Copyright 2003, Society of Petroleum Engineers Inc.

This paper was prepared for presentation at the SPE Annual Technical Conference andExhibition held in Denver, Colorado, U.S.A., 5 – 8 October 2003. This paper was selected for presentation by an SPE Program Committee following review ofinformation contained in an abstract submitted by the author(s). Contents of the paper, aspresented, have not been reviewed by the Society of Petroleum Engineers and are subject tocorrection by the author(s). The material, as presented, does not necessarily reflect anyposition of the Society of Petroleum Engineers, its officers, or members. Papers presented atSPE meetings are subject to publication review by Editorial Committees of the Society ofPetroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paperfor commercial purposes without the written consent of the Society of Petroleum Engineers isprohibited. Permission to reproduce in print is restricted to an abstract of not more than 300words; illustrations may not be copied. The abstract must contain conspicuousacknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.

AbstractSidi El Kilani field produces oil from a Type 1 fracturedcarbonate reservoir. The field has produced 42 mmstboil since first production in 1991, and has consistentlyout-performed expectations. As with many fracturedreservoirs it has proved difficult to describe and modelthe fracture distribution or to predict or control waterbreakthrough. Prior to this study three out of nineboreholes in the field encountered tight rock and are dryholes.

This integrated study used geophysical, geologic, andengineering data simultaneously to improve the reservoirdescription. The reservoir model was proven with thesuccessful completion of a 400m horizontal sidetrack ofone of the dry holes. The successful characterization ofthe fractures was mostly due to the extensive use of 3Dseismic data. Seismic impedance, velocity, density andcoherence were used as input in the integrated fracturemodeling process. Seismic data was also used to derivegeologic drivers. In all nineteen geologic and seismicdrivers were correlated, using artificial intelligence tools,against a fracture porosity indicator computed fromconventional logs. 3D permeability models were thenderived using a whole core porosity-permeabilityrelationship, and input into a reservoir simulator. Theproduction history of most wells was readily matched.

The drilling results indicate that the 3D fracture modelscan be used successfully to locate highly fracturedareas. This was possible by using an integrated fracture *Formerly with CTKP, ** Formerly with KUFPEC+Formerly with ETAP, ++Formerly with (RC)2

modeling approach. None of the individual geologic andgeophysical drivers gave a clear and accuratedescription of the 3D distribution of the fractures. Theirsimultaneous use with a fracture porosity indicator andartificial intelligence tools revealed the complexity of thereservoir and its fracture network.

General InformationThe Sidi El Kilani (SLK) field is located in the NorthKairouan Concession approximately 190 kilometressouth of Tunis. The field was discovered in May 1989and the Exploitation License was granted in December1992. KUFPEC has a 45% working interest and theTunisian Government Oil Company, ETAP, has a 55%working interest in the field. In December 1994, a JointOperating Company, Kairouan Petroleum Company(KPC), (now called Compagnie Tunisio – Koweitienne dePétrole, CTKP) was set up to manage productionoperations. The SLK field produces from a fractured carbonate(Abiod Formation) reservoir at a depth of about 1,600metres and is estimated to contain reserves (P50) ofapproximately 49 mmstb. At the beginning of 2002,approximately 42.1 mmstb oil had been produced andthe field was producing approximately 3,850 stb/d oilfrom four wells. SLK crude oil is 39° API gravity. Ninewells have been drilled on the field (two with sidetracks),resulting in six oil wells, two dry holes and one waterdisposal well.

Regional Setting The Sidi El Kilani field is located in North-Central Tunisiawithin the Pelagian Platform, that forms the forelandarea of the Atlas Thrust Belt. The region is part of theNorth Africa Craton, that has been subjected to severalmajor tectonic events from Palaeozoic to late Tertiarytimes. These include continental collision and uplift inLate Palaeozoic (Hercynian), followed by relativemovements between African and European plates in theMesozoic leading to Jurassic rifting (neo-Tethys) andeventually plate convergence with subduction, andsubsidence during Tertiary.The Tethyan extension from Mid Jurassic onwardscaused fault block rotation, which locally resulted in

SPE 84455

Integrated Fractured Reservoir Characterization and Simulation: Application to Sidi ElKilani Field, TunisiaM. Laribi*,SPE, ADCO ; H. Boubaker CTKP, B.J. Beck KUFPEC; H-K. Chen** Retired; K. Amiri-Garroussi KUFPEC; S.Rassas* ADCO; A. Rourou* ADCO, T. Boufares ETAP, H. Douik+ AGIP; N. Saidi ETAP; A. Ouenes++, SPE, EP Tech

