8.3 The Development of a High Resolution Environmental Modelling System for New York State

Tuesday, 30 January 2024: 5:00 PM
324 (The Baltimore Convention Center)
Anthony Paul Praino, IBM Thomas J Watson Research Center, Ossining, NY; IBM Thomas J. Watson Research Center, Yorktown Heights, NY; and L. A. Treinish and M. Tewari

In our continuing work developing advanced applications of environmental models, we present an overview of a high resolution modeling system for New York State. The computing architecture provides a flexible platform, which supports a multi-model approach. We present the numerical weather prediction component of this effort, IBM’s “Deep Thunder”TM which is derived from a configuration of the WRF-ARW community model. The overall system will ultimately support several connected models for meteorology, hydrology, and hydrodynamics. An orchestrated framework of such models enables efficient and scalable research and solutions development for environmental challenges. An example of this is the operational model ecosystem implemented for The Jefferson Project at both Lake George and Chautauqua Lake where the framework enables daily prediction of weather, precipitation runoff and lake circulation in support of research areas such as water chemistry, food web modeling, invasive species, and harmful algal blooms. The modelling platform as it is currently implemented provides forecast and hindcast capability for all of New York state as well as much of New England, Pennsylvania and New Jersey. The model system builds upon an operational history that spans over two decades producing three-day model forecasts at one kilometer resolution. The current model configuration is composed of three nests at 9, 3, and 1 km with 51 vertical levels and explicit cloud microphysics. Data used to drive the model include NOAA/NCEP RAP and GFS and observations from NOAA MADIS and the New York State Mesonet. The software components and model environments are implemented on a dedicated high performance computing cluster but have been containerized to facilitate deployment, scalability, and scheduling. The current configuration automates the parallelization of the model execution in a native or containerized environment on a high-performance computing cluster. The flexibility of the approach allows the implementation on a cloud computing platform. This approach further extends scaling and flexibility of the model components. We will discuss the current implementation as well as computational and forecasting challenges. As a new capability, we will outline our plans to refine the model for specific applications in the region.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner