Recently I gave each of my 5 kids a copy of a book I was delighted to read, The Curmudgeon’s Guide to Getting Ahead, by Charles Murray, fatherly advice to young people who enter the workforce. I identified with the Curmudgeon, hence the title of this column. Murray’s book started as postings on the internal website of the American Enterprise Institute where he works, with tips for entry level staff and interns such as:
- Excise the word “like” from your spoken English.
- Don’t suck up, meaning don’t flatter your supervisors.
- Stop “reaching out” and “sharing.”
- Rid yourself of piercings, tattoos, and weird hair colors.
- Make strong language count.
As a father I thought all of this is good, solid advice.
With the Curmudgeon in mind, I have a few comments to make about our life as Researchers, pushing the leading edge of technology in our industries. Herding a small group of developer cats at Z-Terra, I have to point them to research areas in the pasture where the grass is not so trampled by the large companies research groups with larger budgets and lots of smart people working on fashionable topics. In small companies it helps to be a contrarian, to develop novel algorithms in areas overlooked by large research groups or the academic groups funded by them.
A few years ago, I read a book by Peter Thiel, “Zero to One.” He starts the book with a question he always asks people he interviews for a job: “What important truth do very few people agree with you on?” It is a contrarian question that is related to the title of his book, he argues that while incremental innovations, making existing things better, is going from 1 to n, inventing new technology is going from 0 to 1. That made me think what would be my answer to his question, and one possible result is Smart Migrations vs. Full Waveform Inversion (FWI) and Reverse Time Migration (RTM).
While a large number of researchers in our industry have joined the stampede on FWI and RTM, high end imaging tools requiring more and more computer resources, fewer have followed John Sherwood’s work on improving Beam Migration and Beam Tomography. I believe the combination of Fast Beam Migration (FBM) and Beam Tomography can lead to high resolution velocity models similar to the ones obtained using FWI, but 100 to 1000 times faster while at the same time the algorithm is very stable, unlike FWI which easily falls into local minima. A while back I attended a local Geophysical Society of Houston (GSH) presentation where John Sherwood was showing examples of Beam Tomography results. One of the results was a velocity model after Beam Tomography where the velocity update was different inside a river channel, somewhere between 1000 and 2000 meters deep. I was blown away by that resolution, never before did I see the results of a tomography update that had so much detail. I came back to the office convinced we have to work on this technology.
In my classification Beam Migrations are a subset of Smart Migrations, a class of algorithms that use information in the prestack input data to guide the migration operator, in contrast with Brute Force Migrations (or Dumb Migrations) such as Kirchhoff or RTM that make the assumption that every point in the subsurface is a diffractor, do not use the stack information available and make no a priori assumptions about the migrated image structure. There are some hybrid algorithms that use the dips from the stack to constrain the aperture, that do not fit neatly in my classification, but the helicopter view still stands. One early and successful commercial implementation of a Smart Migration class algorithm is the Fast Beam Migration developed by John Sherwood at Applied Geophysical Services. The speed of FBM is achieved in two steps:
- A factor of 5-10 in speedup is achieved using beam forming, or beam decomposition of the input data, where the size of the input data is reduced by a factor of 5-10.
- A factor of 10 to 100 in speedup is obtained by spreading each input trace or beam over a beam instead of a full aperture-volume.
Beam Tomography works by combining standard tomography with Beam Migration. The industry standard reflection tomography performed in the post-migrated domain has many advantages over standard tomography performed on prestack data. In general, post-migrated events are much easier to pick, the data volume is more manageable, and the whole process is more robust. The procedure converts common image gather residual picks to velocity changes using 3D tomographic back-projection. In tomographic MVA, fans of rays are used to backproject residual velocities to the places where the velocities errors originated. The state of the art tomography in the early-mid 2000s was based on single value updates, from each (x,y,z) point in the image, a single value for the residual velocity (or time delay) was used to update the velocity along all offset and azimuth rays. The state of the art tomography in the mid-late 2000s, was based on event picking in offset gathers, and generating residual velocity updates from each (x,y,z) point in the image, resulting in a vector of velocity residual values function of offset dV(h) used to update the velocity. Since you have 50-100 bins in an offset gather, the number of independent velocity values for each (x,y,z) point is on the order of 50-100. The state of the art today (2020) is to separate the input data by several main azimuths (typically 5-10), migrate them separately, and update the velocity model using several azimuths and all offsets dV(h;a). 6 azimuths multiplied by 60 offsets generates 360 independent values for the velocity update. In Beam Tomography about 10,000 source-receiver pairs contribute to the velocity update for each (x,y,z) point in the subsurface. That will be the state of the art in the industry in 5-10 years from now.
The faster imaging software allows for more automatic iterations of velocity model building (100-500 iterations, instead of the current 7-10), which enable the processing team to enhance the seismic resolution and imaging of complex geologic structures, and allows for deeper data penetration, steeper dip and sub-salt structure imaging. Improved velocity models in combination with wave-equation imaging provide much greater resolution and accuracy than what can be accomplished today with standard imaging technology. This technology is not yet mainstream in the industry, is a fundamental advance, and is a necessary building block in any seismic processing system that uses wave-equation methods for imaging ultra-deep land and water, complex geological structures, which are the focus of modern oil-and-gas exploration.