2 SPE 84455

facies and thickness variations particularly of Cretaceouscarbonates in North-Central Tunisia. The dominant faulttrends associated with this tectonic phase are NW-SEand NE-SW. During Late Cretaceous compressive stresses resultedin thrusting which created asymmetric folds trending NE-SW. These movements were repeated in Eocene andUpper Eocene to Oligocene times associated with theAlpine/Atlassic tectonic events and the closure of thewestern Tethys Ocean. The SLK structure is a N-Soriented elongate faulted anticline that formed in atranspressional tectonic regime. A major wrench faultdisplaying dextral movement, indicating NNW principlestress field, cross cuts the field.

Structural Evolution The region is mainly characterized by two major post-Paleozoic tectonic cycles. The first phase is extensionaland occurred during the Triassic to Early Cretaceousperiod. The region was affected by N-S extensionalphase leading to the development of mainly E-W andNE-SW fault systems and inducing generation of tiltedfault blocks, horsts and grabens that controlled thelateral variation of thickness and facies distribution. Thesecond phase is compressive and occurred during LateCretaceous to Early Quaternary period and wascharacterized by repetitive compressive phasesseparated by relaxation and led to the generation ofmany tectonic unconformities. These tectonic phaseswere associated with volcanic activity since the MiddleJurassic. The SLK field is transacted by a NW-SEtrending sinistral wrench fault that divides the field into aNorthern and Southern compartments. It should benoted that while salt diapirs are well documented in thenorth-west and offshore areas of Tunisia, only a fewwells drilled in the North Kairouan Concessionpenetrated Triassic salt, and only one SLK wellboreencountered Triassic “evaporites”. Gravimetry data overthe North Kairouan area discounts the presence ofsubsurface salt.

Reservoir and Seal The Abiod Formation carbonates are up to 550 m thickand contain predominantly massive, micritic and chalkywhite limestones with some dolomitised and bioclasticzones. The deposition is primarily in a pelagic marinesetting. Based on clay content, the Abiod has been subdividedinto 3 members: The Lower Limestone Member containsrelatively clean limestones intercalated with shales andmudstones. The Middle Limestone Member is more clayrich with occasional shale interbeds. The UpperLimestone Member is generally clean and massivelybedded with few shale bands. The cleanest limestones within the Abiod Formationcommonly show matrix porosity in excess of 20%.Matrix permeability, however, is very low (< 1mD) due to

narrow pore throat size. Flow capacity is dependent onfracturing with zero or minor contribution from the matrix.

Fractures have been interpreted to exist as swarms in“damage” zones related to systems of tensional orwrench faults. While the majority encountered in wellbores are closed with calcite cement, open fractures,where present, regularly contain oil and provideexcellent flow potential. El Haria Formation shales form the principal seal for theAbiod. The combined thickness of Maastrichtian,Palaeocene and Eocene shales across Sidi El Kilanifield is more than 800 metres. 3D Seismic and Available Field DataA 3D seismic survey was recorded over the Sidi El Kilanifield in 1993 in two parts. SLK-Main, which forms thefocus of this study, comprised 149 km2 of 25m by 25m,30-fold data with an 8-70Hz Vibroseis source. SLK-South was acquired with coarser recording geometryand was not included in the current work. The data wereprocessed with the specific aim of providing a volume ofhigh resolution, zero phase, true relative-amplitudereflection data suitable to help discriminate betweenfractured and non-fractured reservoir. Data quality was avast improvement over the previous 2D surveys, and todate no reprocessing of the 3D data has beenattempted.

In addition to the 3D seismic data a VSP was availableat one well. Using these seismic data, time and depthmaps of the main horizons, Abiod and Aleg, werederived (Fig. 1). A variance (coherency) cube wasderived also from the seismic data and used to extractand interpret fault polygons (Fig. 2).

Porosity and permeability core measurement wereavailble at a few wells and were used to model theeffective porosity and permeability in the reservoir. Awhole core was also available at one well and was usedto find the relationship between fracture porosity andpermeability for the dynamic model.

Raw electric logs and many interpreted logs were usedto derive geological models of various reservoirproperties including effective porosity, water saturation,and volume of shale and limestone. Fracture porositywas estimated from different logs according to twodifferent approaches described in the next sections. Themodeling of the fracture models will use fracture porosityas an indicator of fractures. An integrated fracturemodeling approach briefly described in the next sectionwill be used to create 3D models of fracture porosity.

