Tu-P07-02 Workaround to Build Robust Facies Model with Limited Input Data - A Case Study from North West Kuwait A. Jaradat* (Schlumberger), B. Chakrabarti (Kuwait Oil Company), H. Al Ammar (Kuwait Oil Company), T. Al Adwani (Kuwait Oil Company), R. Srigiriraju (Schlumberger), A. Amer (Schlumberger), N. Banik (Kuwait Oil Company) & A. Abu Ghneej (Kuwait Oil Company) SUMMARY Five wells were drilled in North West onshore Kuwait, targeting the Jurassic Najmah/Sargelu formations. Three major depositional-units are interpreted within these formations viz; basal limestone, overlain by Kerogen and clean limestone with argillaceous intercalations towards the top and base. The objective of this study is to build a 3D facies model that can describe facies architecture across the field. Five litho-facies were described from core study in two wells. These litho-facies have been propagated to non-cored intervals in other wells utilizing a supervised Neural Network technique. The facies trend maps have been developed using various inputs such as paleo shoreface trends, sequence stratigraphic zones and facies fraction. The facies maps were constructed from facies fractions using trend maps. Subsequently variogram maps were built from normalized facies maps. These parameters are in close proximity when compared to regional facies trends. Variogram parameters were used to build the facies model. To validate this workflow a blind well test was run on one well and 65% success has been achieved at the well location and 82% over the whole model. These success rates are considered to be significant considering geographically wide study area and limited number of wells.
Workaround to Build Robust Facies Model with Limited Input Data; A Case Study from North West Kuwait Awni W. Jaradat, Schlumberger, Bhaskar Chakrabarti, KOC, Heyam Al Ammar, KOC, Talal Al Adwani, KOC, Ramachandra Srigiriraju, Schlumberger, Aimen Amer, Schlumberger, Nikhil Banik, KOC, Ali Abu Ghneej, KOC Introduction The objective of this study is to build a reliable 3D facies model that can describe facies architecture across the field and can delineate reservoir, and at the same time will be utilized to distribute porosity. Five litho-facies were described from detailed core study in two wells. These litho-facies have been propagated to non-cored intervals in other wells utilizing a supervised Neural Network technique. The issue at hand was the development of variogram with limited well data and its subsequent application in the stochastic algorithm with reduced uncertainty. Lack of data prevents conventional way to generate facies variogram. The facies trend maps that show the lateral variation of facies have been developed using various inputs such as paleo shoreface trends, sequence stratigraphic zones and facies fraction. The facies maps were constructed from facies fractions using trend maps. Subsequently variogram maps were built from normalized facies maps. Variogram parameters i.e. major, minor range and azimuth are inferred from variogram maps. These parameters are in close proximity when compared to regional facies trends. Analyzed variogram parameters were used to build the facies model. To validate this workflow a blind well test was run on one well and 65% success has been achieved at the well location and 82% over the whole model. These success rates are considered to be significant considering geographically wide study area and limited number of input wells. The produced model has been used to understand the lateral and vertical distribution of facies in the study area and is used successfully to distribute porosity across the field. Geology of the field Five wells were drilled in Kra Al-Maru structure targeting the Jurassic Najmah/Sargelu formations. Najmah/Sargelu constitutes of both, the reservoir and source rock, and is a well-known hydrocarbon play in Kuwait Confined by Gotnia salts/evaporates as top seal, Dharuma Shale as bottom seal and with organic rich black laminated mudstones (known as Najmah Kerogen). Kra Al-Maru structure is well defined four-way closures follows a NW-SE trend and is composed of more than one anticline structure making it an anticlinorium structure that extends towards the southeast Kahlulah field. The structure is intersected by numerous fault systems that exhibit no visible throw on seismic, except two NE-SW trending faults. Representative litho-stratigraphic column (Figure 1) Indicate that the Sargelu Formation mainly composed of limestone with inter-bedded wackestone and packstone. The thickness of the Sargelu is about 120 ft, and it is conformable with the underlying Dhruma and overlying Najmah formations. The Najmah Formation is the uppermost carbonates of the whole Jurassic section. It is composed of interbeds of cemented peloidal packstone, argillaceous and bituminous limestone. The formation is divided into two units, as shown in Figure 1. The lower unit represents the Kerogen, overlaying by argillaceous limestone and covered by considerably thin Kerogen layer. The Najmah Formation was deposited in restricted marine outer ramp and is quite uniform and can be correlated across Kuwait. It corresponds to the widespread, late Callovian flooding event which affected the entire Middle East platform.
