Computational geodynamicist at the University of Mainz, Germany
This minisymposium will feature the use of the differentiable programming paradigm applied to Earth System Models (ESMs). The goal is to exploit derivative information and seamlessly combine PDE-constrained optimization and scientific machine learning (SciML). Speakers will address (1) Why differentiable programming for ESMs; (2) What ESM applications are we targeting?; and (3) How are we realizing differentiable ESMs? Target ESMs include ice sheet, ocean, and solid Earth models.
The accelerating outflow of ice in Antarctica or Greenland due to a warming climate or the geodynamic processes shaping the Earth share common computational challenges: extreme-scale high-performance computing (HPC) which requires the next-generation of numerical models, parallel solvers and supercomputers. We here present a fresh approach to modern HPC and share our experience running Julia on thousands of graphical processing units (GPUs).