Sim-SEQ: A Model Comparison Study for Geologic Carbon Storage

Transcription

1 Sim-SEQ: A Model Comparison Study for Geologic Carbon Storage S. Mukhopadhyay a,, Z. Hou b, L. Gosink b, D. Bacon b, C. Doughty a, J. Li c, L. Wei c, S. Gasda d, G. Bacci e, R. Govindan e, J.-Q. Shi e, H. Yamamoto f, R. Ramanathan b, JP Nicot g, S.A. Hosseini g J.T. Birkholzer a, A. Bonneville b Presented by Curt Oldenburg a a Lawrence Berkeley National Laboratory, Berkeley, CA, USA b Pacific Northwest National Laboratory, Richland, WA, USA c Shell (China) Innovation and R&D Centre, Beijing, China d Centre for Integrated Petroleum Research, Uni Research, Bergen, Norway e Imperial College London, UK f Taisei Corporation, Yokohama, Japan g Bureau of Economic Geology, University of Texas at Austin, Austin, USA

2 Acknowledgment DOE/NETL for providing financial support Susan Hovorka (BEG, University of Texas Austin) for sharing the site characterization and observation data from the S-3 site All participating institutions and the modeling team members

3 Presentation Outline Model uncertainty and the need for a model comparison study for GCS systems An overview of the Sim-SEQ project Brief Introduction to the Sim-SEQ Study Site Comparison of predictive models and lessons learned Future directions

4 Challenges in Modeling CO 2 Storage Vastly differing time scales Processes are coupled and highly nonlinear Heterogeneities on different scales Sparsity of data in field situations Difficult-to-measure and uncertain parameters Wide range of predictions because of different modeling techniques, coupling methods, approaches for multiphase behavior, interpretations of site data

5 Project Objectives There are two different uncertainties associated with making predictions about flow at GCS sites Uncertainties related to the site or the processes (limited information about the geology; sparse or unreliable monitoring data, etc.) Model uncertainties (uncertainties arising from different interpretation of geology and monitoring data) Sim-SEQ addresses model uncertainties: If the same site characterization and injection scenario is given to multiple modeling teams, will they produce identical prediction? If not, what causes the differences in predictions made by different modeling teams?

6 Model Comparison To increase stakeholders confidence in GCS systems, we need to understand the root causes of model uncertainties and, if possible, quantify these uncertainties This is accomplished by engaging in a model comparison study involving both model-to-data and model-tomodel comparison at one or more selected GCS field sites. Model Team 1 Model Team 2 Model Team 7 Site A Model Team 3 Model Team 6 Model Team 4 Model Team 5

7 Sim-SEQ is not Code Comparison or Benchmarking Benchmarking exercises related to GCS problems have been conducted in the past (Pruess et al., 2004; Class et al., 2009; Nordbotten et al., 2012) Sim-SEQ is Model Comparison Model comparison evaluates modeling studies in a broad and comprehensive sense. Comprises all work flow stages - interpretation of site characterization efforts, parameter choices, model assumptions, decisions about domain sizes and boundary conditions, etc. The DECOVALEX project on model comparison in geologic disposal of nuclear wastes (Tsang et al., 2009) serves as an analog for Sim-SEQ.

8 Sim-SEQ Overview In the works since 2009 Actual kick-off meeting in April 2011 After a modest start, currently 15 participating teams A Sim-SEQ web portal has been launched ( - password protected site, access to Sim-SEQ participants only

9 The S-3 Site The S-3 site is patterned after the Southeast Regional Carbon Sequestration Partnership (SECARB) Phase III Early Test in the southwestern part of the state of Mississippi in the USA. The target formation at the S-3 site is comprised of fluvial sandstones of the Cretaceous lower Tuscaloosa Formation at depths of 3300 m Denbury Onshore LLC has hosted (since 2007) the SECARB Phase II and Phase III tests in a depleted oil and gas reservoir under CO 2 flood. The tests are managed by the Bureau of Economic Geology (BEG) at the University of Texas, Austin. Pictures: Courtesy of JP Nicot (BEG)

