The primary goal of the project is to develop a new atmospheric dynamical core designed specifically for heterogenous computing architectures. The foundation of our project is an algorithm that combines the data-locality of spectral element methods with the efficient time stepping of semi-Lagrangian methods. We predict a near-term performance gain of 2x over the current ACME/HOMME dynamical core with further performance gains likely in the longer term. This project simultaneously exploits and further develops synergies between evolving hardware architectures and algorithm design to push the boundaries of resolved scales beyond the 10 km hydrostatic limit and into the non-hydrostatic regime.
We will perform studies and analysis of high-resolution simulations of gravity waves interacting with frontogenesis as a demonstration application. This work will facilitate the direct numerical simulation of multiscale interactions not captured well in today's models.
We also anticipate that explicit resolution of a wider range of scales can reduce sources of uncertainty that arise from non-unique sub-grid parameterizations in lower resolution models.