Advancing Our Understanding of Earth’s Interior with AI for Geophysical Data

Dr. Hongyu Sun, Postdoctoral Scholar in Geophysics, Caltech

Cutting-edge techniques improve our way of visualizing the subsurface with seismic data, thus enhancing our ability to understand the Earth’s interior. AI has transformed seismic data analysis, elevating the role of deep learning in seismology. In this talk, I will outline my contributions to improving seismic monitoring and subsurface imaging with AI. I will first present the Phase Neural Operator (PhaseNO) for earthquake monitoring and seismic phase picking. PhaseNO measures the arrival times of P- and S-waves from continuous seismic data simultaneously across input stations with arbitrary geometries. By leveraging the spatial-temporal information, PhaseNO outperforms single-station AI algorithms by significantly detecting more earthquakes and enhancing measurement accuracy. Additionally, I will show how deep neural networks can overcome the complexities in seismic imaging by being trained to generate seismic waves. These waves, although not directly recorded, are essential for imaging the Earth’s interior. I will provide case studies on full-waveform inversion with activesource seismic data and seismic interferometry with environmental noise. In summary, these AI methods are powerful complements to traditional computational methods and hold significant promise for accelerating energy transition and mitigating geohazards.