Testing Advanced Computational Tools for 3D Seismic Analysis Using the SEG/EAEG Model Dataset

Proposal for final ("close-out") year of Imaging-SEG project, 1999.

Participants: Advanced Data Solutions, Anadarko, ARCO, BHP Petroleum, BP Amoco, Burlington Resources, Chevron, Conoco, Edison Chouest Offshore, Exxon, GECO-Prakla, Marathon, Mitchell Energy, Mobil, Paradigm Geophysical, PGS-Tensor, Phillips, Shell, Texaco, Union Pacific Resources, Unocal, Western Geophysical, Lawrence Livermore, Los Alamos, Oak Ridge, Rice Univ./HARC, Stanford Univ./SEP, Univ. Calif./Davis, Univ. of Houston, SEG.

Statement of Problem

3-D seismic surveys are used to find, develop, and produce U.S. hydrocarbon resources from geological settings, such as beneath salt, that are increasingly complex and difficult to image seismically. 3-D surveys have provided tremendous benefits, but have also produced major challenges for seismic processing and imaging. The high accuracy and resolution needed from the resulting images and the huge volume of input data are major factors driving advances in imaging. In addition, time-lapse (4-D) and multi-component (3- and 4-C) surveys have shown great potential but further add to the complexities of imaging and interpretation and to the volume of data to be processed. This project focuses on developing and testing new techniques to make 3-D seismic imaging and modeling better and faster. It is a collaborative industry/national laboratory/university project and consists of four related but separate tasks, which are unified by their use of the SEG/EAGE salt model and the acoustic numerical model data set calculated from it. The salt model provides a complex geological setting and difficult imaging targets. Since the correct seismic image is known, deficiencies in images can be readily identified. Each task of the project aims to solve a key problem in seismic imaging, modeling, and data analysis, and each has its own set of partners and industry liaison. The industry liaison provides technical leadership and guidance for the task, and facilitates the exchange of ideas and results. This proposal seeks funding for a broadly aimed project, with many participants. The overall goals of the project are to demonstrate the usefulness of the SEG/EAGE numerical model data, and help make the data set more accessible to researchers in industry. More specific goals are listed below for each of the projectís four tasks.

Tasks and Contributions:

This is to be the "close-out" year for this project, and only two of the four project tasks are requesting funding. All four tasks are described here to give an overview of the entire project and its goals. Only tasks 3 and 4 are requesting funding to allow existing efforts to be brought to logical completion. Tasks 1 and 2 are complete within the context of this project. The accomplishments from these two tasks and interest from industry participants stimulated a new project initiative and a new project proposal.

Task 1: Advanced Cost-Effective Imaging Techniques. The goal of this task was to develop and demonstrate new wave-equation-based 3-D imaging methods that are substantially faster than available methods, and which produce high-quality images. The first of these new methods, termed "common-azimuth" imaging transforms data to similar offsets and azimuths then partially stacks traces ahead of migration. This speeds up the imaging in proportion to the number of traces in the partial stack. It has worked well with the SEG/EAGE salt model data, and with a field data set from the North Sea. This work has been successfully completed and no additional funding is requested for this task.

Task 2: Elastic Numerical Modeling. Increased recent interest in multi-component (3-C and 4-C) data collection underscored the importance of understanding elastic wave propagation. Shear waves can have a substantial effect on images, since the imaging assumes acoustic wave propagation. In addition to mis-imaging converted wave arrivals, trace amplitudes can be wrong because of wave conversions. Understanding elastic wave propagation in complicated structures requires elastic wave modeling with realistic structural models. Computational requirements for these calculations may be 100 to 200 times greater than for an equivalent acoustic problem. The goals of this work were to define realistic elastic model parameters and to compute 3-D elastic numerical data from the SEG/EAGE salt model using a numerical modeling code developed by this project. These goals were completed this past year and no additional funding is requested for this task.

