The seismic physical modelling laboratory as a tool for design and appraisal of FWI methods

Sergio Romahn and Kris Innanen

ABSTRACT

We applied full waveform inversion of PP seismic data recorded through the CREWES seismic physical modelling laboratory facility. Although physical modelling introduces certain challenges that must be addressed, it represents a potentially unique way of validating and appraising complex methods involving real measurements of seismic validating and appraising complex methods involving real measurements of seismic waveforms. One key advantage is that we know the subsurface model that we want to solve; therefore, we can monitor model errors almost exactly. Another advantage is that we can control and vary many acquisition parameters. In several respects, we deal with physical modelling data in a similar way we do with real seismic surface data. For example, the wavelet has to be estimated and we must take noise, attenuation, amplitude variations from shot to shot, etc., into account. However, physical modelling data have particularities that need to be addressed, such as source-receiver directivity and changing waveform with offset. We present an early stage, robust workflow for preparation of raw physical modelling data to use as input to FWI; ultimately this will make the CREWES physical modelling lab an almost unique tool for validating and appraising FWI. We show in detail all processing required to make physical modelling data suitable for being inverted. As an important example of practical FWI algorithm, we used iterative modelling, migration and inversion (IMMI) which aims to incorporate standard processing techniques into the full waveform inversion process. This involves several approximations and internal calibration steps. The gradient is approximated for applying on pre-critical reflections using the phase shift plus interpolation (PSPI) migration. We derived non-stationary matched filters from well information to calibrate the gradient. We also iteratively applied Gaussian smoothers to frequency-band fixed migrated data residuals as an alternative form of the frequency multi-scale FWI. The overall recovery of P-wave velocity variations is quite dramatic, though comparison of blind and calibration well information shows that FWI must work hard in order to generate long-wavelength updates that differ significantly from the geology near the calibration wells.

View full article as PDF (3.20 Mb)