Sea ice regulates mass, heat, and momentum exchange between the ocean and atmosphere in polar regions. Its formation and melting drive the global thermohaline circulation, and changes in its extent significantly affect planetary albedo. Moreover, leads or fractures in the ice allow significant amounts of heat and moisture exchange between the ocean and atmosphere at high latitudes. To accurately simulate the complex interactions between sea ice, atmosphere, and ocean in the coupled system, a sea ice model must be able to capture large seasonal changes in ice cover as well as complex deformation and lead opening due to atmospheric winds and oceanic currents. Lagrangian particle methods have certain advantages over Eulerian grid-based methods for sea ice modeling due to their ability to naturally handle advection, maintain a sharp ice edge, and capture large deformations. 

In this talk we describe the elastic-decohesive sea ice constitutive model being implemented within the MPAS-Seaice framework in the Energy Exascale Earth System Model (E3SM).   The more descriptive physics contained in the constitutive model provides a mechanism to model leads explicitly, and to account for frazil ice and dense water formation in leads, for example. The improved representation is expected to generate more realistic coupling fluxes to the atmosphere and ocean.  The model is implemented using the Material-Point-Method (MPM), which couples Lagrangian particles with an efficient mesh-based solution of the equations of motion. MPM provides a Lagrangian representation of sea ice which allows transport of state variables and tracers without numerical diffusion, as well as history along trajectories for constitutive models. Platform portable GPU acceleration is achieved through the polyMPO library which supports material point operations on spherical centroidal Voronoi tessellation/unstructured meshes and is built using the Kokkos library and the PUMIPic library from the SciDAC FASTMath Institute.
To assess model performance, techniques for parameter calibration and model validation (especially for leads) are being developed. These techniques depend on new metrics for analyzing spatio-temporal data with discontinuities, and/or lower dimensional features.