Integrated Fracture Modeling The effects of fractures on the development of manyreservoirs in the world is becoming a major issue.Inapproriate modeling methods fail to predict the

SPE 84455 3

fractures density in the interwell region. Integratedfracture modeling is able to answer many questions andassist in the development of fractured reservoirs.Integrated Fracture Modeling1-2 is a set of tools and aworkflow that allows the creation of fracture models thatutilize all the existing reservoir information. Thisapproach uses a continuous framework where manygeologic drivers could be incorporated in the modelingprocess. This approach was successfully tested in manyfields3-7 and stems from the simple observation thatfracture intensity depends on many geologic drivers (themost commonly known being, structural setting,proximity to a fault, lithology, porosity and thickness).Because all these drivers and their complex interactionmust be accounted for during the modeling process, theuse of a regular 3D grid model similar to the one used ingeologic modeling or in seismic cubes along with acollection of artificial intelligence tools has proved manytimes its viability for the creation of truly integratedfractured models. One of the striking features of theintegrated fracture modeling approach is its ability to fullyutilise seismic data, as will be illustrated in the nextsection. The project described in this paper wasconducted from May to September 2001 and themethods used describe the state of the art at that time.More dramatic improvements were recently made to theintegrated fracture modeling approach8 where all thedrivers are directly derived from the seismic data. Withthese new improvements, it is becoming clear that theproper use of seismic data could solve entirely thecharacterization problems encountered in most fracturedreservoirs. In the Sidi El Kilani study, seismic data wereextensively used to derive geologic drivers by combiningstochastic inversion results with geostatistics.

Stochastic Seismic Inversion The objective of seismic inversion is to convertbandlimited, wiggle-trace seismic data into acousticproperty traces. In this form, these properties may berelated to other petrophysical properties of greaterinterest to reservoir description and management. Theultimate value of seismic data is the wealth ofinformation it can provide in the interwell space, espe-cially for interwell reservoir heterogeneities that maycontrol hydrocarbon distribution. Maximizing thispotential contribution of seismic-based information,however, is dependent upon how the information isextracted. The use of properly conditioned seismicinversion is one method of producing acoustic property(impedance) traces.

A major factor considered in the inversion methodologyis the precise well-seismic tie established for each welllocation. The process of making these ties includesestimation of the inherent waveform, and adjustments tothe depth-time relationship. It is not possible todecouple the position of the reflection coefficients in timewith the wavelet extraction (i.e., they must besimultaneously estimated). The convolutional equation

has one known (seismic trace) and two unknowns(reflectivity trace and waveform). The seismic pilot traceused for wavelet extraction is formed from a time-variantdip stack of a number of traces in the well vicinity.Precise depth-time functions permit sample-by-samplecorrelations between the well logs and the seismic data.This is extremely important if the inversion traces are tobe used with the well logs in any subsequent reservoirproperty modeling (e.g., facies and/or porosity).

Wavelet estimations and simultaneous depth-timeanalyses were performed using the available log andseismic data. The pre-inversion processing stepsinclude: spectral whitening, FX deconvolution, sparse-spike reflectivity estimation, polarity reversal, model-based reflectivity balancing and sub-seismic resolutionimpedance modeling. Both deterministic and stochasticinversion algorithms were applied to the dataset.Although the original seismic data was sampled at 4ms,the stochastic inversion results were provided with avertical resolution of 1 ms. The end products of thestochastic inversion are the impedance (Fig. 3), density(Fig. 4) and velocity (Fig.5) cubes.

A variance (coherency) cube was also provided. Thevariance data provides information on abrupt changesfrom trace to trace, which facilitates identification ofseismic faults. The variance cube is based solely onseismic data and therefore has a poor vertical resolution,while the stochastic inversion is derived from well data,giving the inversion a much higher vertical resolution.This high resolution seismic information could be usedwith log data and geostatistics to derive effective rockproperties.

Rock Property ModelingThe main objective of this task is to construct seismicbased rock property models of the Abiod reservoir to beused in the fracture modeling phase of this project. Themodels to be constructed include:

1. Facies model2. Effetive Porosity model3. Effective Permeability model4. Effective Water saturation model5. Volume of clay model6. Volume of limestone model

These models will aid in the understanding of theeffective rock property of the Abiod reservoir and will beused as drivers in fracture modeling. The first step in the geologic modeling phase is thefacies modeling. It is very important to understand thefacies distribution in the reservoir. In the case of Sidi ElKilani, the major facies present is the limestone, withdolomite present only near the faults. The facies modelwas derived by using lithology logs and the seismiccubes derived from the stochastic inversion. A 3Dmodel of limestone percent is shown in Fig. 6.

