ABSTRACT
This chapter investigates strategies to scale global adjoint tomography to unprecedented levels by assimilating data from thousands of earthquakes. It focuses on improvements targeting not only current supercomputers, but also next generation systems, such as the Oak Ridge Leadership Computing Facility's Summit. Scalability improves as the scale of the simulation becomes larger and larger. Adjoint tomography workflows consist of a series of iterations. When researching which workflow engine would be the most appropriate for global adjoint tomography, the need to restrict the signification of the word workflow emerges. Indeed, workflow and workflow management have very different meanings depending on the application domain. Assimilating a large number of seismic time series from numerous earthquakes requires improvement of legacy seismic data formats, including provenance information and the ability to seamlessly integrate in a complex data processing chain. Seismic tomography involves the construction of images of Earth's interior by minimizing a predefined misfit function involving observed and synthetic data.
