An Operational Update and Roadmap
Brett Wilt, James Cipriani, John Wong, David Heeps
The Weather Company, an IBM Business
Since August 2018, The Weather Company (TWC), an IBM Business, has provided operational global numerical weather prediction solutions utilizing NCAR’s Model for Prediction Across Scales (MPAS) community model. While the primary focus of early implementation was to replace legacy Global Weather Research and Forecasting Model (GWRF) applications, IBM and UCAR embraced a collaboration opportunity in developing a next-generation high performance computing global model based on IBM’s POWER9 processor and NVIDIA Tesla V100 GPU acceleration.
As a result of the IBM-UCAR collaboration, the world’s first high-resolution, convective allowing, hourly-updating global weather forecast model was introduced by IBM in 2019 – marketed as IBM Global High-Resolution Atmospheric Forecasting System (GRAF). Using a customized MPAS variable-resolution global 15/3km hexagonal mesh of 24 million cells, GRAF provided hourly updating precipitation forecasts (5-minute temporal resolution) for TWC’s short-term weather forecast platforms and applications. In addition, an extended version of GRAF provided a 4x daily 72-hour forecast on a variable 15/4km mesh, with the higher resolution 4km refinement regions across CONUS and Europe. The extended GRAF solution became a foundational asset and tool across media applications, including widespread use throughout the broadcast television weather industry. As part of the overall GRAF research and development cycle, TWC leveraged weekly feedback from a diverse user group of broadcast meteorologists, as well as a proactive engagement of social media feedback. This not only empowered clients to directly contribute to the ongoing success of GRAF, but also resulted in key precipitation and temperature forecast skill improvements.
In 2022, planning and development commenced for the next-generation GRAF initiative for continued scientific investment, including a significant HPC expansion and unification of multiple GRAF applications. With an effective 2.5x increase in compute power, a new IBM weather prediction paradigm was made possible via a high-resolution, convective allowing, hourly-updating global 72-hour forecast. Data assimilation will be an integral component for the next-generation GRAF, with a focus on transitioning from a non-cycled GSI (Gridpoint Statistical Interpolation) approach to a fully cycled, rapidly updating JEDI (Joint Effort for Data assimilation Integration) implementation. The TWC JEDI development path will include an incremental progression from 3DVar, Hybrid 3D EnVar, to Hybrid 4D EnVar, using a wide range of observation datasets such as radar, satellite, aircraft, radiosondes, conventional, and TWC proprietary sources.
The presentation will provide an overview of GRAF HPC, mesh configuration, scalability challenges, model updates, forecast skill statistics, and a GRAF roadmap summary.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner