← Back to Sessions

Extensions for the Reversibility of First-arrival Travel-times using PINNs

Description: 

The numerical solution of the Eikonal equation can be used to determine first-arrival travel-times in laterally varying media. However, after caustic formation, the information in first arrivals become increasingly incomplete as the wavefronts fold back on themselves, resulting in a loss of reversibility using just first-arrivals. This can be improved by smoothing the medium which moves caustic formation out to greater distances. It can also be addressed by using full ray-tracing in phase-space which is reversible. Nonetheless, first-arrival front-tracking continues to be popular for many seismological applications. Here we investigate the reversibility of first-arrivals using physics informed neural networks (PINNs) which once trained offer advantages in flexibility and speed in comparison to traditional numerical solutions. The occurrence of caustics, if not too densely occurring, can be identified from discontinuities in slope of the first-arrival fronts at the receivers. Once caustics have formed, reverse wavefront tracking can be thought of as finding a minimum-norm, initial wavefront solution. A general solution can then be obtained using prior information of L/a, the ratio of distance to heterogeneity scale, and heterogeneity strength to find an updated estimate of the initial wavefront while still using first-arrival tracking. Several tests are performed using forward and reverse propagation of corrugated wavefronts in laterally varying media. Further examples are then given of forward and reverse wave-front tracking from variable shaped earthquake faults.

Session: Scientific Machine Learning for Forward and Inverse Wave Equation Problems [Poster]

Type: Poster

Date: 4/15/2025

Presentation Time: 08:00 AM (local time)

Presenting Author: Robert

Student Presenter: No

Invited Presentation: 

Poster Number: 138

Authors