberger References 1. Connolly, P., “Elastic impedance”. The Leading Edge, April, 1999, 438-452. 2. Goodway, W, Chen, T. and Downton, J.: “Improved AVO fluid detection and lithology discrimination using Lame petrophysical parameters; λρ, μρ and λ/μ fluid stacks from P and S inversions.” 1997 Internat. Mtg., Soc. Expl. Geophys., Expanded Abstracts. 3. Simm R., Kemper, M., Deo J.:”AVO Impedance: A New Attribute for Lithology and Fluid Discrimination,” paper presented at the 2002 Petroleum Exploration Society of Great Britain PETEX conference, London. 4. Jenkins, C., Ouenes, A., Zellou, A., Wingard, J.: “Quantifying and predicting naturally fractured reservoir behavior with continuous fracture models,” AAPG Bulletin, V. 93, No 11, Nov. 2009. 5. Ouenes, A.: “Practical application of fuzzy logic and neural networks to fractured reservoir characterization,” Computers and Geosciences, Shahab Mohagegh (Ed.) v. 26, no 7, 2000. 6. Zellou, A., Ouenes, A: “Integrated Fractured Reservoir Characterization Using Neural Networks and Fuzzy Logic: Three case studies “, Soft Computing and Intelligent Data Analysis in Oil Exploration, Elsevier, Amsterdam, 2003 7. Ouenes, A., Robinson, G., Zellou, A., “Impact of Pre-stack and Post-stack Seismic on Integrated Naturally Fractured Reservoir Characterization”, SPE 87007, presented at the 2004 SPE Asia Pacific Conference on Integrated Modelling for Asset Management, Kuala Lumpur, Malaysia

6 128417

8. Boerner, S, Gray, D., Todorovic-Marinic, D. Zellou, A., Schnerck, G., “Employing Neural Networks to Integrate Seismic and Other Data for the Prediction of Fracture Intensity,” paper SPE 84453 presented at the 2003 ATCE. 9. Al-Dosari S., Marfurt K., “3D volumetric multispectral estimates of reflector curvature and rotation,” Geophysics 71, no 5, 2006, P41-P51. 10. Ouenes, A., et al. “Seismically Driven Characterization, Simulation and Underbalanced Drilling of Multiple Horizontal Boreholes in a Tight Fractured Quartzite Reservoir: Application to Sabria Field, Tunisia,” SPE 112853 presented at the 2008 North Africa Technical Conference and Exhbition, Marakech March 2008. 11. Ouenes, A., et al..: “Integrated Property and Fracture Modeling Using 2D Seismic Data: Application to an Algerian Cambrian Field” paper SPE 109272, presented at the 2007 SPE ATCE, Anaheim. 12. Pinous, O. et al., “Application of an Integrated Approach for Characterization of the Naturally Fractured Reservoir in The West Siberian Basement: Example of Maloichskoe Field”, SPE 102562, presented at the 2006 Russian Oil & Gas Technical Conference and Exhibition, Moscow, Russia. 13. Christensen, S.A. et al. , “Seismically Driven Reservoir Characterization Using an Innovative Integrated Approach: Syd Arne Field,” SPE 103282, presented at the 2006 ATCE. 14. Laribi, M. et al..: “Integrated Fractured Reservoir Characterization and Simulation: Application to Sidi El Kilani Field, Tunisia.”, Journal of Petroleum Technology, August 2004, and paper SPE 84455 presented at the 2003 SPE ATCE, Denver.

Figure 1: Depth map showing the major faults and the multitude of wellbores drilled from mostly two platforms.

[128417] 7

Figure 2: Gamma ray (GR) plotted versus the shear velocity (Vs) plotted from one well. The circles are inversely proportional to the compressional velocity (Vp) and the background color represents the density (Rho). Notice the very distinct rock facies that appear as the result of using Vs, and Vp with other logs.

Figure 3: Cross section of the best facies model that was validated by two thirds of the wells kept as blind wells.

8 128417

Figure 4: Cross section of the best GR model that was validated by two thirds of the wells kept as blind wells.

Figure 5: Cross section of the best matrix porosity model that was validated with many blind wells

[128417] 9

Figure 6: Cross section of the best matrix permeability model that was validated with many blind wells

Figure 7: Cross section of the best fracture frequency model that was validated with many blind wells

10 128417

Figure 8: Fracture directions at the bottom of the reservoir showing the two most important fracture directions.

Figure 9: Reservoir scale permeability calibrated to the well test data.

[128417] 11

Figure 10a: Imposing the liquid rate (top left), the water cut (top right cyan), oil rate (bottom left) and pressure (bottom right red curve) are matched at well 07

Figure 10b: Imposing the liquid rate (top left), the water cut (top right cyan), oil rate (bottom left) and pressure (bottom right red curve) are matched at well 11

12 128417

Figure 10c: Imposing the liquid rate (top left), the water cut (top right cyan), oil rate (bottom left) and pressure (bottom right red curve) are matched at well 17

Figure 10d: Imposing the liquid rate (top left), the water cut (top right cyan), oil rate (bottom left) and pressure (bottom right red curve) are matched at well 50