New method estimates complex subsurface properties

Stein mixture variational inference helps represent multiple plausible Earth models.

Researchers introduced Stein mixture variational inference (SMVI). This method provides an efficient way to perform Bayesian geophysical inversion.

Bayesian geophysical inversion uses a probabilistic framework to estimate subsurface properties. It uses a posterior probability density function (pdf) to represent uncertainties.

SMVI combines parametric variational inference with Stein variational gradient descent (SVGD). The method approximates the posterior pdf using a mixture of distributions.

The method was tested on two synthetic seismic inverse problems. It successfully recovered non-Gaussian, complex and multimodal posterior solutions.

SMVI was applied to real seismic data through Love wave tomography of the British Isles. It produced results consistent with previous Monte Carlo solutions.

Around the East Irish Sea sedimentary basins, SMVI captured bimodal uncertainty. This reveals alternative subsurface structures compatible with observations.

Key facts

  • SMVI can recover non-Gaussian, complex and multimodal posterior solutions.
  • The method captures bimodal uncertainty near the East Irish Sea sedimentary basins.
  • SMVI results for the British Isles are consistent with previous Monte Carlo solutions.

All sections

Edition