Task 3: Physical Modeling. Two data sets were collected from a physical model of the SEG/EAGE salt structure. One simulated a conventional marine survey, the other a vertical receiver cable survey. The two data sets were collected so that 3-D imaging results from the two different acquisition methods could be compared. The vertical receiver cable data has been imaged by researchers at Texaco and the University of Houston. The marine survey data set has been partly imaged by researchers at Marathon and Los Alamos. A portion of this data set has been imaged, but this has taken much longer than was anticipated. As a result, more work is needed to image enough of the marine survey data set to allow meaningful comparisons between it and the vertical receiver array data. Imaging at Los Alamos is being done with a 3-D split-step algorithm, which was developed by another project. Imaging at Marathon will be done with a 3-D Kirchhoff method. These imaging results will allow comparison of results from the two different imaging techniques, as well as from the two different acquisition geometries. Researchers at Los Alamos, the University of Houston/AGL and Marathon will be doing this work. The industry liaison is Robert Wiley, Marathon.

Task 4: Unconventional Seismic Processing, Analysis and Inversion. This task consists of two sub-tasks. One is developing methods for faster and more reliable event picking and tracking to improve 3-D velocity estimation. The other is developing methods to solve large inverse problems and to estimate rock properties from seismic and well log data.

Building a reliable 3-D velocity model is an essential part of 3-D imaging, yet manual picking and tracking of seismic events in a 3-D volume is time consuming and error-prone, even for an experienced interpreter. A highly successful neural network technique for 2-D event picking and tracking is being extended to work with 3-D data. Event picking and tracking with a prestack 2-D data set is a 3-D problem, with the addition of the source-receiver offset. Working with a prestack 3-D seismic data set is a 5-D or 6-D problem, since it adds two source-receiver offsets, and, depending on the design of the acquisition might add source-receiver azimuth. The automated methods that are being developed exploit the concepts of deformable contours and "snakes" from the field of computer vision research to carry the analysis through the 5-D or 6-D data space that is occupied by 3-D seismic data. Researchers at Lawrence Livermore will lead the work on this sub-task in collaboration with U.C. Davis.

New methods for solving large inverse problems and for seismic data synthesis and data fusion have been developed and applied to problems of determining residual statics and estimating rock properties from seismic data. These are large problems and the estimation of uncertainty in the solutions can be almost as important as obtaining the solution itself. This work has developed a unique method for estimating the uncertainty in the solutions to these large problems. The logical completion of this work is to combine two portions that were developed separately into a single technique for more reliably predicting rock properties and pseudo-logs (gamma ray, resistivity, and porosity). The single technique will predict the rock properties and pseudo-logs, and provide meaningful uncertainty bounds for the predictions. Researchers from Oak Ridge will lead the work on this sub-task.


Task 1: Advanced Cost-Effective Imaging Techniques. No additional funding is requested for FY 00.

Task 2: Elastic Numerical Modeling. No additional funding is requested for FY00.

Task 3: Physical Modeling. Additional portions of the data set from the marine survey will be imaged with wave-equation based 3-D prestack depth migration. We will not be able to image the full data set, but plan to image a significant portion of it. Additional imaging of this data set may be done at Marathon with a 3-D Kirchhoff method. If the Kirchhoff imaging is done, we will be able to compare results from the two imaging methods as well as from the two data sets.

Task 4: Unconventional Seismic Processing, Analysis and Inversion. The event picking and tracking sub-task will make final technical improvements to the 2D and 3D event picking algorithms. Also, the software and related expertise that was developed at Lawrence Livermore will be provided to Shell and other participants who are interested. The second sub-task will combine methods of neural network rock properties estimation with nonlinear uncertainty analysis to predict pseudo-logs (gamma ray, resistivity, and porosity). Research results from both sub-tasks will be documented in manuscripts that will be submitted to peer-reviewed journals.

Funding Table:

  FY99 (Last Year) FY00 (This Year) FY01 (Next year)
Total DOE $ 730K $ 425K $ 0K
Industry Contributions $ 350K $ 250K $ 0K
Contracts to universities (included in DOE) $ 165K $ 60K $ 0K

Contact: Leigh House, phone: 505-667-1912, FAX: 505-667-8487, email: house@lanl.gov


The overall project goals, to demonstrate the usefulness of the SEG/EAGE numerical model data, and help make the data set more accessible to researchers in industry, have been satisfied. Specific accomplishments are listed below by year for the past three years of the project. Accomplishments in 1999 are broken down by project task, those for 1997 and 1998 are summarized by year.

