Generative AI and Diffusion Models in Geophysics: A Survey
Year: 2026
Type: preprint
Abstract: Generative modeling has become a practical prior family in geophysics, with score-based and diffusion models now used for seismic processing, inversion, subsurface monitoring, digital rock physics, well-log synthesis, and benchmark-driven reproducibility. Diffusion is best read as a restoration and posterior-sampling prior whose value depends on forward-model consistency, benchmark design, baseline coverage, and explicit uncertainty diagnostics rather than visual plausibility alone. The taxonomy links generated objects, model families, conditioning interfaces, physical coupling, and validation burden. Current evidence is strongest for seismic restoration and controlled posterior-sampling studies, and weakest for calibrated uncertainty in field-scale decisions. Scope is bounded to exploration and reservoir geophysics; ground-motion synthesis and electromagnetic, magnetotelluric, and potential-field methods are out of scope.
Source: Preprints.org
Author: Rong Lu
Subfield: Geophysics
Field: Earth and Planetary Sciences
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7166649080
Cites:
Cited by:
Related to: