Discrete Element Method for High-Resolution Sea Ice Models
On the DEMSI project, we will develop a new sea-ice dynamical core using the discrete element method, where collections of floes are explicitly modeled as discrete elements. This shift to a particle-based method from the current continuum model is necessary to represent the dynamics of sea ice at high resolution.
Current DOE Earth system modeling efforts use a sea-ice component that approximates the sea-ice cover as a continuous material and assumes sufficient cracks exist within each model grid cell to ensure an isotropic distribution of crack orientations. Such models are only appropriate at low resolutions, but DOE modeling efforts use much higher resolutions. To correct these deficiencies we will develop a new sea-ice dynamical core using the discrete element method, where collections of floes are explicitly modeled as discrete elements. We will use the Large-scale Atomic/Molecular Massively Parallel Simulator code developed at Sandia National Laboratories for the dynamical core, and modify this code to better utilize heterogeneous computing architectures with the kokkos programming model. An element contact force model will be developed that will more accurately reflect the brittle, highly heterogeneous and anisotropic nature of the ice pack. Methodologies to account for the element deformation and destruction, as well as creation of new elements, will be developed, incorporating a new multi-scale sea-ice morphology. We will also develop coupling techniques appropriate to couple discrete element models to the other fixed-grid based components of Earth system models.
This DEMSI project includes FASTMath staff on the development team. The main technical overlaps are with regards to performance portability and software engineering that are common to both projects. We will look for opportunities to pull in other FASTMath capabilities such as load balancing and unstructured grid tools for coupling grids where they can have an impact.