Deep learning for deep 3-D Earth imaging

The past decade has seen a formidable progress in the 3-D imaging of the Earth’s deep interior. Huge, continent-size provinces have been mapped at the core-mantle boundary, about ~3,000 km deep beneath our feet, and complex patterns of flow in the deep mantle are being unravelled. This has been possible thanks to the free availability of huge global seismic datasets along with advanced computational methods for simulating seismic wave propagation in 3-D media and for inverse modelling. While millions of seismic measurements are now routinely used in the global seismic imaging of Earth’s deep interior – seismic tomography – the full richness of the data available is still under-explored and full 3-D models of seismic wave propagation are still difficult to use due to very heavy computational costs. In this project we will use novel deep learning approaches based on emulators to accelerate the calculation of accurate wave propagation simulations in full 3-D media. We will also explore the use of deep learning techniques to discover new seismic observables that enhance the resolution of the images. This will enable the next generation of high resolution global seismic images and to better understand the structure and evolution of our planet over millions of years, and their impacts on the Earth’s surface, with wide impacts across Earth Sciences and beyond.