NOAA rolls out AI weather models that promise faster, more accurate forecasts

It marks a shift from physics-based models to data-trained AI systems

Serving tech enthusiasts for over 25 years.
TechSpot means tech analysis and advice you can trust.

What just happened? The National Oceanic and Atmospheric Administration (NOAA) has launched a new generation of artificial intelligence-based weather models that the agency says will produce faster and more accurate forecasts while using far less computing power. The technology became operational early Wednesday, signaling a significant step in NOAA's broader effort to modernize US weather prediction systems.

The models were built and deployed by NOAA's Environmental Modeling Center in coordination with the National Weather Service. A spokesperson for the service, Erica Grow Cei, told CBS News the machine learning technology is not intended to replace existing numerical models that run on complex physical equations. Instead, it supplements those methods, drawing in part from their underlying data. Cei said the traditional models remain one of the information sources used to train the AI program.

For decades, NOAA's core forecasting tool has been the Global Forecast System, or GFS – a physics-based model that simulates the atmosphere through mathematical equations to generate data on temperature, wind, precipitation, ozone, and soil moisture. Within the larger system, individual components address land, ocean, and atmospheric conditions.

The Global Ensemble Forecast System, or GEFS, was later developed to mitigate certain biases within the GFS by running multiple simulations to capture uncertainty across different weather scenarios.

This AIGFS forecast map for December 10, 2025, shows heavy precipitation from an atmospheric river hitting the US Pacific Northwest

According to Daryl Kleist, deputy director of NOAA's Environmental Modeling Center, the new AI systems were trained using decades of data from those traditional models. Kleist said that "for these AI models, much of the gain in skill is owed to the fact that they were trained on analysis data," which originated primarily from the older numerical modeling frameworks.

// Related Stories

• OpenAI will burn hundreds of billions of dollars through 2030, says HSBC
• Google and Yale's new AI just made a major cancer discovery

The agency estimates that the new AI systems will use between 91% and 99% less computing power than conventional forecasting models, dramatically reducing the resources required for real-time prediction. They may also extend forecast range by as much as 18 to 24 hours.

However, Kleist noted that while the models consume significantly less energy during operation, those numbers do not account for the energy-intensive process of training the AI systems.

Three models form the foundation of NOAA's new AI forecasting portfolio. The first, the Artificial Intelligence Global Forecast System (AIGFS), is described as a "weather forecast model that implements AI to deliver improved weather forecasts more quickly and efficiently" than the traditional GFS.

NOAA said a single 16-day forecast using the AIGFS uses only 0.3% of the computing resources of the operational GFS and finishes in approximately 40 minutes. That level of speed, the agency said, gives forecasters access to updated projections far sooner than before.

The Artificial Intelligence Global Ensemble Forecast System (AIGEFS) builds on that platform by generating a range of potential forecast outcomes rather than a single deterministic projection. A third model, the Hybrid-GEFS, combines the new AI technology with NOAA's existing ensemble system to refine forecasts that account for forecast uncertainty.

Work on the systems remains ongoing. NOAA said scientists are continuing to enhance the AI models' performance in hurricane forecasting and in the range of possible outcomes generated by the ensemble system.