The massive supercomputers predicting the atmosphere's next 16 days

Space · Published 18 October 2026 Swipe more like this →

In short: Weather forecasting is a battle of brute-force mathematics. The Global Forecast System (GFS) uses a global computer model to simulate the atmosphere, running four times a day to predict conditions up to two weeks out. It is the backbone of global weather, powering both government agencies and private companies.

The GFS operates by dividing the atmosphere into a grid. Its current dynamical core, the Finite-Volume Cubed-Sphere (FV3), features a horizontal resolution of approximately 28 km between grid points. Vertically, the model is sliced into 127 layers extending up to the mesopause, roughly 80 km high. To manage computational demands, the model's spatial resolution intentionally drops to 70 km for forecasts spanning one to two weeks, as accuracy naturally decays over time. The system generates output every hour for the initial 120 hours. After that, the frequency shifts to three-hourly intervals through day 10. By day 16, the model provides updates every 12 hours. These calculations also help produce essential model output statistics.

The system has undergone significant technological evolution to keep pace with competitors like the European Centre for Medium-Range Weather Forecasts (ECMWF). In 2015, the GFS notably struggled with Hurricane Sandy's trajectory, leading the National Weather Service to invest in supercomputers that boosted processing power from 776 teraflops to 5.78 petaflops. Further upgrades in 2018 pushed power to 8.4 petaflops. A major shift occurred in 2019 when the model transitioned from a spectral method to the FV3 dynamical core, which improved large-scale prediction and hurricane track accuracy. The agency previously tested an icosahedral model called FIM but abandoned it in 2016. This decision followed a lack of substantial improvement over the existing GFS. The 2017 upgrade also improved the tracking of the Saharan Air Layer.

Modern iterations, such as the GFSv16 implemented in March 2021, now couple the model with the WaveWatch III global wave model. This integration allows for a 16-day forecasting window and increased vertical resolution. Because GFS data is in the public domain, it serves as a primary foundation for numerous commercial and foreign weather services worldwide, balancing high-level scientific complexity with open-access utility. The GFS also supports a 30-member ensemble known as the GEFS. This ensemble combines with Canada's Global Environmental Multiscale Model to create the NAEFS. Additionally, the system includes a 32nd member for aerosol predictions.

General information, not professional advice. Spotted an error? Flag the card or see corrections and complaints.

Intellect Plus is a visual educational discovery platform for exploring fascinating facts and ideas. Brain Food, Not Brain Rot. What is Intellect Plus?

Related

• How supercomputers turn current weather observations into tomorrow's forecast using complex mathematical models.
• Without parallel computing, tomorrow's weather would be predicted 1,000 years later
• Why even the most advanced supercomputers cannot perfectly predict tomorrow's weather
• Can a machine learning model outperform massive, expensive weather forecasting supercomputers?
• Three-quarters of modern weather accuracy relies on eyes in the sky
• The invisible weight of the atmosphere shapes our daily weather
• Why the sky doesn't actually signal the ground's next move
• The lost art of reading the sky for weather shifts
• The mysterious North Atlantic cold blob might be a warning sign
• The global weather system is a product of celestial mechanics
• The archer's bow that can devastate entire cities
• Could a single asteroid impact trigger a storm the size of North America?

More in Weather · More in Space · Latest · A–Z · All topics