Locating critical minerals, monitoring geologic storage of CO₂, managing aquifers, remediating land, and tracking permafrost change are just a few examples of geoscientific problems that require us to characterize the subsurface. Much like how medical data, such as an MRI or X-rays, are used for non-invasive imaging, geophysical surveys collect data that are sensitive to the physical properties of the subsurface. Reconstructing the distribution of subsurface physical properties from measured data gives rise to a PDE-constrained optimization problem that requires efficient simulations and sensitivity calculations for large-scale PDEs. Because the inverse problem is ill-posed, it is essential to incorporate prior information such as petrophysical, geologic, and geochemical measurements, as well as complementary geophysical data sets. Such information can be incorporated into the inversion through the model parameterization, design of an appropriate regularization, or through joint inversion strategies. In this talk, I will outline how geophysical inverse problems are formulated and solved, and will highlight current research combining machine learning with physics-driven methods to characterize the subsurface. The methods that we develop are applicable to a range of geophysical data types, but I will focus on electric and electromagnetic methods in the context of mineral exploration and environmental monitoring.

Refreshments will be served preceding the talk, starting at 2:45.