Panoma (Council Grove) Geomodel Summary Formation-Member Tops Coverage 11000+ wells, formation level tops Tops Check (in Petrel) Data busts readily apparent (rotated view looking north). Panoma Field Model Boundary ~500 Well Model 500 wells with proper log suite for facies and petrophysical modeling Data Structure Grid 1000’ X 1000’ cells Interval Isochores “Proportionally” Layered Model Structural Cross Section Grid Model Architecture Data Management: Tops data set 24,879 Total wells in initial PETRA project, including regions outside of model 12,097 Wells having at least Council Grove top pick by KGS geologist 11,367 Wells in initial structure model. 10,836 After screening for tops busts. Further screening reduced the well count to10,700. Digital well log data set 1,103 Sufficient Council Grove penetration 536 After removing wells with bad curves, data gaps or other problems 469 Final count after further screening SH LM "Dummy" Total A1 23 41 12 76 B1 19 16 12 47 B2 12 15 12 39 B3 20 15 12 47 B4 17 18 12 47 B5 8 34 12 54 C 28 61 12 101 Layers per Model Cells in model XY = 1000 X 1000 feet 5,200 square mile model Average model 8.6 million Maximum15 million (C cycle) Minimum 5.7 million (B2 cycle) 0% 12.5% 25% Phi Porosity Model 2nd Iteration 0.0001 0.001 1 0.1 0.01 10 100 Perm (md) Permeability Model 2nd Iteration Shrimplin GU-2HI A1 Interval Upscale Porosity Core Lith Upscaled Log Phi X-Plot Log Phi 0.2 30 0.2 29 Shrimplin GU-2HI A1 Interval Upscale Permeability Core Lith Upscaled Log Phi X-Plot Log Phi Upscaled Perm. 0.2 30 0.2 29 E-4 27 Petrophysical Model Raw log data was quality checked and aliased. During the processing a cross- plot porosity curve was generated using the Neutron and Density Porosity curves. No shale correction was made in the first models. Similar to facies modeling, porosity curves were "upscaled" at the wells for modeling porosity in cells between the wells (Sequential Gaussian). Shrimplin GU-2HI Core Lith Upscaled Core Lith Predicted Lith Upscale Lithofacies A1 Interval 3D Lithofacies Model 3D Lithofacies Model A1 Interval Lithofacies Model (1) Upscale lithcode curves wells having facies predicted by NNets. (2) Populate cells at these wells with upscaled facies. (3) Model cells between wells for lithofacies (Sequential Gaussian). A constant average curve was fitted creating a constant distribution (from "lith-code" curves) equal to the average probability of that facies. Predicted Lithofacies at Well (upscaled) A1 Interval NM Silt & Sd NM Shly Silt Mar Shale & Silt Mudstone Wackestone Dolomite Packestone Grnst & PA Baf Slice and Dice Seven cycles modeled For efficiency while maintaining fine heterogeneity, reservoir model was divided into 7 marine-nonmarine cycles (A1-C) for lithofacies, porosity and permeability modeling. The 7 perm models were then joined for simulation. Proportional layering in cycles Layer (N) proportional to interval thickness (h). Nm (marine) = (mean h + sd) and Nnm (nonmarine) = mean h. “Dummy” layers hold places of other cycles in each model (12 per model) Define Framework 1. Create a "skeleton grid” 2. Construct top horizon for Council Grove Group (top A1_SH) 3. Create isochores for subjacent zones (2/cycle) and “hang” on top horizon 4. Proportionally layer the zones 5. QC structural framework by sectioning Biasing Lithofacies Geometry Model facies using different biasing parameters based on geologic understanding. Examples include “non- biased” using a spherical range of 100,000' x 100,000', slightly biased (2:1 ratio to NE-SW), heavily biased to NE- SW, and bias based on mapped regional facies distribution patterns. Facies Distribution Biased B1-LM (Grnst-PAbaf) Heavy Trend Biased (NE-SW) A1-LM Grnst-PA) 2:1 Trend Biased (NE-SW) A1-LM Grnst-PA) Non-biased A1-LM Grnst-PA) Initial Simulations (Single Well) 10^2 10^3 10^4 10^5 10 1975 1983 1991 2015 2007 1999 Mo n t hly P ro d. 0.4 0.8 1.2 1.6 2.0 0 60 120 180 240 300 ALEXANDER D2 - PANOMA BHP/Z Flowing 0.4 0.8 1.2 1.6 2.0 0 60 120 180 240 300 ALEXANDER D2 - PANOMA Gas Cum (BCF) BHP/Z Flowing Pressure Cum. Gas (BCF) 1.0 0.75 0.75 1.0 Alexander D2 1.6 BCF cum. Three Runs 1. Upscaled from well 2. GeoModel, Rate Specified 3. Geomodel, Pressure Specified Parameters ! 640 Acre Section ! Cell Size: 390’ X 415’ ! Layers: Upscale Well – 6 Geomodel -- 41 ! Well Location: Center ! .6’ X 315’ Fracture ! 100 Year Run ! Sw 30 % ic ! BHP 260 psia i Frac Analysis on Panoma Well Newby 2-28R 12/03/1997 Upscaled from Well Model Core Properties ! Tied to Lithofacies ! Corrected for Klinkenberg ! Filter Data> . 