Extended Abstract (228K)5.9
Modeling Environment for Atmospheric Discovery
Robert B. Wilhelmson, National Center for Supercomputing Applications, Univ. of Illinois, Urbana, IL; and K. Droegemeier, S. Graves, M. Ramamurthy, D. Haidvogel, B. Jewett, J. Alameda, and D. Gannon
The scientific advances and societal benefits associated with the accurate prediction of hurricanes and severe storms are enormous; however, presently available computational and data management frameworks have not yet incorporated the full benefits of grid and web technologies for carrying out and analyzing/mining ensemble simulations. To help address existing limitations and to remove unnecessary human effort, we are developing a scalable framework -- known as Modeling Environments for Atmospheric Discovery (MEAD) -- for use in ensemble prediction and parameter studies. With MEAD it will be possible to launch hundreds of model simulations on a variety of grid resources and manage distributed model data stores and streams. Each simulation will not have to be submitted and catalogued individually in the traditional painstaking method of research. Metadata and the resulting large volumes of data (10’s to 100’s of terabytes) will then be made available through the MEAD portal for analysis, datamining, machine learning, and visualization.
The key modeling components of MEAD include the Weather Research and Forecasting Model (WRF) and the Regional Ocean Modeling System (ROMS). The MEAD effort includes the coupling of these two models for studying hurricanes. MEAD will develop and adapt existing infrastructure as appropriate. Components include datamining with D2K (Data to Knowledge) and ADaM (Algorithm Development and Mining System), ESML (Earth System Markup Language), ROMS and WRF ports to the Intel Itanium platforms, HDF5 parallel I/O, wall visualization, and Chimera (A Virtual Data System).
Key technology challenges include model and grid workflow management, scientific and derived metadata for model simulation and derived data including visualizations, data management of very large computed and derived data sets, model coupling and incorporation of HDF5, interactive analysis/mining and visualization capabilities for large distributed ensemble datasets, and enabling access of MEAD technologies by the scientific community for use on their own clusters as well as the TeraGrid. Funding for MEAD is provided by the National Computational Science Alliance which includes the National Center for Supercomputing Applications
Supplementary URL: http://www.ncsa.uiuc.edu/expeditions/MEAD/
Session 5, Internet Applications and Web Portals
Tuesday, 11 February 2003, 8:30 AM-1:30 PM
Previous paper