AWS for Industries

On-demand seismic processing on AWS using GeoTomo’s technology

The seismic methods are the most common and effective ways for subsurface imaging to delineate and characterize oil and gas reservoirs. Seismic data are acquired in the field by deploying a seismic source (vibrator) that radiates elastic waves into the subsurface. The waves travel through the formations and are reflected back due to the variation in the elastic properties of the geological layers. The reflected, refracted, and diffracted seismic waves and the accompanying various types of noise that contaminate the seismic signal are then recorded by the receivers – geophones on land and hydrophones in marine seismic surveys. By seismic processing, raw data are transformed to a subsurface image using a workflow comprised of several steps that are I/O intensive and computationally intensive. The seismic image obtained from seismic processing along with other subsurface related data such as well logs are used for making multimillion-dollar business decisions in the oil and gas industry.

Seismic data processing comprises three principal stages: signal processing of recorded data, velocity estimation, and seismic imaging. Land seismic data processing includes the additional stage of near-surface modeling for statics corrections. Signal processing is aimed at enhancing reflections and suppressing coherent and random noise in the recorded data, and includes a number of steps for single-channel processing and multichannel processing with residual statics corrections interleaved between the single and multichannel processing steps. Estimation of stacking velocities, rms velocities, and interval velocities are I/O intensive and computationally intensive, especially for 3D seismic data. These velocities are used for stacking, time and depth migrations, respectively. Seismic imaging is performed both in time and in depth by way of prestack time and depth migration algorithms, which, again, are I/O intensive and computationally intensive.

Seismic processing jobs, are executed by specialist seismic processor companies using on-premises high-performance computing (HPC) centers that require extensive IT infrastructure planning to setup with major capital investments upfront. There is a push in the industry to move to higher resolution and denser seismic surveys, where closer survey spacings result in much greater amounts of raw data for processing, and we can expect orders of magnitude increase over the coming years. Cloud technologies can relieve most of the pain points related to HPC as discussed in this article. In a report on the state of computational engineering, 78% of engineers are using cloud-based HPC with over 50% using it consistently for engineering and science workloads.

Problem definition and industry challenges

Each step or job executed in the seismic processing workflow involves gigabyte to terabyte scale of data. The total process to go from the raw data to the seismic image could take from weeks to months to complete depending on the size of the acquisition and the workflow applied. Different vintages of seismic processing products are obtained by applying different algorithms or variations of the seismic processing workflow.

Many major exploration and production (E&P) companies are using on-premises HPC centers for executing seismic processing workflows. Contracts are also awarded by the E&P companies to specialized seismic processing service providers for turn-key seismic data processing services. Executing seismic processing jobs in an HPC environment requires detailed business and IT infrastructure resources planning. The IT infrastructure for I/O and computational requirements – increases exponentially as we collect more seismic data in the field and use more advanced algorithms to get better seismic images from the collected data. HPC evolution in the oil and gas industry shows the exponential increase in compute requirements with advancements in the seismic imaging. This heavy resource overhead inhibits the ability to experiment and innovate by executing processing workflows based on immediate business needs. As a result, many of our customers have expressed interest to have an on-demand seismic processing solution.

The On-demand seismic processing (OdSP) solution

An OdSP is a seismic processing HPC solution in the cloud. The OdSP solution in the AWS Cloud executes seismic processing jobs in a cost-managed, fully scalable, and turn-key delivered HPC environment. A HPC cluster for OdSP can be created in the cloud at any time, scaled as needed, and dissolved when the project is complete. There’s no limit of how many HPC clusters can be created as needed to run multiple seismic processing jobs. Such approach will eliminate a need to wait in queue for a completion of another job.

OdSP on AWS using GeoTomo’s technology reference architecture

Figure: OdSP on AWS using GeoTomo’s technology reference architecture

The OdSP solution on AWS starts with uploading the data from data centers to the AWS cloud into Amazon S3 buckets. A wide variety of cloud data migration tools from AWS can be used for this purpose. AWS Parallel Cluster helps in the automatic orchestration of the cluster resources on AWS. The cluster definition is done using a configuration file that contains the details of the cluster, for example: the type of instances for head node and compute nodes, the number of compute nodes and others. Session Manager from AWS Systems manager can be used for launching the cluster. Alternately, a cloud IDE such as AWS Cloud9 can also be used for the same process. AWS parallel cluster has options to manage HPC jobs using REST API. AWS Parallel Cluster makes a custom Cluster CloudWatch dashboard to monitor the cluster performance. This dashboard can be customized to include metrics of interest such as memory consumption. The seismic processing software is installed in a separate instance referred as the seismic processing software node. This includes GeoTomo software that users use to submit seismic processing jobs on-demand to the cluster and visualize the results obtained from the execution of the jobs. In this case, a g4dn instance is used for the seismic processing software node. Nice DCV visualization client provides virtual desktop streaming capability for remote visualization. Amazon FSx for Lustre high performance storage system provides high performance I/O capabilities for this solution. AWS Parallel Cluster mounts storage automatically to the compute node when the job starts. The storage system is synchronized with Amazon S3 bucket for retaining results of running the seismic processing jobs in Amazon S3 bucket when the rest of the resources are released after the job is completed.