4 SPE 84455

Effective Porosity models were constructed based onthe facies model. The process used for the porositymodeling was the Sequential Gaussian Simulation withCollocated Cokriging. The collocated soft data were theseismic impedance cube, and the log used was theeffective porosity of the matrix. Variograms and porosityhistograms are also used as input in this process. Theresulting 3D porosity model is shown in Fig. 7. Theexamination of this porosity model shows that the highlyproductive areas which corressponds to the highlyfractured zones appears to have low matrix porositywhich is an expected result. The next geologic driver isthe effective permeability.

The permeability model is based mainly on the derivedporosity model as well as the permeability logs (ifavailable). In this case, the core data relationshipbetween core porosity and core permeability is used inplace of the permeability logs.

The method used for generating the permeability modelis the cloud transform. This method preserves the spatialrelationship between the core porosity and corepermeability. The variogram used in this case is thesame as the variograms that were used in the porositymodel. The next geologic driver is the water saturationmodel.

An Initial water saturation model was constructed usingthe impedance model. It is believed that the initial watersaturation in the fracture porosity is 0% and the fracturesif open are initially 100% filled with oil. The watersaturation model will be used as one of the drivers forthe fracture modeling.

Given all these geologic drivers derived from thesimultaneous use of the high resolution inversion cubes,log and core data, and geostatistics one can start theintegrated fracture modeling

SLK Fracture Drivers To fully understand the reservoir, 3D fracture models willbe created. Prior to the modeling process the differentinput data that are used for the integrated reservoirmodeling process are examined.

The modeling process is initiated by considering a 3Dgrid with the following properties. The grid is divided inthe East-West direction with Nx = 80 cells, and in theNorth-South direction with Ny= 100 cells. Each cell is100m wide and 100m long. The vertical gridding isbased on 2 markers (Top Abiod and Top Aleg) thatdefine one zone, divided in 20 layers.

The total number of layers is 20 and the overall grid ismade by 80 x 100 x 20 = 160,000 gridblocks. The size ofthis grid is the one that will be carried on to thesimulation. All the input data used for the modelingprocess are defined on an 80 x 100 x 20 grid.

The availability of seismic data is critical in the SLKproject. The first benefit of the seismic data is to providea very high resolution structure map (Fig.1). Thestructure is used to compute derivative maps in fourdirections: East-West, North-South, Northwest, andNortheast . The second benefit is the fine resolutionimpedance model used in creating the lithology andporosity models, which is critical for the fractures since itcontrols the rock brittleness. The third benefit is toprovide variance (coherency) results and faultinterpretation that play a major role in SLK.

A fault interpretation was also available. Fig. 2 showsthe closest distance to the fault map overlaid with thefault pattern. There is a major strike-slip fault in theNorthWest-SouthEast direction and the major fault trendis parallel to the strike-slip fault. The distance to closestfault is computed over the entire volume, on the 20layers, which will be used as a driver.

In addition to tectonics and structural information, fourseismic cubes will be used as drivers: the seismicimpedance volume, the variance cube ,the velocity anddensity volumes created from the decoupled inversion.Finally, the 3D matrix porosity, water saturation, effectivepermeability and the lithology models will be used asdrivers. In addition to the geologic and geophysicaldrivers, a pseudo-thickness 3D volume was included inthe analysis. In all, nineteen drivers will be used as inputin the integrated fracture modeling approach. The nextstep is to define the fracture intensity.

SLK Fracture IntensityFractures will be described with different means usingboth image logs and conventional logs. Three indicatorswill be used to quantify fracture intensity: a) a SecondaryPorosity Indicator using the Sonic/Density-Neutron logs(SPI 1), b) a Secondary Porosity Indicator using thelaterologs (SPI 2) and c) a 3D model of the fracturecounts (FC) using the image logs.

The SPI1 fracture porosity indicator was derived usingthe difference between density-neutron porosity and theone derived from sonic logs. The logic was that sonic logsees mostly the matrix porosity while the density toolsrecords the bulk density which includes both matrix andfracture porosity.

The second fracture intensity is the Secondary PorosityIndicator SPI2 derived from the invasion effect using thefollowing formula9

SPI 2 = [ Rmf ( RLLS-1 – RLLD-1 ) ]a

where SPI 2 : fracture porosityRmf : drilling mud resisitivityRLLS : resistivity reading of the shallow laterologRLLD : resisitivity reading of the deep laterologa : constante generally taken as 2/3

SPE 84455 5

Finally, the processed image logs on seven wells (SLK2to SLK8) were provided and used to define a fractureintensity based on the fracture count. A total of 1460fractures were found over the 7 wells. Out of 1460fractures, 276 were classified as open (18.9%) and 230were classified as closed (15.7%).

