Ultra-high-resolution CCSM
Over the last several years we have made concerted efforts to significantly improve the scalability of the various components of the upcoming Community Climate System Model (CCSM4). We initially modified the Parallel Ocean Program (POP) to improve its efficiency at 0.1° resolution on large processor counts. We next improved the simulation rate of the Community Ice CodE (CICE) at 0.1° on large processor counts. In collaboration with the Climate and Global Dynamics (CGD) division, the CCSM coupler was significantly redesigned to increase the flexibility of the coupling architecture. These improvements to the coupler, along with scalability enhancements to component models, enable CCSM4 to execute efficiently at high resolution on systems with large numbers of processors.
Our attention to the scalability of the entire CCSM system has enabled the first-ever coupling within the U.S. of an eddy-resolving ocean and sea ice model at 0.1° to an ultra-high resolution atmospheric model and land model at both 0.5° and 0.25°. Using a grant through the Lawrence Livermore National Laboratory Institutional Grand Challenge program, we have competed 2.5 years of a fully coupled simulation using 0.5° atmosphere, and 2 years of a fully coupled simulation using a 0.25° atmosphere.
The winter season precipitation
rate (mm/day) in the North Atlantic from CCSM using 0.5° atmosphere
and land models coupled to a 1° (left panel) and a 0.1° (right panel)
ocean and sea ice model. The more realistic representation of the Gulf
Stream in the 0.1° Parallel Ocean Program (POP) causes the atmospheric
winter storms to become stronger and track along the Gulf Stream well
offshore versus the lower-resolution model where the maximum precipitation
area is confined closer to the coast.
Our discovery is significant because it demonstrates that it is possible to simulate the Earth System on currently available supercomputer resources at resolutions that are 100 times as computationally demanding as the current production CCSM simulation. Our work demonstrates that it will be possible to utilize the upcoming NSF petascale system to finally resolve many of the important physical processes with the Earth System that have been previously only been parameterized.
Future plans include continuing to improve the scalability of each component model and the entire coupled system through improvements in partitioning and decomposing the computational grid across processors. We also plan to complete the addition of the Parallel I/O (PIO) library into all component models.
This work advances NCAR's strategic priority of "Conducting research in computer science, applied mathematics, statistics, and numerical methods." Our work is supported through the NSF cooperative Grant NSF01, and through the Department of Energy CCPP program grant #DE-FC03-97ER62402.