Physics-Informed Deep Learning for Understanding Earthquake Slip Complexity

Brittany A. Erickson, & Cody Rucker

Published September 10, 2023, SCEC Contribution #13173, 2023 SCEC Annual Meeting Poster #138

Faults are home to a vast spectrum of event types, from slow aseismic creep, to slow-slip to fast earthquakes followed by postseismic afterslip. Understanding the physical mechanisms for the diversity of slip styles is crucial for mitigating the associated hazards and yet persistently difficult. Direct observations of earthquake nucleation and propagation are few and yet the next decade will likely see an unprecedented increase in indirect, surface observations that could be integrated into modeling efforts. Traditional numerical approaches for solving the partial differential equations governing earthquake processes have seen incredible growth in the past century in terms of convergence theory and high-performance computing. However, their integration with noisy or sparse data is currently limited, and their use in solving inverse problems can be prohibitively expensive. Machine learning (ML), on the other hand, excels in the presence of large data and is an actively growing field in seismology, with applications ranging from earthquake early warning to ground‐motion prediction. Despite this increase in activity, not all ML methods incorporate rigorous physics, and purely data-driven models can predict physically unrealistic outcomes due to observational bias or extrapolation. Our work focuses on a new Deep Learning technique that has recently emerged called the Physics-Informed Neural Network (PINN), which seamlessly integrates sparse and/or noisy data while ensuring that model outcomes satisfy rigorous physical constraints. PINNs do not outperform traditional numerical methods for forward problems (except in high-dimensional settings), but they offer advantages over traditional numerical methods in that both forward and inverse problems can be solved in the same computational framework, enabling both data-driven solutions and data-driven discoveries. We are developing a computational PINN framework that integrates observational data in order to better understand earthquake fault processes. Our immediate goal is to answer how on-fault frictional and off-fault rheologic heterogeneity vary in space and time and how these asperities influence slip behavior, but first target important model verification questions such as: What are the benefits and limitations of PINNs to the study of earthquake fault processes? What tests are needed to ensure confidence in PINN model outcomes? How should we quantify sensitivity of model outputs to model inputs?

Key Words
machine learning, earthquake physics

Citation
Erickson, B. A., & Rucker, C. (2023, 09). Physics-Informed Deep Learning for Understanding Earthquake Slip Complexity. Poster Presentation at 2023 SCEC Annual Meeting.

Related Projects & Working Groups
Fault and Rupture Mechanics (FARM)