In most fractured reservoir studies, finding fractureintensity is a challenge. In SLK, there are three fractureindicators and the challenge is to find which one is themost representative. For each fracture indicator, 3Dintegrated fracture models were derived by usingartificial intelligence tools. For each fracture indicator,the fracture porosity and permeability were estimatedand input into a reservoir simulator.

SLK Fracture ModelsPrior to any fracture modeling the nineteen drivers wereranked, using a fuzzy logic algorithm2, to understand themajor geologic parameters controlling the fractureintensity considered. The examination of the differentranking results was a first test for the adequacy of eachfracture indicator. The examination of the ranking results provided by theuse of SPI1 indicates the importance of structural driversfollowed by the seismic indicators. When using SPI2, theseismic indicators became more important. Finally, theuse of the fracture count FC derived from image logswas mostly pointing to fault related drivers and structuralinformation.

Following the ranking, the modeling process is initiated.The methodology used to build fracture models is to usea sub-set of the data. The process, described astraining, will allow the neural network to adjust its weightso that the fracture indicator predicted by the neural netmatches the actual one. When using SPI1, 176 valuesfrom the wells were available. The modeling processdoes not use all the data and randomly picks 70% of thedata, (i.e. 104 values), for the training. The neuralnetwork is trained (Fig. 8) with a correlation coefficient of0.7 or higher on the 104 SPI1 values. Once the model isbuilt, testing its ability to predict the SPI1 fractureporosity at the remaining locations where the SPI1 isknown is performed (i.e. the remaining 72 SPI1 values).The models considered valid are those that achieve alarge testing correlation coefficient (Fig. 9), and are ableto reproduce the trends over the SPI1 value range.

As the fracture model building process depends on thewells selected for training, many realizations aregenerated. The only realizations used for furthermodeling are those that have a high testing correlationcoefficient and are able to reproduce the trends. Fiverealizations are kept, and another average realization ofthe five is also generated. Fig. 10 shows the upper layerfor the average realization when using SPI1 as a fractureindicator.

The same process is repeated with the SPI2 fracture

indicator and the upper layer of the average modelderived from five realizations is shown in Fig. 11.

For the third fracture indicator, the same process isrepeated using the fracture count. An averagecontinuous fracture model of fracture count was used toconstrain a Discrete Fracture Network (DFN) which wasin turn used to derive a fracture porosity andpermeability model3. Unlike the two previous fractureporosity models, the one derived from the DFN modelhas to go through multiple cumbersome steps andrequired additional fracture information such as fractureapertures and directions that were not readily available.

When computing the volumetrics of the reservoir usingthe three fracture porosity models, the fracture porositymodel derived from the DFN constrained by the fracturecount continuous model led to a very low oil in place -smaller than the oil already produced. Because, theSPI1 and SPI2 models were leading to bettervolumetrics they were kept for further modeling and thefracture count derived from image logs was deemedinadequate for SLK.

When comparing the two fracture porosity models usingSPI1 and SPI2, it is noted that the two models are verysimilar and tell the same story. As expected, all thefractures are fault related but the models show aninteresting feature where it appears that there are twotypes of faults: the “good” faults and the “bad” faults.The good faults still have the good fracture porositypreserved, whereas the bad faults have all their fracturesentirely mineralized. This difference between the faultsand the mineralization of the fractures is related to themigration of oil. When oil migrated through a fault and itssurrounding fractures, it protected the open fracturesand prevented their complete mineralization. As a result,a remnant fracture porosity that resembles vugs kept thefracture network partially open/connected and filled withoil. In the absence of oil, mineralizing fluids passingthrough the faults and fractures succedeed in cementingcompletely the fractures. A good illustration of thisphenomenon is shown in Fig. 12 where a core isshowing both situations. At the top of the core, a “vug”representing the remnant fracture porosity is visiblewhile at the bottom fracture porosity was completeleyocclided by layers of cementing minerals. The SLK fieldis a good example that illustrates the importance of theevents that happen after the appearance of the fractures(as a result of tectonic events). Using geomechanicalmodeling and structural reconstruction to derive a stress/ strain model was tried on SLK and led to the drilling ofa dry hole. The zone with the highest strain had all thefractures mineralized.

In addition to explaining many field observations, theSLK fracture models could be used as input in areservoir simulator.

