German STORM consortium to drive climate modelling beyond the IPCC Fifth Assessment

3 Jun 2010 Hamburg - Prof. Dr. Jochem Marotzke from the Max Planck Institute for Meteorology in Germany was the first speaker in the session on climate modelling at the ISC'10 Conference. Climate science is a driver for HPC requirements, he told the audience. His institution has been participating in the STORM consortium for high resolution climate modelling. During his talk, Prof. Dr. Marotzke showed a few examples of what supercomputing simulation is capable of in climate modelling.

Prof. Dr. Marotzke started out by giving an overview of the requirements by climate science. This research is characterized by a higher complexity: more model components involve more complexity. Longer runs are needed to simulate ice age cycles. A much larger ensemble size provides the means to cope with uncertainty, predictability, and data assimilation in the model-data fusion. A higher resolution offers better results.

For the ocean a 10-fold higher resolution per horizontal dimension is needed and for the atmosphere the researchers equally require a 10-fold higher resolution per horizontal dimension to represent hurricanes for example, explained Prof. Dr. Marotzke.

Within the STORM consortium a three-times better physical process representation has been achieved. More biogeochemical processes were modelled explicitly by a factor of 3 and a much higher resolution from 200 km to 20 km was acquired. There were five times more of large ensembles and much longer runs.

The consortium has been focusing on the challenge of better horizontal resolution, explained Prof. Dr. Marotzke. Simulating hurricanes requires a resolution of a few 10s of km. In IPCC there is typical grid spacing.

The STORM simulations involved multi-century climate change simulations at the highest possible resolution in both atmosphere (40km) and ocean (10 km). The consortium consists of most of the major German ocean and climate research institutions.

The simulations have been performed on a local machine at DKRZ with the following parameters:

  • use of MPI-OM TP6M to avoid convergence of meridians at the North Pole
  • 3602x2394x80 or a total of 969 million grid points
  • Blizzard uses 1920 cores and 2.8 TB of memory and issues 197 forecast days per day
  • the output is 1TB Netcdf4

The STORM simulations also provided an ocean model.

At present the science focus lies on ocean circulation and internal waves. It is climatically the most important component of the ocean circulation. The "meridional overturning circulation" is associated with deep sinking at high latitudes, explained Prof. Dr. Marotzke.

He noted that 2 TW of mechanical power is needed to support the meridional overturning to supply small-scale kinetic energy required for deep-ocean ocean mixing. The energy is thought to come from the breaking of internal waves. One TW is thought to come from winds, and one TW from the tides.

Prof. Dr. Marotzke believes climate science will remain a major driver for HPC enhancements, requiring orders of magnitude of more power. The high resolution simulations are essential for capturing curcial elements of climate dynamics. These were demonstrated for ocean eddies and internal ocean waves.
Leslie Verswevyeld