Our Planet

Convective-Scale Numerical Weather Prediction with the NASA GEOS Model

Image from animation visualizing Hurricane Idalia landfall over Florida
This visualization of radar reflectivity shows a 4-kilometer-resolution NASA Goddard Earth Observing System (GEOS) model forecast accurately following the path of Hurricane Idalia over Florida, at landfall along the northern Gulf Coast, through Georgia and South Carolina, and until its breakup in the Atlantic Ocean. William Putman, NASA/Goddard

Vigorous up-and-down motions of the atmosphere—a process known as convection—vertically transport heat and moisture to fuel the strongest storms. Due to intense computational demands, convective-scale numerical weather prediction (NWP) has traditionally been reserved for regional, limited-area computer models. The flexibility of the NASA Goddard Earth Observing System (GEOS) atmospheric model and assimilation system, combined with access to NASA high-end computing resources, allows researchers to configure a global numerical weather prediction model with 2- to 4-kilometer (km) resolution for convective-scale weather over the continental United States. This combination also allows continuous and seamless integration of global data from the GEOS Forward Processing (GEOS-FP) data assimilation system within the scale-aware global model. The global nature of this application enables researchers to extend the predictability of convective-scale NWP beyond two days with a global accuracy consistent with state-of-the-art global NWP forecasts. These forecasts give an unprecedented look at severe storms, tornadoes, and hurricanes at medium-range timescales, with increased fidelity of local impacts from extreme precipitation, winds, and heat.

Quick Facts

As weather forecasting continues to evolve, global forecast systems—including NASA’s GEOS model and data assimilation system—will emerge as critical tools for producing medium-range forecasts of convective-scale weather. This example with the GEOS model is one step along that path to providing an essential forecast resource for severe weather within the context of seamless global numerical weather prediction applications.

William Putman,
NASA Goddard Space Flight Center
  • GEOS 4-km forecasts ran on the NASA Center for Climate Simulation’s Discover supercomputer using 13,000 Intel "Cascade Lake" nodes, while similar forecasts with 2-km resolution ran on the NASA Advanced Supercomputing Facility’s Aitken supercomputer using 53,000 AMD "Rome" nodes.
  • GEOS convective-scale forecasts are an extension of the global forecast model utilizing a “stretched grid” approach to enable finer resolution over the continental U.S, while seamlessly predicting global weather five days in advance.
  • These convective-scale forecasts from GEOS are being analyzed alongside leading NWP regional models as part of the National Oceanic and Atmospheric Administration (NOAA) Hazardous Weather Testbed Spring Forecast Experiment in 2023.
  • These convective-scale forecasts continued to run through the summer and into the fall of 2023, capturing numerous extreme weather events including hurricanes Idalia and Lee and tropical storm Ophelia.

Researcher

  • William Putman, NASA Goddard Space Flight Center

More Information