10 Modeling Challenges The DAS area comprises fluvial deposits of considerable heterogeneity located in the water leg of an active CO 2 -EOR field with a strong water drive. These features add significant complexity when approximating the natural system, and challenges arise in dealing with boundary conditions. In addition, presence of methane has been confirmed in the brine, which can potentially exsolve and impact pressure buildup history and CO 2 plume extent Acknowledgment: JP Nicot (BEG)

11 Injector BHL = (248267, ) OBS #1 BHL = (248494, ) OBS #2 BHL = (248633, ) X & Y coordinates in feet Depth below ground surface (feet) Formation Top Surface Fault trace at reservoir level Reservoir is nominally 880 feet thick this ssurface + 80 ft.

12 Petrographic Analysis and Heterogenity Facies 1: Standstone (orange) Mean perm: md Variance: 3.39 Porosity: 0.27 Facies 2: Sandstone and conglomerate (hot pink) Mean perm: md Variance: 2.21 Porosity: 0.26 Facies 3: Everything else Mean perm: 9.07 md Variance: 6.63 Porosity: 0.16 Acknowledgment: Diana Bacon and Ramya Ramanathan (PNNL)

13 Injection Data

14 Preliminary Conceptual Models of the S-3 Site Acknowledgment: JP Nicot (BEG) F1 F2 F3 Conceptual models developed by the following teams are included in the comparative analysis presented here (in alphabetical order) 1. Imperial College London, UK (ICL) 2. Lawrence Berkeley National Laboratory, USA (LBNL) 3. Pacific Northwest Laboratory, USA (PNNL) 4. Shell (China) Innovation and R&D Center, China (Shell) 5. Taisei Corporation, Japan (Taisei) 6. Uni Research, Norway (UNR)

15 Comparison of Preliminary Model Results F1 ~70 m ~30 m F2 F3 Acknowledgment: JP Nicot (BEG)

16 Comments Qualitative model comparison illustrates that Model conceptualization plays a significant role in deciding outcome To improve model prediction, need to include more sitespecific parameters/factors (relative permeability, entry pressure, residual saturation, etc.) not considered Model refinement using observation data will likely reduce the range of predictions

17 Ongoing Activities & Future Plans Iterative model refinement utilizing observation data from the S-3 site. Evaluation of the influence of methane on CO 2 flow History-matching and calibration of the predictive model using pressures from the two observation wells and CO 2 arrival times Impact of far-field production and injection, and preferential flow paths on model prediction Quantitative model comparison and uncertainty analysis Reactive transport modeling Extension to other GCS sites Integration with NRAP

18 Thank You Questions?

19 Attributes of the Six Selected Conceptual Models ITEM ICL LBNL PNNL Shell Taisei URN Software Eclipse/E300 TOUGH2/EOS7C STOMP- MoReS TOUGH2- VESA CO2e MP/ECO2N Components Water+CO 2 Water+CO 2 +CH 4 Water+CO 2 + Water+CO 2 + Water+CO 2 + Water+CO 2 Salt Salt Salt Grid Type 3-D, Rectangular, Uniform 3-D, Voronoi tessellation, Irregular 3-D, Rectangular, Irregular, Orientation No Tilted (2 o ) Top boundaryfitted Model extent 2,000 m 2,000 m 4,000 m 5,200 m 3,218 m 3,218 m Number of layers Number of gridblocks 3-D, Rectangular, Irregular Top boundaryfitted 5,000 m 5,000 m 3-D Cylindrical, Voronoi tessellation, Irregular Top boundaryfitted 2-D, Rectangular, Uniform 1,200 m 610 m 610 m (radius) , , , ,901 40,000 No

20 LBNL PNNL Total Gridblocks: 4,968 Total Gridblocks: 44,944

21 Taisei, Japan Gridblocks = Connections = 887,915 (4478 x 50 layers + 1 well) Total gridblocks: 67x68x40 = 182,240 Shell, China

22 Boundary and Initial Conditions ICL LBNL PNNL Shell Taisei URN Fault N/A No flow No flow No flow N/A N/A Top/bottom boundaries Side boundaries Closed Closed Closed Closed Closed Closed Constant pressure Constant pressure Constant pressure Semianalytical aquifer model or closed Constant pressure Constant pressure Initial pressure ~32 MPa ~32 MPa ~32 MPa ~32 MPa ~32 MPa ~32 MPa Initial temperature 125 o C 127 o C 128 o C 128 o C 100 o C 125 o C Initial salt N/A N/A 157gpl 150 gpl 123 gpl N/A Initial CH 4 N/A Water saturated with dissolved N/A N/A N/A N/A Pure CO Pure CO Pure CO Pure CO CH 4 Injected fluid Pure CO 2 92% CO 2 +8% CH

23 Rock Properties ITEM ICL LBNL PNNL Shell Taisei URN Perm/poro Homogeneous (layerwise) Homogeneous (layerwise) Heterogeneou s (facies) Heterogeneou s (facies) Homogeneous (layerwise) Homogeneous Source Core data (F- 2 & F-3) Logs & sidewall cores ( F-1) Core data (F- 2 & F-3) Core data (F- 2 & F-3) Log data (F- 3) Average value of core data (F-2 & F-3) Anisotropy No Yes Yes Yes Yes No Upscaling No Yes Yes Yes Yes No Rel perm van Corey From cores Corey van Brooks-Corey Genuchten Genuchten P C van van Brooks-Corey Core data van No P c Genuchten Genuchten Genuchten S gr and

24 PNNL Realization #1 (perm/poro constant in each facies) Realization #4 (perm/poro random in each facies) Realization #7 (perm/poro Gaussian in each facies) (1) (2) (3) (4) (5) (6) Sequential indicator Truncation Gaussian Anisotropy (2:1) GeoModel1 GeoModel4 Anisotropy Anisotropy (1:2) (1:1) GeoModel2 GeoModel3 GeoModel5 GeoModel6 Permeability scale 0 (cyan)-400 md (red) Shell

25 Uncertainty Analysis Generalized linear model (GLM) analyses are used to test the significance of linear and interaction terms/factors for each response variable Y i n 0 j * pi, j i, ~ N (0, j 1 An Akaike's information criterion (AIC) based backward removal approach is used to identify the most significant parameters. The statistical significance of input parameters is evaluated through null hypothesis tests A t-statistic and a P-value are calculated for each input parameter If the P-value is larger than the significance level of the test (e.g., 0.1), one can accept the hypothesis that the corresponding parameter is not significant. i iid 2 )

26 Statistical Significance of Parameter Factors F-2 Breakthrough Time, day F-3 Breakthrough Time, day Max Pressure 30d, MPa Max Pressure 180d, MPa Max Pressure 365d, MPa RMSE bpw2, MPa RMSE bpw3, MPa RMSE MPa Perm * *** ** * Poro ** * *** **** ** Perm:p oro * ** ** R zx ** * *** * **** ** R zy ** ** * R yx ** R zx :R zy ** * * Scheme ** * *** **** ** **** ** ** Modeling scheme plays the most significant role Response variables are more sensitive to the input parameters at intermediate times Simplified models (homogeneous and isotropic) produce the largest deviances Highly complex models (3-D heterogeneous and anisotropic) models have more simulation errors compared to models with intermediate levels of complexities

27 Comments Results are inconclusive The preliminary statistical analyses are based on a limited number of experiments (only 14) and a large number of response variables (8). Need to include more parameters/factors (relative permeability, entry pressure, residual saturation, etc.) not considered in this study. More refined observation data are needed for more conclusive results

28 Site Characterization and Injection Data 1. Location of the injection formation (e.g., top and bottom coordinates of the Tuscaloosa Formation; formation thickness; fault trace) 2. Construction details and bottomhole locations of the injection (SS31F1) and observation wells (SS31F2 and SS31F3) 3. Open hole logs from the injection and two observation wells 4. Porosity/permeability from core plugs and sidewall cores from the injection and two observation wells 5. Porosity/permeability from other nearby wells (SS ~1900 m from DAS; SS ~2,200 m from DAS; SS ~3,450 m; and SS ~3,950 m from DAS) 6. Relative permeability and mercury injection test data from SS Petrographic analysis of the injection formation 8. Geochemical analysis of the formation brine 9. Carbon dioxide injection rate; bottomhole pressure and temperature data from the injection well

