I believe Full Waveform Inversion (FWI) has been oversold by the geophysical technical community. When FWI is labeled “the Ultimate Tool”, the “Final and Ultimate Solution“, the old timers in the industry get skeptical. Art Weglein wrote in an article in The Leading Edge (TLE) in October 2013:
A central purpose of this article is to bring an alternative voice, perspective, and understanding to the latest geophysical stampede, technical bubble, and self-proclaimed seismic cure-all, the so-called “full-waveform inversion” or FWI. If you think this is exaggerated, I refer to the advertisement/announcement of the 2013 SEG Workshop on FWI whose opening line is, “Full-waveform inversion has emerged as the final and ultimate solution to the Earth resolution and imaging objective.”
Besides representing language, attitude, and a viewpoint that have no place anywhere in science, and, in particular, in exploration seismology, the fact is that the method, as put forth, is from a fundamental and basic-principle point of view (aside from, and well before, any practical considerations and track record of added-value are considered) hardly deserving of the label “inversion”, let alone all the other extreme and unjustified claims and attributes, as being the “deliverance” and the last and final word on the subject.
Evgeny Landa and Sven Treitel write in the March 2016 issue of TLE:
Current FWI algorithms fit the computed seismic response of an iteratively updated initial structural model to the observed data. Weglein (2013) calls these inversion methods “indirect,” since they require a forward calculation to fit the observed data. He has proposed an alternate approach, based on the Lippmann-Schwinger scattering series. This series is invertible in terms of an inverse scattering series (ISS), which can be used to solve reflection-seismic inverse problems. Because this method works directly with the data and a reference (or initial) model, Weglein call his approach “direct” to distinguish it from indirect FWI methods.
Weglein has done the exploration seismic community a great service by highlighting some of the pitfalls and problems associated with indirect FWI methods. It is our contention, nevertheless, that all seismic inversion methods, whether direct or indirect, rest on theoretical foundations making it impossible to find unique solutions replicating ground truth.
And later in the same article:
If we knew how to describe these two operators analytically, we might be able to invert for the subsurface structures directly and accurately from the data. This could be done were we able to record noiselessly and with perfect accuracy everywhere on the recording surface, and that we had a full and perfectly accurate understanding of the physics of seismic wave propagation in a heterogeneous earth. Such a goal is of course a pipe dream, and so we content ourselves with the approximate procedures listed in the introduction.
I believe there is a good place for FWI in the arsenal of seismic processing and imaging tools. But it needs to be presented with caution and circumspection, we are a conservative industry.
In 2006 I saw a presentation by Bee Bednar at a GSH (Geophysical Society of Houston) meeting, showing FWI examples on the 2D Marmousi model. At the time, the results were impressive. 10 years later, when I see the same results on the 2D Marmousi, I cringe. It is OK to show the Marmousi results part of a large set of different synthetic and real data models, but when the final and ultimate results of a presentation are only the 2D Marmousi, I think the work is 10 years late. Why not show SEAM model results?
Leave a Reply