7.2.b.2. The Model for Prediction Across Scales (MPAS)

The Model for Prediction Across Scales (MPAS) is a next-generation modeling system being developed for weather, regional climate, climate, and Earth system research and prediction. It is designed to simulate the interactions of small-scale phenomena (e.g., clouds, small hydrologic basins, and small estuaries) with large-scale phenomena (e.g., planetary atmospheric waves and Earth-ocean circulations). A critically important aspect of this project is to develop MPAS components (atmosphere, ocean, land surface models, etc.) that scale well on new computer architectures and can simulate all necessary scales for NCAR and community research and production applications.

In FY2013, NESL scientists achieved a number of accomplishments including:

  • As of June 2013, MPAS is freely available to the community in a joint release with DOE.
  • MPAS-Atmosphere (MPAS-A) is now running as a atmospheric component in CAM/CESM using PIO. Initial testing (aquaplanet (APE) and AMIP testing) is just beginning.
  • The CESM/CAM radiation package has been implemented; the GFS physics port is still in progress.
  • MPAS-A has been used in real-time NWP testing for the 2013 hurricane season where daily 10-day forecasts were produced on a uniform 15 km global mesh and a variable-resolution 60-15 km mesh where the high-resolution region was centered over the Atlantic hurricane basin.  While the Atlantic hurricane season has been quiet, forecasts from both MPAS configurations captured many of the observed tropical storms.  The strong typhoons in the western Pacific have been well forecast by the uniform resolution 15 km MPAS configuration.  Yearlong regional climate simulations have also been produced using MPAS-A and a data ocean; these results are currently being analyzed.
  • MPAS-A has been run using uniform global meshes of 30, 15, 7.5 and 3 km as part of an NCAR/ASD allocation on Yellowstone.  The initial scaling results have met expectations with regard to the solver performance and have demonstrated MPAS ability to perform capability computations on Yellowstone. The results also have uncovered problems associated with the scaling of the Parallel IO package in MPAS.
  • MPAS-A is now fully coupled to the ensemble Kalman filter in the Data Assimilation Research Testbed (DART).  Preliminary results with uniform meshes found no significant problems with MPAS-A, and variable-resolution mesh testing is underway.
  • The variable-resolution testing of MPAS has been undertaken within the tropical cyclone forecast experiment.  The development of scale-aware physics is focused on the deep convection parameterization, and several efforts are being led by investigators in other labs (EPA, NOAA, DOE) in collaboration with MPAS developers.
  • Under the NCAR/ASD allocation, hydrostatic- and nonhydrostatic-scale tests have been performed that verify the accuracy of the MPAS-A global solver at these scales.  Most important, convective-scale structures are well simulated at the nonhydrostatic scales, and are comparable to those produced with state-of-the-art could models (e.g. WRF).  An example of these convective structures is given in the Figure below.

  Figure: Supercell thunderstorms simulated in global MPAS using a 3 km mesh in a 4 day 3 hour forecast valid 3 UTC 27 October 2010.  The left panel depicts the column-maximum radar reflectivity that shows the isolated severe convective cells ahead of the cold front.  The middle panel shows the cold pools associated with the convection in the warm sector ahead of the cold front.  The right panel shows the classic splitting supercell structure in the upper level vertical velocity field.