6 SPE 84455

Porosity and Permeability for the Dynamic Models Since SLK has no matrix oil contribution it can besimulated as a single porosity reservoir where thefractures provide both the porosity and permeability. ADual porosity model was tested and proved its inability tomimic the reservoir past production and pressure history.These simulation tests confirmed the single porositynature of SLK - a conclusion supported by all thegeologic and production data.

Since the SPI1 and SPI2 represent the fracture porosity,they can be used directly as input in the reservoirsimulator. The initial water saturation in the fractures iszero and the only remaining unknown in the problem isthe fracture permeability. The only available informationwas a whole core analysis that gave a relationshipbetween the fracture porosity and the fracturepermeability (Fig. 13). This relationship - fracturepermeability is equal to fracture porosity to the power of3.6 - was used to derive a fracture permeability by usingthe derived fracture porosity models. The use of similarrelationships in a conventional reservoir will most likelylead to erroneous permeability models but in SLK, thefractures look like “rough pipes” and seem to exhibit apower law relation between fracture permeability andporosity. A good way to test the validity of thisrelationship is to input the fracture porosity andpermeability in a reservoir simulator and to evaluate itsability to reproduce the past history of the field.

History Matching The major problem with any reservoir characterizationand simulation effort is to be able to input the geologicmodels and reproduce the past reservoir history.Although “miracles” of reproducing reservoir history didoccur with conventional reservoirs where seismic datawas incorporated using geostatistical algorithms, itremains an unseen phenomenon in fractured reservoirs.Many problems plague the history matching process infractured reservoirs and the main one is the lack of anaccurate and systematic workflow for the estimation offracture porosity and permeability.

In the case of SLK, the effort spent in the geologicmodeling process led to realistic fracture porosity andpermeability models. The reservoir simulation wasperformed on the third generation simulator, SURE,using an unstructured grid (Fig. 14). The grid cell size inthe area of interest is about 100m and multiple cells withmultiple faces were generated to accommodate thecomplex geometry introduced by the two fault systemsand a horizontal well.

The observation of the pressure history in the north andsouth block indicates that a major water influx iscompensating for the oil production. Furthermore, afterfew years of production, the pressure is increasing dueto the increase in this water influx. Using numerousproduction and geologic data and testing many possiblescenarios, it was concluded that the water is coming

from deeper formations through the faults cut or near theexisting producing wells. When a well intercepted a fault,the water cut was high even during early production.When a well is close to a major fault, the well producesoil with a low water cut for a short time; then when thepressure draw-down effects reach the fault, water flowincreases and a rapid increase in the water cut isobserved. The best well is the one located far enoughfrom the fault to benefit from the fractures but still able toavoid the water influx problem. The water cut problem isvery complex and involve many factors that are notlimited to the distance to the nearest fault. The challengefor simulating SLK is to model correctly the water influxthrough the fault system. Many assumptions were madein this regard.

The first assumption is dictated by the variousobservations indicating that the external water is comingfrom an aquifer in the Triassic below the Abiod reservoir.This phenomenon is supported by a “water analysis”report that describes different properties of the producedwater. The next assumption made was the path of thedeep water source to the reservoir. It is apparent fromseismic observations that in SLK, the faults defining theflower structure originated deep below the Abiod. Hence,the flow path of the deeper water source could beupward along the conductive faults until the water getsinto the Abiod reservoir. In the reservoir, the fracturenetwork conducts the water to the wells, more so in theSouth block then the North block. The fracture networkin the south being very dense present a situation similarto a “text book” waterflood. The North block, howeverdoes not have such a dense and nearly uniform fracturenetwork, and thus the water path is more of a channelingeffect.

The location of the conducting faults was originallyselected based on the throw of the faults in theunderlying Aleg Formation. All faults with a throw higherthan 30 meters were considered as potential conductivefaults. This initial hypothesis required furtheradjustments, as it appeared that some faults with smallerthrow are also conductive and are responsible forchanneling water. The final permeability distribution forthe conductive faults is shown in Fig. 14 where cellsaffected by the presence of the conductive fault have apermeability larger than the background permeability ofonly 0.1 md. In other words, the model is allowing waterto channel from the lower 5 layers by forcing it throughthe conductive faults.

The Abiod reservoir was represented with 20 layerswhere the porosity and permeability derived from thefracture modeling was directly input. These values willnot be modified or changed. The SLK aquifer isapproximately located in the ten lower layers (layer 11 to20).

To model the water influx from lower formations, extralayers were added to represent the deepwater source.

SPE 84455 7

Seven more layers were added to the geologic model.Layers 21 and 22 simulate the conductive faults from thedeep water source, while layers 23 through 27 representthe deep water source itself. Layers 21 and 22 are in all200m thick and have 0.1% porosity and 0.1 mdpermeability except where the conductive faults arepresent (Fig. 14) where permeability varies between 20and 2500 md. These two layers act as an interfacebetween the Abiod and the deep source. The distributionof permeability shown in Fig. 14 was derived by historymatching. The 5 layers representing the water sourceitself have a porosity of 100% and a permeability of 0.1md and they are 375 meters thick. A cross sectionthrough all the wells (Fig. 15) shows the relativethickness of the additional layers as compared to theAbiod itself.

After estimating the permeability near the conductivefaults in layers 21 and 22 at the bottom of the Abiod andwithout modifying the fracture porosity and permeabilityin the 20 layers of the Abiod, the pressure and water cuthistory of all the SLK were matched as shown in Figs.16-19.

A first test for the validity of this match is the examinationof the water level. The final water level at the end of thesimulation (Fig, 15) was compared to actual fieldmeasurements and shows the same profile. Notice, howfracturing creates an uneven water level.

Based on the simulation and the geologic model manyreservoir management strategies were considered andimplemented. Among all these intiatives is the drilling ofa side-track from a dry hole.

Testing the Models with Actual DrillingIn the summer of 2002, a horizontal side-track wasdesigned and drilled (Fig. 20) The target was to reach ahighly fractured area west of the existing dry hole.Drilling was stopped due to major mud losses after 400mlateral offset and the final target was not reached withthe side-track. The well tested and produced oil and iscurrently producing 500 bbls/day. Following the drillingan image log was used to compare the models with theactual reservoir.

The first important conclusion is the comparisonbetween the SPI1 and SPI2 fracture models. From theactual FMI it appears clearly that the SPI1 modelprovides a better representation of the distribution of thefractures. In Fig. 20 the image log distinctly show fivedifferent regions. The first one is the region near theexisting dry hole where all the fractures are mineralized.The second region is showing an increase in openfractures that are represented on the FMI log by a largenumber of conductive features. The third region is abrutal drop in fracture porosity easily seen in the FMI logwhere conductive features are almost absent over anentire length of about 80m. The fourth region shows arapid increase in open fractures represented in the FMI

logs by distinct conductive features. Finally, the fifthregion is the decrease in fracture porosity that can beinterpreted also on the FMI logs where the vanishing ofthe fractures is evident.

The comparison between the actual FMI data with thefracture porosity model shows clearly that even whenusing a 100m model, one can predict very accurately thefractures in very complex reservoirs. This predictivecapability was possible by the appropriate use ofadequate fracture indicators which in this case turnedout to be the SPI1 secondary porosity indicator and theextensive use of seismic data in an integrated modelingapproach.

Conclusions Based on the extensive collaborative work done by alarge multidisicplinary team from May to September2001 on the SLK field the following conclusions can bedrawn:1. The extensive quantitative use of 3D seismic data inan integrated fracture modeling approach is a viablesolution to model complex fractured reservoirs.2. The Abiod fractures can be modeled in 3D and theremnant fracture porosity filled with oil could bepredicted if seismic data are available3. The secondary porosity indicator SPI1 is the bestfracture indicator for modeling SLK fractures.4. Once the fracture porosity is know in 3D, whole coredata analysis could provide a good estimation of fracturepermeability that seems to be related to fracture porositythrough a power law.5. The most realistic explanation for the water influx inthe Abiod formation is that water flows from lowerformations along conductive faults6. The most difficult unknow parameter in SLK is thepermeability of the faulted zones responsible for thewater influx in the Abiod. Pressure and water cut historymatching is the only way found to estimate thepermeabilty of these conductive faults.

Acknowledgements The authors would like to thank the management ofCTKP, KUFPEC, and ETAP for authorizing thepublication of this paper. Special thanks to the (RC)2 consultants who participatedin this project: Justin Hoffman, Gary Robinson (EPTech), Alex Sackefyio (ExxonMobile), Abdel Zellou (EPTech), and Andras Kiraly (HOT Engineering). This work would not have been possible without thefruitful discussions, comments and assistance providedby many people at CTKP, KUFPEC and ETAP amongthem Mike McMurty (KUFPEC), Sami Bahri (BritishGas), Attia Mohamed (ETAP), Moncef Chekir (CTKP),Ghazi ben Djemia (KUFPEC), and Faouzi Messaoudi(ETAP).

