Estimating geological stacking patterns using seismic waveform detection: A gateway to using large language models for seismic interpretation of geological facies

David J. Emery, Marcelo Guarido, Daniel O. Trad

Interpreting seismic waveforms to determine geological facies stacking patterns has been an accepted principle since the 1970s. The advent of seismic computer-based 3D interpretation has focused more on peaks and troughs, which has started to minimize this style of interpretation. Fortunately, machine learning permits the automation of pattern recognition, and this report concentrates on the autodetection and classification of two closely spaced reflection coefficients (RC).

Large language models have successfully translated between variations in grammar and context. The hope is to use similar techniques to translate geophysics into geology. The results from autodetecting seismic waveforms and tying them back into reflection coefficients are intended to provide the lettering for a geological profile. The lateral association of the waveforms should provide the wording and words along the profile describing the depositional sentence.

This style of analysis infers a few fundamental principles: first, the reflectivity is sparse (limited), thin-bed tuning reflects the amplitude response, and the seismic reflector dip and continuity can be used to determine geological profiles.