1 Perm,>.01 Perm Lithofacies Layering ! Used 6 Layers to capture different facies & zones ! Simulation under performed actual 20 53 3015.5 2962 0.276967 6.722 6 2.5 2.5 2961.5 2959 0.006298 4.956 5 6 21 2958.5 2937.5 0.009621 5.530 4 12.5 15 2936.5 2921.5 0.221282 8.391 3 11.5 14 2920.5 2906.5 0.107800 8.475 2 5 5 2905.5 2900.5 5.864239 8.679 1 Net Gross Bottom Top Perm Porosity Layer 20 53 3015.5 2962 0.276967 6.722 6 2.5 2.5 2961.5 2959 0.006298 4.956 5 6 21 2958.5 2937.5 0.009621 5.530 4 12.5 15 2936.5 2921.5 0.221282 8.391 3 11.5 14 2920.5 2906.5 0.107800 8.475 2 5 5 2905.5 2900.5 5.864239 8.679 1 Net Gross Bottom Top Perm Porosity Layer Results Upscale well, 6 layers Used a single layer to represent unit wide properties (Øh = 3.8). Provided better results. Upscale well, 1 Layer Perm, Layer 1 0.2 7.6 Range (md) Pressure Specified Run Step down pressure curtailment (Back Pressure Curve) Initial Pressure: 243 Psia Recovery 2.5 BCF Rate Specified Run Rate Curtailment Initial Rate: 589 Mcf/D Recovery 2.6 BCF Model (by Rate) Well Model 2.6 BCF 1.4 BCF Cum. Gas vs. Time Simulation Conclusion ! Rate specified decline provided better match than pressure specified ! Petrel model performed better than upscaled well data alone Simulations Using Geomodels Lith A1_LM B1_LM B2_LM B3_LM B4_LM B5_LM GR 3D view of Facies 8 (Grnst/PAbaff) This view of the occurrences of Facies 8 illustrates the discontinuous nature of this particular facies in 3D. Discontinuity, to different degrees, is the norm for all facies. Lithofacies 8 (grnst/pa) A1_LM Permeabiltiy Lithofacies 8 (grnst/pa) A1_LM 3D view of Permeabilty (Facies 8) A view of a portion of the adjacent slide illustrates the permeability distribution with of Facies 8. 3D views of Porosity and Permeability Porosity was extrapolated in three dimensions from the wells having upscaled facies and porosity using Sequential Gaussian Simulation. This yielded a cellular model with facies and porosity at each cell. Permeability was then calculated at each cell by using Byrnes lithofacies specific porosity- permeability transform equations that are based on empirical data and absolute permeability. Acknowledgements: We wish to thank the following industry partners whom have supported this ongoing project financially and with data: Pioneer Natural Resources USA, Inc. Anadarko Petroleum Corporation BP America Production Company ConocoPhillips Company Cimarex Energy Co. E.O.G. Resources Inc. OXY USA, Inc. W. B. Osborn This paper is a snapshot of an ongoing effort with the ultimate goal of creating a robust three-dimensional geomodel that is suitable for accurate reservoir analysis and simulation. The work to date demonstrates: 1. Eight lithofacies have distinct suite of petrophysical properties and use of those properties in calculating volumetric original gas in place (OGIP) reduces error in the calculation. 2. Neural network models accurately predict lithofacies in non- cored wells using log curves and geologic constraining variables after training on wells with cores. 3. The petrophysical model, synthetic capillary pressure curves, petrophysical transform equations, facies prediction and porosity correction appear to be validated by property-based volumetric OGIP that matches material balance OGIP. 4. Initial simulations using cellular model appear to be working well. 5. Vast detailed geomodel is made possible by extremely large set of detailed tops, digital logs, core data and automation (facies prediction and OGIP calculator, for example). FURTHER WORK: Additional effort will be in several broad areas; 1) “ground truthing” lithofacies prediction and extrapolation, 2) increasing coverage, 3) improving the neural network model and Petrel models, and 4) expanding reservoir simulation. 1. Test the Nnet models by comparing core lithofacies (from undescribed, available core) with those lithofacies predicted. 2. Test Petrel's stochastic facies modeling by comparing its results to Nnet predictions at wells that were not used in conditioning the Petrel model. 3. Consider alternative workflow: populate Petrel cells with variables and use Nnet models to predict facies at cells. 4. Increase well coverage (wells with appropriate log curves). 5. Improve NNet models, if possible, by incorporating other statistically based classification techniques (e.g.: fuzzy logic and Markov chains). 6. Explore use of Nnet facies probabilities in Petrel facies geometry biasing. 7. Establish a detailed, field-wide free water level. 8. Analyze GIP by comparing property based volumetric with material balance OGIP and compare both with production history. 9. Simulate at a variety of scales. CHALLENGES: The primary challenge in this project is to develop single model with sufficient detail to be useful at the field, region, area and well scale. Related to the overall challenge are individual hurdles: 1. Manage vast data sets required for detailed characterization. 2. Balance upscaling against model utility. 3. Physical limitations of software and hardware. 4. Upscale simulation exercises from well to field.