6B.4 Numerical Weather Prediction at IBM/The Weather Company: Overview of a Global Rapidly-Updating Forecast System

Tuesday, 5 June 2018: 11:30 AM
Colorado B (Grand Hyatt Denver)
Todd Hutchinson, The Weather Company, Andover, MA; and B. A. Wilt, J. Wong, and J. P. Cipriani

The Weather Company (TWC), an IBM Business, has been running numerical weather prediction (NWP) systems internally for the past 20 years, and has referred to these systems under the trade name “RPM”. Independently, IBM Research has also been running localized NWP systems for over two decades and has referred to these systems as “Deep Thunder”. The systems have ranged from local domains driven by MM5 to continental-scale convective-allowing forecasts driven by WRF. Currently the next generation system is under development as a global NWP system. It will acquire the name “Deep Thunder”, following on from the name given to the IBM research system.

The Model for Prediction across Scales (MPAS), developed by the National Center for Atmospheric Research (NCAR), will serve as the core global numerical weather prediction system, while the Community Gridpoint Statistical Interpolation (GSI) system will serve as the core data assimilation (DA) system. Initially, GSI will be run as a 3DVAR system and later transition to a hybrid ensemble-variational system. GSI will be used to assimilate a variety of observations, including satellite (GOES-R, Himawari, POES, etc.), conventional (METARs, SYNOPs, cell-phone pressures, radiosondes), and radar. MPAS will integrate the GSI analysis forward in time to provide rapidly-updating global forecasts for the day ahead time period. Horizontal resolution will vary from convective-allowing over the populated areas of the world to coarser (~15km) resolution over the rest of the world.

TWC, IBM Research, and NCAR have partnered to develop a GPU-accelerated software port of MPAS to IBM Power9 systems with nVidia GPUs. It is expected to require less resources to achieve the same performance compared to conventional CPU-only computing systems. This advantage translates to smaller initial investment and ongoing operating costs.

The forecast output from the system will be used throughout TWC product lines, including consumer (e.g., television, web and mobile forecasts), aviation (e.g., turbulence and convective hazards), energy demand, and insurance applications.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner