Efforts are underway through two DOE Biological and Environmental Research (BER) funded partnership projects to advance simulation workflows to better understand sea-level rise due to changes in land and sea ice. Land ice workflows will tightly couple Omega_h GPU accelerated mesh adaptation to better resolve the evolving ice front, fractures, and areas of with large discretization errors in velocity fields. Conservative field transfer methods are critical to the quality of the subsequent analysis on an adapted mesh are are likewise being supported through the Parallel Coupler for Multimodel Simulations (PCMS) and MeshFields libraries.
Sea ice workflows are coupling the PUMIPic to MPAS-Seaice-MPM to provide GPU acceleration of computationally expensive steps of the material point method. Use of MPAS mesh based on spherical centroidal Voronoi tessellations (SCVT), and the dual mesh, are crucial. Material point and mesh operations of interpolation and reconstruction, and tracking of material points, on SCVT-based meshes are fully supported on GPUs by the polyMPO library build on top of PUMIPic.
Primary Contacts: Onkar Sahni and Cameron Smith
SCOREC Researchers: Castillo, Nath, Sahni, Shephard, Smith
External collaborators: Sandia National Laboratories, University of New Mexico
Grounding line for Greenland Icesheet extracted with Compass (top).
Detailed view of the Simmetrix mesh within the red rectangle (bottom).
Developments to Support the Modeling of Sea and Land Ice
Applying GPU accelerated mesh adaptation to land ice domains requires the extension of the existing COMPASS ice sheet mesh generation workflow to provide a high resolution representation of the initial geometric model features and the meshes classification on them. During the adaptive analysis, mesh motion, mesh adaptation, and geometric model topology and geometry update procedures are applied to track the level set representing the grounding line and calving front. Omega_h developments include data structures and methods to store and query non-manifold geometric model information to maintain and improve the geometric approximation during mesh adaptation.
Critical to supporting the material point method for sea ice analysis are data structures that effectively support storing, accessing, and distributing material point information associated with mesh elements. To handle relatively low number of material points per mesh element, fast and efficient data structures supported by PUMIPic are used. In particular, polyMPO library has been developed to support all relevant material point (MP)-related operations such as initialization of MPs, transfer of fields from mesh-to-MP and MP-to-mesh (on spherical meshes), MP tracking with specialization for a centroidal Voronoi tessellation, addition and deletion of MPs to account for freezing and melting of ice, and MP migration for multi-GPU cases. polyMPO library is implemented in C++ and is interfaced with the Fortran-based MPAS code using a well-defined set of APIs based on iso_c_binding for language interoperability.
GPU-enabled reconstruction of velocity field based on material point data on a SCVT-based mesh
GPU-enabled reconstruction of velocity field based on material point data on a SCVT-based mesh
GPU-enabled reconstruction of velocity field based on material point data on a SCVT-based mesh