8 SPE 84455

References 1. Ouenes, A., Richardson, S., Weiss, W.: “Fractured

reservoir characterization and performanceforecasting using geomechanics and artificialintelligence,” paper SPE 30572 presented at the1995 SPE Annual Technical Conference andExhibition.

2. Ouenes, A.: “Practical application of fuzzy logic andneural networks to fractured reservoircharacterization,” Computers and Geosciences, ShahabMohagegh (Ed.) v. 26, no 7

3. Ouenes, A. and Hartley, L.J.: “Integrated FracturedReservoir Modeling Using Both Discrete andContinuum Approaches,” paper SPE 62939presented at the 2000 SPE Annual TechnicalConference and Exhibition, Dallas.

4. Zellou, A., Ouenes, A., Banik, A.: “ImprovedNaturally Fractured Reservoir CharacterizationUsing Neural Networks, Geomechanics and 3-DSeismic,” paper SPE 30722 presented at the 1995SPE Annual Technical Conference and Exhibition.

5. Zellou, A., Ouenes, A: “Integrated FracturedReservoir Characterization Using Neural Networksand Fuzzy Logic: Three case studies “, SoftComputing and Intelligent Data Analysis in OilExploration, (2003), Elsevier, Amsterdam

6. Zellou, A, Hartley, L.J, Hoogerduijn-Strating, E.H, AlDhahab, S.H.H, Boom, W., Hadrami, F. “IntegratedWorkflow Applied to the Characterization of aCarbonate Fractured Reservoir: Qarn Alam Field,”paper SPE 81579 presented at SPE 13th Middleeast Oil Show & Conference, Bahrain.

7. Gomes J.S., Ribeiro, M.T., Fouchard P., TwombleyB.N., Negahban, S., SAl-Baker, S.: “GeologicalModeling of a Tight Carbonate Reservoir forImproved Reservoir Management of a Miscible WAGInjection Project, Abu Dhabi, U.A.E.” paper SPE78529 presented at the 2002 Abu DhabiInternational Petroleum Exhibition and Conference.

8. Shen, F., Ouenes, A.: “Seismically Driven IntegratedFracture Modeling,” paper SPE 84460 presented atthe 2003 SPE Annual Technical Conference andExhibition, Denver.

9. Saidi, A. M.: Reservoir Engineering of FracturedReservoirs, Fundamental & Practical Aspects, TotalEdition Presse, (1987) ISBN 2-905 143-09-06

SPE 84455 9

Fa tim

Fintercube

ig. 1: Top of the Abiod derived by the combined use ofe horizon map, well tops, and geostatistics

igp

Fig. 3: Cross section of the impedance model

. 2: Distance to closest fault map showing thereted faults derived from the variance (coherency)

Fig. 4: Cross section of the velocity model

Fig. 5: Cross section of the density model

10 SPE 84455

Figsei

usi

. 6: 3D model of percent limestone derived usingsmic data and lithology logs

Fig. 8: Training results when using SPI1 as a fracture

Fig. 7: 3D model of effective matrix porosity derivedng seismic data, and porosity logs

indicator

Fig. 9: Testing results when using SPI1 as a fractureindicator

SPE 84455 11

Fig. 10: Upper layer of the SPI1 fracture porositymodel

Fmodfault

ig. 11: Upper layer of the SPI2 fracture porosityel. Notice the desctruction of porosity near some of thes.

pom

Fig. 12: Core showing at the top remnant fracturerosity in a large fracture and at the bottom completelyineralized fracture

12 SPE 84455

Figfrom w

Figunstrurespon

Figfree w

10000

13: Permeability-Porosity relationship derivedhole core analysis

Fig. 16: Pressure (white curve) and water-cut (blue curve) match at well 1.

1

10

100

1000

0 2 4 6 8

Fracture Porosity SPI (%)

K a

ir (m

d)

. 14: Permeability in Layer 21 showing thectured grid and the conductive zones near the faultssible for the water influx.

. 15: Cross section showing oil saturation and theater level at the end of the simulation.

Fig. 17: Pressure (white curve) and water-cut(blue curve) match at well 6.

Fig. 18: Pressure (white curve) and water-cut(blue curve) match at well 7.

SPE 84455 13

aftbef

Fig. 19: Pressure (white curve) and water-cut(blue curve) match at well 3.

Fero

ig. 20: Comparison between actual FMI recorded drilling and the fracture porosity model derivedre drilling. Notice the ability of the model to predict

a 100m zone free of fractures (red arrow) followed byanother very fractured zone (cyan arrow) and thevanishing of the fractures at the end of the wellpath.