Forecasting Extreme Weather at Local Scales with the Global NASA GEOS Model
NASA’s Goddard Earth Observing System (GEOS) model produces highly accurate representations of a variety of weather and atmospheric phenomena ranging from hurricanes to tornadic supercells to aerosol atmospheric rivers. NASA researchers are currently using this state-of-the-art model to run a very high-resolution global 2-kilometer (km) stretched-grid system for forecasting extreme weather over the continental United States. Comparisons with global satellite observations demonstrate GEOS to be a true digital replica of the global Earth system. Additionally, GEOS contributes these convection-allowing forecasts to the Spring 2024 Hazardous Weather Testbed (HWT) at the National Oceanic and Atmospheric Administration (NOAA) Storm Prediction Center. The NOAA HWT examines and evaluates a variety of model output daily to explore different types of guidance and provides focused feedback to model developers. GEOS production modeling and assimilation systems run daily around-the-clock on NASA Center for Climate Simulation (NCCS) supercomputing resources. During the highly active severe storm season of 2024, GEOS accurately forecasted severe thunderstorm and tornado outbreaks up to three days in advance.
The extensive amount of high-resolution 2-kilometer forecast data will be used to inform AI models, improving model parameterizations and enhancing our predictions at conventional modeling and forecast resolutions. All this must happen in concert with expected advancements in supercomputing speed and processing, a key complement to the atmospheric science developments.
Gary Partyka,
NASA Goddard Space Flight Center
Quick Facts
- Current GEOS production systems utilize 210 Intel Cascade Lake nodes (31% of the Cascade Lake nodes at NCCS, 4% of all CPUs at NCCS).
- The experimental 2-km stretched grid model is running daily on 660 nodes of the latest AMD Milan nodes at NCCS (47% of the AMD Milan nodes at NCCS, 36% of all CPUs at NCCS).
- The results of the experimental 2-km stretched grid represent important regional- and local-scale impacts of weather in highly accurate 3-day forecasts, with global weather prediction skill comparable to global numerical weather prediction systems.
- The goal is to run these high-resolution models in order to make routine 10-day forecasts as the flagship numerical weather prediction system at NASA’s Global Modeling and Assimilation Office.
Researchers
- Gary Partyka, NASA Goddard Space Flight Center Please enable Javascript to see email address.
- Bill Putman, NASA Goddard Space Flight Center Please enable Javascript to see email address.