OdSP using GeoTomo’s technology

The seismic processing software in the OdSP solution on AWS uses GeoTomo’s software – TomoPlus and GeoThrust. TomoPlus is a comprehensive near-surface solutions package with multiple algorithms designed to obtain accurate near-surface velocity models and derive long and short-wavelength statics solutions for land, ocean bottom nodes and shallow marine seismic data. GeoThrust is an advanced 2D/3D Seismic Data Processing system that goes from raw field data through to Prestack Time Migration and Prestack Depth Migration. The OdSP solution using GeoTomo’s technology gives an end-to-end seismic processing workflow – from execution to visualization and analysis of results. The solution enables customers to run multiple scenarios simultaneously and narrow uncertainty in the results. The solution also provides flexible configuration options to quickly iterate resource selection and ensure cost optimization.

3D seismic volume visualized using GeoTomo’s software

Figure: 3D seismic volume visualized using GeoTomo’s software

Conclusion

The OdSP solution eliminates the need for elaborate IT planning and prioritization of seismic processing jobs. This brings agility to the business by providing results faster which leads to E&P companies significantly reducing their exploration cycle. Cost reduction is achieved when virtual seismic data centers can be instantiated as needed in the cloud and follow customers demand cycle. Ideal price/performance ratio guarantees low cost for each individual run making the whole processing job cost effective. Contact us today to find out how we can help optimize your downstream business outcomes.

Vishal Das

Vishal Das

Vishal Das is an Applied Scientist in the Machine Learning Solutions Lab (MLSL), Amazon Web Services (AWS). Prior to MLSL, Vishal was a Solutions Architect, Energy, AWS. He received his PhD in Geophysics with a PhD minor in Statistics from Stanford University. He is committed to working with customers in helping them think big and deliver business results. He is an expert in machine learning and its application in solving business problems.

Dmitriy Tishechkin

Dmitriy Tishechkin

Dmitriy Tishechkin is Principal Partner Technical Lead, Energy, Amazon Web Services. Dmitriy has over 20 years of experience of architecting and delivering enterprise solutions to customers, and 15 years spent in Energy industry. For 4 years with AWS Dmitriy has been working with partner community to build, migrate, and launch their Exploration and Production workflows on AWS. Dmitriy is interested in renewable energy and reducing carbon footprint technologies.

James Jackson

James Jackson

James C. Jackson is CEO at GeoTomo. He has over 40 years in the oil and gas industry. James worked for Baker Hughes for 26 years where he held a variety of management and engineering positions. During this time, he gained vast international experience while on assignments in Saudi Arabia, Venezuela, and Canada. He has several United States Patents relating to the geophysical industry and has authored/co-authored many SEG and EAGE technical publications.

Kate Sposato

Kate Sposato

Kate is a Sr Partner Solutions Architect at AWS. She’s passionate about working with customers to optimize workflows to achieve key business objectives. She specializes in upstream energy businesses and workflows and is always looking for new ways to streamline legacy processes.

Neil Peake

Neil Peake

Neil Peake is the Business Development Manager for GeoTomo and has spent 30+ years in the oil and gas industry as a technical and business development specialist in various seismic techniques. Neil has travelled extensively around the world providing integrated and innovative subsurface solutions to clients. He is excited about the increasing application of geophysics to the nascent carbon storage and geothermal industries. Neil enjoys scuba diving, playing bass and guitar, and spending time with his family.

Oz Yilmaz

Oz Yilmaz

Dr. Oz Yilmaz is CTO at GeoTomo. He received his Ph.D. in Geophysics from Stanford University. He previously worked for Schlumberger, Western Geophysical, and Paradigm. Dr. Yilmaz received the Virgil Kauffman Gold Medal Award from SEG in 1991 and the Conrad Schlumberger Award from EAGE in 1992. He served SEG as Vice-President from 1993-1994, and as the 1996 Spring Distinguished Lecturer. He is the author of 3 books; "Seismic Data Analysis", “Land Seismic Case Studies for Near-Surface Modeling and Subsurface Imaging” and “Engineering Seismology with Applications to Geotechnical Engineering” published by SEG.

Dhruv Vashisth

Dhruv Vashisth

Dhruv Vashisth, a principal solutions architect for Global Energy Partners at AWS, brings over 19 years of deep experience in architecting and implementing enterprise solutions, with a 15-year tenure specifically in the energy industry. Dhruv is dedicated to helping AWS energy partners in constructing upstream and decarbonization solutions on AWS. Since joining AWS in 2019, Dhruv has been driving the success of energy partners by leading solution architecture, solution launches, and joint go-to-market strategies on AWS.