Multi-stage Warm-up Seismic Full Waveform Inversion Informed by Electromagnetic Data

Abstract

Abstract Full waveform inversion (FWI) provides high-resolution velocity models by exploiting the complete information embedded in seismic waveforms. However, the advantages of FWI come with cycle skipping when sufficient information is not available to build an accurate initial model. To integrate non-seismic data into seismic FWI, this study establishes a direct conversion method, leveraging the correlation between multi-physical models. The diffusive nature of electromagnetic (EM) data makes it an independent, low-cost information source that complements seismic low-frequency data. Unlike joint inversion approaches, our flexible FWI framework initializes FWI using EM resistivity inversion results. This novel approach extracts low-frequency skeleton structure from electromagnetic data to construct an initial FWI model. By design, it avoids strong assumptions about petrophysical correlation and structural alignment and does not require complex multi-physics objective functions, resulting in a straightforward implementation. Our three-stage implementation begins by inverting EM data to generate a blurred resistivity image. Pixels in this image are then clustered into models, each assumed to have a constant velocity. To prevent cycle skipping, a preliminary FWI is performed to roughly fit seismic data by determining velocities for these models. Finally, a full-bandwidth FWI refines the preliminary velocity model, which already incorporates EM and low-frequency seismic information, aiming for the highest possible resolution. Applied to the Marmousi model, our approach outperformed conventional seismic-only FWI methods in terms of data fitting and computational performance. Using EM to warm-up FWI is theoretically similar to employing an initial velocity model by smoothing the true model, because EM surveys can be regarded as a low-pass filter of the subsurface structure, as EM surveys physically average subsurface structures. Our findings underscore the importance of incorporating non-seismic data in seismic imaging and offer a robust workflow for joint multi-physical data acquisition and analysis.