Synergizing Seismo-geodetic Coupling and Slip Models with Optimal Transport and Machine Learning to Determine if Megathrust Earthquake Ruptures are Slip-deficit Controlled
Description:
Geodetic coupling models are a widely recognized tool for evaluating earthquake hazards. Conventionally, megathrusts marked as highly coupled in these distributions are expected to slip in subsequent large earthquakes, while creeping areas are believed to act as barriers to rupture propagation. However, recent studies indicate that creeping faults may host earthquakes, while locked areas could impede ruptures, thereby challenging our current assessments of earthquake hazards. Unfortunately, our seismo-geodetic observational record of subduction interfaces -typically spanning less than twenty years at a single megathrust, a mere fraction of the timescales over which seismic cycles operate - does not allow us to determine whether seismic slip and coupling are linked.
Here we develop an innovative framework that merges optimal transport principles with machine learning, employing the Wasserstein Generative Adversarial Networks algorithm (WGAN). We use this framework to train neural networks on decades-long geodetic coupling distributions, and compute geodetically and physically consistent geometric mappings between megathrusts into a single unifying coupling model. These coupling mappings are then utilized to project coseismic slip across megathrusts, integrating finite slip models from dozens of earthquakes from Chile, Japan, the Himalayas, and other subduction zones into a cohesive space. This approach extends our seismo-geodetic observational record several-fold, enabling us to assess statistical correlations between the unified coupling model and projected seismic slip, and to investigate whether earthquake slip is driven by interseismic slip deficits. This is particularly crucial for regions such as the Cascadia subduction zone, where historical earthquakes are rare, and seismic hazard is assessed based on coupling models.
Session: Predictability of Seismic and Aseismic Slip: From Basic Science to Operational Forecasts [Poster]
Type: Poster
Date: 4/16/2025
Presentation Time: 08:00 AM (local time)
Presenting Author: Bar
Student Presenter: No
Invited Presentation:
Poster Number: 44
Authors
Bar Oryan Presenting Author Corresponding Author bar.oryan@columbia.edu University of California, San Diego |
Alice-Agnes Gabriel algabriel@ucsd.edu University of California, San Diego |
|
|
|
|
|
|
|
Synergizing Seismo-geodetic Coupling and Slip Models with Optimal Transport and Machine Learning to Determine if Megathrust Earthquake Ruptures are Slip-deficit Controlled
Session
Predictability of Seismic and Aseismic Slip: From Basic Science to Operational Forecasts