Task 1: Advanced Cost-Effective Imaging Techniques. The common azimuth technique for 3-D prestack depth imaging was developed and tested with the SEG/EAGE data set. In some portions of the data set common azimuth imaging results were superior to results from Kirchhoff migration. The common azimuth imaging technique was refined by adding multiple reference velocities. The resulting images are uniformly superior to those from Kirchhoff. A two-pass 3-D prestack depth migration method was used to image a portion of the SEG/EAGE data set. The salt body was generally well imaged, both top and bottom. Reflectors below the salt were poorly imaged, however. This was expected, since the velocity changes in the structure violate the basic assumptions made by the two-pass method. Nevertheless, these results confirm that the 2-pass method can be a fast way to get an initial velocity model and to image down to the base of salt.

Task 2: Elastic Numerical Modeling. Elastic numerical model data for a 12 shot "mini-survey" of the SEG/EAGE salt model were computed. Shots were recorded by simulated streamer surveys and ocean bottom cables. Calculations were done with the 3-D finite-difference wave propagation code, E3D, and took about 3,400 cpu-hours on an SGI Origin2000 system. The simulation used elastic parameters derived from the acoustic velocity structure. The original 20 m model grid interval of the SEG/EAGE salt model was tri-linear interpolated to 12 m grid spacing to allow use of a source wavelet with 8 Hz central frequency. Results will be presented at a Research Workshop after the 1999 SEG meeting and shared with participants at a later project meeting.

Task 3: Physical Modeling. Imaging of the marine survey data set was started, and is partially finished. The depths of layer interfaces in the velocity model had to be adjusted to get reflector depths to match interfaces. Two-pass migration of selected lines provided reference images. A swath surrounding line 351 was chosen for 3-D depth imaging by wave-equation based methods. Although it will take considerable computing time, the wave-equation based imaging methods should yield an image that is superior to what can be obtained from Kirchhoff methods. Marathon plans to carry out Kirchhoff 3-D depth migration of the full marine survey data set, which will allow comparison of the images from the two methods. Initial results will be presented at the 1999 SEG meeting.

Task 4: Unconventional Seismic Processing, Analysis and Inversion. The 3-D implementation of a neural network method for event picking and tracking was improved by adding validation of picks through use of deformable contours and "snakes" from the field of computer vision research. Results of this work stimulated further development of automated event picking and tracking methods at Shell. A unique method for systematic nonlinear incorporation of uncertainties in reservoir parameter estimation was developed. This new technique provides robust estimates of the uncertainty in the derived reservoir parameters. Conventional methods of estimating errors assume a linear relationship between observed and derived values. They are inappropriate for this application because the reservoir parameters are calculated from the seismic and well log data using nonlinear relationships.

Accomplishments in 1998:

The common-azimuth imaging technique was developed and successfully tested with data from the SEG/EAGE model. Elastic model data were calculated from a single shot positioned within the SEG/EAGE salt model. The E3D elastic modeling code was provided to Phillips for testing. Portions of the vertical cable survey data from the physical model were migrated. Shallow portions of the model are well imaged, but the image quality beneath the salt body is disappointing. A preliminary 3-D implementation of the automated event picking and tracking was developed and provided to Shell. A new global optimization method was developed and applied to estimation of residual statics. The method received an R&D 100 award.

Accomplishments in 1997:

A method for shifting the effective source-receiver azimuth and source-receiver offset of traces was developed and tested. The versatile elastic modeling code, E3D was written with the combined resources of several projects. Collection of a second data set from the physical model was completed; this simulated a survey using vertical receiver cables. Automated event picking and tracking methods were implemented and tested in 2-D. Shell estimated an equally successful 3-D implementation would reduce the time needed for 3-D velocity analysis from 12 weeks to 1, and reduce the cost from about $75,000 to $6,000. A new method for solving large optimization problems was developed and partly tested by applying it to estimating residual statics.

Results of project work have been reported in 54 publications. Eight are peer-reviewed papers, five are trade-journal publications, and 37 are talks at professional meetings. The project is supporting research by several graduate students. One completed his Ph.D. thesis in 1999. Results from this project were essential components of four workshops at professional society meetings (SEG and EAGE).

Copyright 2000 U.C. - Disclaimer - Jan 14, 2000

Comments, suggestions to house@lanl.gov