Litho-facies Classification In the ideal world, the classification of litho-facies should be based on both petrophysical and geological information. However, this requires sufficient information from both disciplines. Detailed core description in two wells was used in the Identification of the litho-facies of the reservoir. From which, five litho-facies were identified. Due to the limited control points of the cored wells and to be able estimate the lateral distribution of each litho-facies with relative confidence, estimation of litho-facies at non-cored wells was required; Najmah Najmah LST Hardground Kerogen MFS - 3 Lower Kerogen Sargelu MFS - 2 Hardground Dharuma 10,000 m Figure 1: Well Section Showing the Facies Distribution of Najmah Sargelu Formation Neural Network Technique was used to estimate the litho-facies based on relationship between log response and specific litho-facies inferred from core description. The relationship between open-hole log response and identified litho-facies was examined. The log combinations clearly demonstrate relationships to the litho-facies. The core description of the cored section was used as a train model in the Supervised Neural Network to predict the litho-facies of non-cored sections and to generate a log derived litho-facies (= electro-facies) curve for all wells in the study area. For defining the Neural Network, GR, VCL, DT and PHIE logs were selected as these logs are able to distinguish between the different facies adequately and are also present in all of the wells. A supervised neural network model was defined in these two wells using the training facies log and the open-hole logs (GR, VCL, DT and PHIE). Once the neural network was run electro facies were generated for all wells in the field (Figure-1). The robustness of the generated facies log was validated and enough confidence was built up on the developed Neural Network facies logs. The developed facies log was successful in re-generating the training facies in the two cored training wells. Litho-facies Distribution Model The general workflow of facies modeling is given in Figure 2. The 3D facies modeling exercise has two steps: The first step is to generate the facies trend maps as explained above, and the second step is to model the generated litho-facies log in 3D reservoir space. Facies Trend maps In facies and petrophysical modeling, trend maps can be used to control the probability of Occurrence of certain facies or to describe the change of geometry of facies bodies across the field or to control
property values. A litho-facies trend map was prepared using various inputs such as paleo shoreface trends, sequence stratigraphic zones and averaged values (fraction of each facies in each zone) from the generated litho-facies log readings. The trend maps have to be normalized (i.e. the sum of facies fraction in each zone should equal to one). A simple workflow was developed in this work to be used in normalizing the trend maps. The workflow in summary starts with generating a summation map of all facies trend maps for the specific Well-log lithofacies Figure 2: Litho-Facies Modelling Workflow Variogram Lithofacies Probability Maps Sequential Indicator Simulation (SIS) Lithofacies model was built using geostatistical SIS method. Honoring the upscaledwell-log lithofacies data. Honoring relative fractions of each lithofacies. Honoring the depositional characteristics through use of lithofaciesprobability maps. Honoring sequence stratigraphic interpretation. zone, then each facies trend map in that zone will be divided by the generated summation map. This will assure that total facies fraction in each zone will be equal to one. One point of caution here to watch for is making sure that the maps value is to be between zero and one. Subsequently variogram maps were built from these normalized facies maps (Figure 3). Variogram parameters i.e. major, minor range and azimuth are inferred from variogram maps. These parameters are in close proximity when compared to regional facies trends. Analyzed variogram parameters were used to build the facies model. Kerogen Algal LST Figure 3: Probability Map examples and the Resultant Variogram Maps Facies Modeling After a careful up-scaling of the facies log into the 3D grid using the most of method, the litho-facies distribution model was constructed using the Sequential Simulation Indicator algorithm and
controlled by the litho-facies trend map. The resulted facies model captured the lateral distribution of the field The developed 3D facies model formed the basis and used as a constraint for the petrophysical modeling all other reservoir properties like porosity, permeability, water saturation. The facies model will also be used in the extensive STOIIP uncertainty analysis. A blind well test was run to validate the model which gave us the excellent opportunity to examine the predictive capability of the model. Figure 4 shows comparison of predicted and actual reservoir properties. The success rate achieved at well location is 65% and a rate of 82% was achieved over the whole model. Figure 4 shows beyond doubt that the developed facies model is a robust one and has good predictability for future drilling locations. Figure 4: Model Validation Conclusion Litho-facies classification based on core description and the relationship between litho-facies and open-hole log response was carried out using Neural Network Technique. Litho-facies distribution maps were very effective in guiding the facies distribution across the field. The variogram data obtained from facies probability maps are realistic and in conformity with regional facies belt. The workflow can be generalized and applied to other fields with limited well data References 1. 3D Geological Modeling of a Carbonate Reservoir, Utilizing Open-Hole Log Response - Porosity & Permeability - Lithofacies Relationship K. Akatsuka, Japan Oil Development Co., Ltd. 2. 3D Facies Modeling of a Jurassic Carbonate Reservoir - A Case Study from Kuwait Subrata K. Chakraborty, Christian Perrin, Schlumberger 3. Application of Neural Networks Technique in Lithofacies Classifications used for 3-D Reservoir Geological Modeling and Exploration Studies. - A Novel Computer-Based Methodology for Depositional Environment Interpretation. (X-Field Example, Niger Delta, Nigeria). EMEKA M. ILOGHALU, NNAMDI AZIKIWE UNIVERSITY, AWKA, NIGERIA. 4. Frequentist Meets Spatialist: A Marriage Made in Reservoir characterization and Modeling Y. Z. Ma, SPE, Schlumberger; A. Seto, SPE, Pengrowth Corp.; and E. Gomez, Schlumberger 5. How a Conceptual Model from Core to Seismic Can Drive Reservoir Description: A Case Study from a Large Carbonate Brown Field A. Carrillat, S. Sharma, K. Bora, G. Iskenova, and T. Grossmann - Schlumberger