Back to search

University of Leeds

Traveltime eikonal tomography with physics informed neural networks

Abstract

dc:description.abstract

This thesis develops and validates three innovative methodologies for surface wave tomography using physics-informed neural networks (PINNs): pinnET (PINN-based eikonal tomography), pinnEAET (PINN-based elliptical-anisotropic eikonal tomography), and pinnTET (PINN-based teleseismic eikonal tomography). These methods progressively address challenges in seismic tomography from isotropic to anisotropic tomography, and ambient noise to teleseismic earthquake applications. By integrating the eikonal equation as a physical constraint while leveraging neural networks' approximation capabilities, PINN-based surface wave tomography demonstrates several key advantages, including significant memory efficiency, physics-guided interpolation for sparse data regions, simultaneous multi-frequency processing, and flexible evaluation at arbitrary locations. However, challenges remain in computational efficiency and automated parameter optimization. When applied to the seismic dense array in northeastern Tibetan Plateau using both ambient noise (10-40 s periods) and teleseismic data (20-80 s periods), these methods reveal significant lateral heterogeneity in velocity structure and azimuthal anisotropy. Notably, they achieve comparable resolution quality with only approximately 20 % or even less of traditionally required data. The results indicate prominent low-velocity zones beneath the western Qilian Orogen, western Qinling Orogen, and Songpan-Ganzi Terrane, contrasting with high-velocity zones in the Ordos Block and central Qinling Orogen. These findings provide new insights into the region's complex crustal and upper mantle structure while demonstrating the practical utility of PINN-based approaches in seismic tomography. Comprehensive uncertainty analysis, checkerboard resolution tests and cross-validation with traditional approaches confirm the methods' reliability and resolution capabilities. This work demonstrates that PINN-based approaches provide valuable alternatives for seismic tomography, particularly in regions with limited data coverage, while establishing a foundation for future developments in physics-constrained seismic tomography methods.

Degree

thesis:*
Name dc:type.qualificationname
Ph.D
Level dc:type.qualificationlevel
doctoral
Grantor dc:publisher.institution
University of Leeds
Year dc:date.issued
2024

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Chen, Yunpeng
Advisors dc:contributor.advisor
  • De Ridder, Sjoerd
  • Rost, Sebastian
ORCID person:identifier.orcid

Rights

Language dc:language.dcterms_iso639_2
eng

Chain of custody

source
Harvested from
White Rose University Consortium
Base URL
etheses.whiterose.ac.uk/cgi/oai2
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
related terms
citation

Chen, Yunpeng. Traveltime eikonal tomography with physics informed neural networks. doctoral thesis, University of Leeds, 2024.