AI Hurricane Forecasts: What Changed for the Gulf Coast

This is the time of year when Gulf Coast homeowners watch the forecast cone. The models behind that cone changed in the last two years. Artificial intelligence (AI) weather models now run at the major forecast centers, next to the physics-based models that forecasters have used for decades.

The AI models

Europe. The European Centre for Medium-Range Weather Forecasts (ECMWF) took its Artificial Intelligence Forecasting System (AIFS) into operations on 25 February 2025.¹ ECMWF says the AIFS outperforms physics-based models on many measures, including tropical cyclone tracks, “with gains of up to 20%.”¹ It also says the system uses approximately 1,000 times less energy to make a forecast.¹

United States. On December 17, 2025, the National Oceanic and Atmospheric Administration (NOAA) announced three operational AI models: a global forecast model (AIGFS), an AI ensemble that provides a range of probable outcomes (AIGEFS), and a hybrid ensemble that combines AI with NOAA’s flagship ensemble model (HGEFS).² NOAA reports a significant reduction in tropical cyclone track errors at longer lead times.² NOAA also reports that version 1.0 shows a degradation in tropical cyclone intensity forecasts.²

Google DeepMind. In June 2025, Google DeepMind released an experimental cyclone model that generates 50 possible scenarios, up to 15 days ahead.³ The company said that in its own tests on 2023 and 2024 storms, its 5-day track prediction was, on average, 140 km closer to the true storm location than the leading European physics-based ensemble.³ It also announced a partnership with the U.S. National Hurricane Center.³

How the 2025 season went

A WAFB report on the National Hurricane Center’s 2025 results gives the first full-season record.⁴

• The Google DeepMind model slightly outperformed the National Hurricane Center’s official track forecasts in the short term, up to about three days out.⁴
• Among the traditional models, the European model performed best at four to five days out.⁴
• Intensity errors were higher than average in 2025.⁴
• There were 23 rapid intensification events, and the National Hurricane Center’s forecasts were generally more accurate than the models in those situations.⁴

What is new this month

On 3 August 2026, ECMWF described a correction for a known weakness of its AI model: it underestimates the most intense storms.⁵ In the 2025 evaluation, the uncorrected forecast had a global wind-speed bias of almost −29 knots, and the correction reduced it to about −2 knots.⁵ ECMWF says the correction reduced the intensity error by roughly a factor of three.⁵ The correction is research work. ECMWF says the next steps are to implement it as a real-time system.⁵

On August 17, Singularity Hub reported on Google DeepMind’s cyclone model. The report says existing physics-based forecasts extrapolate two days ahead, and that DeepMind says its model extends the warning period to three days without a loss of accuracy.⁶ The same report says the AI model will not replace conventional models, and that the two give human forecasters better information.⁶

These are forecast models. None of the sources describes them as agents, and a human forecaster still issues the official forecast.

What this means on the Gulf Coast

1. Track forecasts are better, earlier. A firmer track at three to five days gives more time to secure a property, protect a roof, or move a boat.
2. Intensity is still uncertain. Rapid strengthening is the case the models handle worst. Prepare for a stronger storm than the forecast shows.
3. The official forecast is the one to act on. The National Hurricane Center publishes it at nhc.noaa.gov. An AI model is one input to that forecast.

If a storm damages your home or business, photograph the damage before you repair it, and keep every receipt. Our emergency response page explains what we do in the first days. Call (985) 255-2435.