2:00 PM-2:45 PM
New Hampshire Ballroom
Chair: Layne T. Watson, Virginia Polytechnic Institute and State University,USA
Numerical weather prediction has always been a leading-edge application on supercomputers, because of the need to provide timely forecasts using sophisticated models with spatial resolution as high as possible. Currently, ECMWF runs a global spectral model with 60 unequally spaced levels in the vertical and a horizontal resolution equivalent to a gridlength of 60 km; research forecasts have already been run using this global model with a horizontal resolution of 25 km. Computational requirements dictate that the models must run on parallel machines. Algorithmic improvements have also led to a speedup by a factor of 50-100 over the past decade or so.
Providing initial conditions for the model integration (the "data assimilation" problem) presents an even bigger task, which in the current ECMWF system is formulated as an enormous four-dimensional variational problem; this generates a further et of computational challenges.
The speaker will discuss these aspects of the numerical weather prediction problem and present some recent high-resolution results.Clive Temperton