New method improves subsurface mapping using neural representations

A framework using continuous neural representations helps reconstruct elastic properties of the subsurface.

Researchers developed a robust implicit neural representation (INR) framework for multiparameter elastic full-waveform inversion (EFWI). This framework uses continuous neural representations to parameterize the subsurface model. It provides an effective model-space constraint during inversion.

Multiparameter elastic full-waveform inversion (EFWI) is a way to reconstruct subsurface elastic properties. This process faces challenges like strong nonlinearity and parameter crosstalk. It also deals with different sensitivities of seismic data to individual elastic parameters.

The study investigated different elastic parameterizations. Combining P-wave velocity (VpV_p), P-wave impedance (IpI_p), and the Vs/VpV_s/V_p ratio provided the most robust performance. This combination improved parameter separation and produced more reliable reconstructions.

The approach was tested on the Marmousi and Overthrust models. It was also applied to the field Volve ocean-bottom-node dataset. Results show the method improves inversion stability and suppresses parameter cross-talk and inversion artifacts.

The results produce geologically coherent multiparameter models. This highlights the importance of considering both model representation and physical parameterization in multiparameter EFWI.

Key facts

  • The VpV_p, IpI_p, and Vs/VpV_s/V_p ratio combination improves parameter separation.
  • The framework was applied to the Marmousi, Overthrust, and Volve ocean-bottom-node datasets.
  • Implicit neural representation (INR) provides an effective model-space constraint.

All sections

Edition