About the Author

A Novel Computational Method for Real-time Seismic Imaging

By Karthika Swamy Cohen

The process of monitoring for seismic activity and vibrations under the Earth’s subsurface involves massive data collection with thousands of seismic sensors collecting information and transferring them to a centralized system for processing. The entire operation is costly and spans several days or even weeks and months in some cases. 

Oil and gas companies as well as earth sciences communities invest a great deal of research into finding new and faster ways to obtain and process this information. The ability to process this data in real time would greatly reduce costs and risks associated with the exploration and production sector. 

At the SIAM Conference on Computational Science and Engineering being held in Atlanta, GA, this week, as part of the minisymposium on Modeling and Computational Methods in Network Science and Applications, WenZhan Song of the University of Georgia presented an innovative Real-time In-situ Seismic Imaging (RISI) system. RISI is a mesh sensing computing system capable of monitoring subsurface structures and dynamics in seconds, and can thus compute and visualize the 3D subsurface in real time.

Seismic imaging applications with large and dense networks present a great deal of bandwidth and energy constraints that make it challenging to collect raw seismic data in real-time. This calls for in situ computing.

Song remarked that just as in other areas and disciplines, the Earth sciences has been inundated with huge amounts of data, which presents a data resolution problem. The Internet of Things is transforming the way we live our lives but the added convenience and efficiency comes at an extra cost. “In situ computing is necessary for the Internet of Things,” Song said.

Researches used seismic imaging to investigate Mount St. Helens, a volcano in Washington that erupted in 1980. Image courtesy of Wikipedia.
Song began by describing the process used for seismic imaging under the Mount St. Helens volcano ten years ago. The approach, which used real-time sensors deployed by helicopters, offered a lot of lessons to Song.

Seismic imaging of the volcano presented a challenged network in a hardy environment, which caused sensors to become unstable, leading to huge gaps in data under harsh winter conditions. More data was lost during the afternoons due to high temperatures.

“Traditional internet design assumes that the network is stable and has a very high bandwidth. For many applications today you have a scenario where you cannot convey all the data to one place to process. You have to communicate them to a network,” Song said. “So I thought: why do we need to connect all data? Can we just process them in networks?”

And hence the idea was born.

Instead of data collection followed by post processing at a central place, the mesh network performs distributed data processing and inversion computing under bandwidth and resource constraints, generating an evolving 3D subsurface image as events occur. 

Seismic imaging uses multiple methods. One well-known method is seismic tomography, which is similar to CT scans or MRI technology.

The existence of electrical waves under magma makes activity amenable to measurement by such techniques. However, when using such a method, the wave propagation speed under different media tends to be different with waves arriving at sensors at different times.

This can be accounted for mathematically by picking arrival times. The problem is formulated in terms of the size of the wave penetrating the ground from the source and reaching the sensors; this can be illustrated by one linear equation per sensor and source. Accounting for thousands of such sensors and sources makes this an inversion problem. The basics of inversion computing are used to solve a least-squares problem.

Seismic imaging algorithms in general use cannot be directly utilized in sensor networks, which are centralized and hence need global information at the start of the operation. Thus real-time seismic imaging requires a distributed or decentralized strategy, which can process and compute seismic data and images in-situ in real-time despite network constraints such as bandwidth, energy, and computing power.

Song’s group has developed multiple in-situ distributed seismic imaging algorithms including tomography and migration, and validated them using synthetic and field seismic datasets. The team designed synchronized and asynchronized algorithms, which allow every node to compute and transmit data without necessarily waiting for its neighbors. Researchers also made comparisons of centralized computing with distributed computing results.

RISI was seen to be fast, inexpensive, and more faithful to its goal relative to existing subsurface cameras. 

  Karthika Swamy Cohen is the managing editor of SIAM News