Recognition: unknown
Optimal Transport for Seismic Full Waveform Inversion
read the original abstract
Full waveform inversion is a successful procedure for determining properties of the earth from surface measurements in seismology. This inverse problem is solved by a PDE constrained optimization where unknown coefficients in a computed wavefield are adjusted to minimize the mismatch with the measured data. We propose using the Wasserstein metric, which is related to optimal transport, for measuring this mismatch. Several advantageous properties are proved with regards to convexity of the objective function and robustness with respect to noise. The Wasserstein metric is computed by solving a Monge-Ampere equation. We describe an algorithm for computing its Frechet gradient for use in the optimization. Numerical examples are given.
This paper has not been read by Pith yet.
Forward citations
Cited by 1 Pith paper
-
Mitigating cycle skipping in full waveform inversion using max-pooling-based approximate envelope and shot patching
Max-pooling approximate envelope combined with shot patching reduces cycle skipping in FWI more effectively than Hilbert-transform methods when initial models are poor.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.