Physics-informed deep learning approach for modeling crustal deformation.

Area of Science:

• Geophysics
• Computational Seismology
• Deep Learning Applications

Background:

• Earth's crust and upper mantle deformation are crucial for understanding earthquake processes and seismic hazards.
• Dislocation models traditionally represent earthquake faults as defects in a continuum medium for crustal deformation analysis.

Purpose of the Study:

• To propose a physics-informed deep learning (PIDL) approach for modeling crustal deformation caused by earthquakes.
• To leverage neural networks for representing complex displacement fields and material properties in geodynamic systems.

Main Methods:

• Utilizing neural networks to represent continuous displacement fields within arbitrary geometries.
• Incorporating governing physical equations and boundary conditions into the neural network's loss function.
• Employing a polar coordinate system to precisely model fault displacement discontinuities as boundary conditions.

Main Results:

• Demonstrated the validity and utility of the PIDL approach through example problems involving strike-slip faults.
• Successfully modeled crustal deformation with accurate representation of fault slip.

Conclusions:

• The proposed PIDL method offers a powerful and flexible alternative to conventional crustal deformation modeling techniques.
• This approach shows significant potential for extension to high-dimensional, anelastic, nonlinear, and inverse problems in seismology.