Earth Sciences Seminar: Dr. Sihong Wu

Title: AI-enabled electromagnetic imaging with quantified uncertainty from near-surface to deep earth.

 

Electrical resistivity is a fundamental physical property that is highly sensitive to lithology, fluid content, and salinity. Electromagnetic (EM) methods are therefore widely used in groundwater investigation, environmental monitoring, resource exploration, and studies of Earth structure and geodynamics. However, translating EM observations into reliable subsurface images remains challenging due to data noise, inherent non-uniqueness of inversion, and the growing computational demands associated with large-scale datasets. Recent advances in artificial intelligence (AI) offer new opportunities to address these challenges in EM data analysis. In this talk, I will present an AI-enabled EM imaging framework that spans the full workflow, from data denoising and forward modeling to inversion and uncertainty quantification. This integrated approach improves sensitivity to deeper electrical structure, enables near-real-time subsurface imaging, and facilitates quantitative assessments of uncertainty. I will feature two continental-scale case studies that illustrate these capabilities. The first is the largest conductivity model of Australia’s shallow subsurface, imaging structures to depths of ~650 m with quantified uncertainty and characterizing resistivity variations across major geological and hydrogeological units, as well as regions hosting critical mineral occurrences. The second is an AI-driven analysis of USArray magnetotelluric data that yields a first-order view of the electrical structure of the U.S. down to ~400 km while assessing data quality and 3D effects. In summary, AI-enabled EM imaging facilitates scalable Earth characterization with quantified uncertainty and holds strong potential to advance both scientific understanding and applied geophysical decision-making.