29 Acknowledgment: JP Nicot, BEG

30 Organization Chart PI: Sumit Mukhopadhyay (LBNL) No. Organization/Institution Name of Software/Model Further Information 1. Bureau of Economic Geology, USA CMG-GEM 2. Bureau de Recherches Géologiques et Minières, France TOUGH2/Eclipse /Petrel Geological Storage Consultants, VESA Gasda et al. (2009) USA 4. Imperial College, UK Eclipse 5. Institute of Crustal Dynamics, China CCS_MULTIF Yang et al. (2011a,b), Yang et al. (2012) 6. Lawrence Berkeley National Laboratory, USA TOUGH2-EOS7C Oldenburg et al. (2004) 7. Pacific Northwest National STOMP-CO2E White et al. (2012) Laboratory 8. Research Institute of Innovative Technology for the Earth, Japan TOUGH2- ECO2N Pruess and Spycher (2007) 9. Sandia National Laboratory, USA Not available 10. UFZ, Germany OpenGeoSys Shell, China MoReS Wei (2012) 12. Taisei Corporation, Japan TOUGH2- MP/ECO2N 13. Uni Research, Norway VESA Gasda et al. (2009) 14. University of Stuttgart, Germany DUMUX University of Utah, USA STOMP-CO2E

31 Heterogeneity From Well Log Data non reser voir non reservoir 64 6% 16% Rock facies based on porosity and gamma logs Good reservoir 44 Core data from F2 and F3 Poor sand Good Sand

32 PNNL Transition Probability Based Facies Model Facies 1: Standstone (orange) Mean perm: md Variance: 3.39 Porosity: 0.27 Facies 2: Sandstone and conglomerate (hot pink) Mean perm: md Variance: 2.21 Porosity: 0.26 Facies 3: Everything else Mean perm: 9.07 md Variance: 6.63 Porosity: 0.16 Petrographic Analysis from F2 Observation Wells F2 and F3 Core Data Porosity : %; mean 21.76% Permeability: md; mean: 2.91 md

33 Sources of Model Uncertainty Need to predict the fate of injected CO 2 Uncertainty of subsurface processes and of their spatial/temporal scales Uncertainty of the subsurface geology and of the distribution of parameters (flow, PVT, geochemistry, etc.) Choices made by modelers: software to be used, which processes, coupling of processes, multiple length scales and grid discretization, boundary conditions These choices cause a wide range in model predictions

34 Gantt Chart Year FY11 FY12 FY13 FY14 Quarter Task 1: Data Review and Site Selection Task 2: Model Development Task 3: Predictive Simulations Task 4: Model Refinement Task 5: Technical Team Participation Annual Reports - - -

35 LBNL Layer Properties Layers are equal thickness except for one thinner lowpermeability layer inferred to be a shale baffle Well-log permeabilities are scaled to well-test permeability (multiply by 1.76) Arithmetic average for horizontal permeability Harmonic average for vertical permeability, times anisotropy factor of 0.5 (literature value) No lateral heterogeneity, except well column permeability decreased to represent skin effect (well-test analysis)

36 Bibliography Journal, multiple authors: Mukhopadhyay, S., Birkholzer, J.T., Nicot, J.P., and Hosseini, S.A., 2012, A model comparison initiative for a CO 2 field injection test: An introduction to Sim-SEQ. Environmental Earth Sciences, doi: /s , available at: Curt, I am attaching the presentation for the IEAGHG meeting next week. There are 27 slides (the last one is just a thank you slide for your convenience). You may delete slides and 23-25, which will bring the total slide count to about 18. I kept those slides (9 in total) so that you have all the details while making the presentation. You can just rush through them or put them at the end so that, if anybody has a specific question, you can use them. I have also added a few more slides (after slide 27) with some details, just in case you need those additional details. Let's meet sometime today/tomorrow, after you have time to go through the slides, particularly if you need any other data/info. Thanks for doing the presentation Sumit