Fourier neural operator tsunami surrogates for global hazard, source inversion and inundation analysis
Year: 2026
Type: preprint
Abstract: Preliminary version (v0.1) for a seminar on 14 Oct 2026; not peer reviewed. Later versions will update the manuscript, code and data. Draft manuscript, supporting information and research code for fast tsunami models: Fourier neural operator (FNO) surrogates trained on our own GPU nonlinear wet/dry shallow-water solver. For the Japan Trench the surrogates reproduce the solver coastal maximum water level (median absolute log error 0.07-0.11) about 1,700 times faster, and are tested in probabilistic hazard, slip inversion and inundation of the Sendai plain. Regional models for 20 subduction zones are compared with pooled models and with GeoClaw in six regions, and are served through a web service. Contents: preprint (version 0.1, 13 Oct 2026), supporting information, code (solver, training/evaluation, multi-region experiments, GeoClaw comparison, figure scripts) and README. Trained weights, event ensembles, bathymetry and third-party observations are not included in this version.
Author: Meng, Lingsen
Institution: University of California, Los Angeles
Language: English
Subfield: Statistical and Nonlinear Physics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7221832070
Cites:
Cited by:
Related to: