The Birth of Digital Seismology and Origins of ERL
Thursday, January 31, 2019
Led by MIT alumnus Enders Robinson, MIT’s former Geophysical Analysis Group transformed the field of geophysical recording and data processing.
In the late 1940s, a conversation between an MIT geologist and a mathematician led to an innovative collaboration that would revolutionize geophysics and the exploration energy industry: MIT’s Geophysical Analysis Group (GAG), the precursor to the Earth Resources Laboratory (ERL) in MIT’s Department of Earth, Atmospheric and Planetary Sciences (EAPS).
Before the earliest computers, researchers, students, and geophysicists scrutinized seismic data—laboriously interpreting peaks and valleys on a seismic trace—to map subsurface features and find likely reservoirs of hydrocarbons. To obtain this time-series data, exploration geophysicists would trigger ground motions with explosives sending energy through the earth. The “echoes” were recorded as a waveform on a strip of photographic paper: a seismograph. As the waves traveled through layers with different porosities, they would bend, distort, reflect, and reverberate, providing information about the local geology and potential fossil fuel resources.
However, seismic data are notoriously “noisy.” Seismographs pick up irrelevant motions in the earth or capture the same wave multiple times as it reflects off of underground features. In the addition model that industry used at the time, all of these wave sources combined into a single waveform trace in analog—a process called “convolution” —which scientists had to visually tease apart.
The new consortium at MIT, GAG, was about to change all of that, making computation easier, quicker, and more accurate.
It was during a carpool that MIT Professor George Wadsworth, a mathematician applying time-series methods to weather prediction, was discussing the use of mathematics in geology with Professors Robert R. Shrock and Patrick M. Hurley, both of whom worked in MIT’s Department of Geology and Geophysics. Wadsworth needled Hurley because weather and seismic traces behaved similarly and wondered why time-series analysis had not yet been applied to seismograms. Wadsworth, now interested in geophysics, set a new graduate student, Enders Robinson ’50, S.M. ’52, PhD ’54 to the task of determining if he could use time-series analysis to find wave reflections in the record that would help estimate the properties of the Earth’s subsurface.
Time-series analysis fascinated Robinson. After finishing his bachelor’s in mathematics at MIT, Robinson returned to MIT in the fall of 1950 and, working under Professors of Mathematics Wadsworth and Norbert Weiner, began applying time-series analysis to weather prediction and seismic exploration, using traces provided by petroleum companies. Simultaneously, Robinson pursued a master’s degree in economics with Paul A. Samuelson and Robert Solow, who also worked with time-series—a move that would prove useful with his geology problem.
While trying to find underlying innovations in economic data, Robinson learned that technological advances could not be predicted, so when he crafted mathematical equations to reflect this, he found that there should be a measurable prediction error in the data when one occurs. Robinson proposed applying this to geophysics—treating digitized seismic traces as economic series and carrying out prediction-error filtering, now called deconvolution. The method worked. Excited by the initial results and the technique’s potential, Hurley drummed up interest from the oil and gas industry while Robinson learned to code on the Whirlwind, MIT’s first digital computer.
In February of 1952, GAG was born in the Department of Geology and Geophysics, and a year later, it became a consortium—with oil and service companies, MIT researchers, and graduate students. Raytheon was contracted to help with calculations using the FERUT computer, while GAG used the Whirlwind. Despite computing setbacks, technology continued to improve and over several years, GAG consistently showed the consortium’s advisory committee the promise of deconvolution using digital computing. By 1953, it became apparent that industry liked the deconvolution method, but not digital processing. They insisted on investigating the properties of noise and analog filters to boost the signal to noise ratio—the cost and inconsistency of digital processing deterred them.
When Robinson first started leading GAG in 1952, his objectives were to make deconvolution operable on a production basis with the Whirlwind, demonstrate that deconvolution worked on assorted seismic records, and provide a geophysical model that justified deconvolution. When he submitted his doctoral thesis “Predictive decomposition of time series with applications to seismic exploration” in the summer of 1954, GAG had achieved this. By introducing the convolutional model, GAG showed that the signal and noise are related and that the seismic trace is the sum of wavelets arriving with random strengths and arrival times. In Robinson’s words, this turned the seismic world upside down.
For the remaining four years, GAG continued to perform significant research: fitting the model to the data and differentiating between different types of noise, but interest and guidance from industry petered out. By June 1957, GAG shut down and its members scattered into industry.
In the early ’60s, former GAG graduate student Sven Treitel ’53, SM ‘55 PhD ’58, then working at Amoco, revived GAG’s work and, with Robinson, began adapting it for the needs of the fossil fuel industry. Together, they developed Fortran software, as well as writing and republishing papers in layman’s language to function as a teaching tool. By the mid-1960s digital memory improved significantly and digital processing overtook geophysics, making it the first scientific field to do so. Former GAG members, now leaders at oil and service companies, were on board, and the early ’70s saw the “Golden Era” of industry-sponsored university consortia, including MIT’s ERL, which continues to this day in a similar form, to tackle geophysical challenges leveraging the latest in mathematics, machine learning and Earth sciences.
Story Images: Punch tape courtesy of Dr. Sven Treitel; oilfield seismogram courtesy of Bill Gafford, Geophysical Society of Houston Geoscience Center.
The Earth Resources Laboratory (ERL) is MIT’s home for geophysical research driven by technological questions. The laboratory is comprised of a dozen faculty members and their groups, active in areas ranging from seismology to geomechanics, rock physics, flows in porous media, and methods of inversion, inference, and uncertainty quantification.
WATCH THIS: Sven Treitel ’53, SM ’55, PhD ’58 discusses the history of digital seismology at MIT