JACKSONVILLE, FL – The National Hurricane Center (NHC) is teaming up with Google DeepMind and Google Research to test a cutting-edge artificial intelligence (AI) model that could improve how we track and predict hurricanes, also known as tropical cyclones.
Hurricanes are among the most destructive natural disasters on Earth, responsible for over $1.4 trillion in damages worldwide over the past 50 years. While forecasting has improved over the decades, predicting exactly where a storm will go, how strong it will get, and when it will form remains a major challenge. Even small changes in the atmosphere can dramatically alter a storm’s behavior.
Enter AI.
What’s New with AI Forecasting?
Traditionally, hurricane forecasts rely on complex physics-based computer models, like those run by the National Weather Service and the European Center for Medium-Range Weather Forecasts (ECMWF). These models simulate the atmosphere and oceans using the laws of physics, crunching through vast amounts of data to predict a storm’s path and strength.
Now, Google is introducing a different approach. Their new AI model uses machine learning, meaning it’s trained to recognize patterns by analyzing huge amounts of past weather data. It doesn’t try to simulate every physical process — instead, it makes predictions based on what it has “learned” from millions of past observations.
The model, part of Google’s new Weather Lab project, can generate 50 different possible scenarios for a hurricane’s track, size, and intensity — looking up to 15 days ahead. That gives forecasters a broader view of what could happen, helping them understand the range of possibilities as a storm develops.
How Accurate Is It?
According to Google, early results are promising. In tests, the AI model’s five-day track forecasts for hurricanes in the Atlantic and East Pacific were, on average, 87 miles closer to the actual storm track than those from the respected ECMWF model during the 2023 and 2024 seasons.
But it’s important to note: this technology is still experimental. The AI model has not yet been peer-reviewed, and it’s not being used for official public forecasts — at least not yet. That’s where the National Hurricane Center comes in.
What Does This Mean for the Public?
The NHC will evaluate how well this AI model performs in real-time during this hurricane season. The goal is to see whether the AI predictions can complement existing tools to provide even more accurate, earlier warnings, giving people more time to prepare, evacuate, or take other safety measures.
The NHC emphasizes that AI won’t be replacing traditional forecasting models or meteorologists anytime soon. But it could become a powerful tool to assist them, especially as hurricanes become harder to predict in a changing climate.
Andrea: First Storm of the Season, but a Limited Test
The first named storm of the 2025 Atlantic season, Tropical Storm Andrea, offered a limited opportunity to see Google’s AI model in action. Andrea was first flagged as a potential tropical cyclone on June 22, organized further on June 23, and officially named on June 24 — only to degenerate about 12 hours later.
Because Andrea formed late in its lifecycle and lasted such a short time, forecasters had very little lead time to work with. The brief nature of the storm means it may not provide a sufficient case study to properly judge the AI model’s performance compared to traditional models like ECMWF.
However, the storm still provided an early-season opportunity to test the AI’s ability to predict storm formation and initial track — critical pieces of information for communities watching the tropics.
Weather Lab: A Glimpse Behind the Scenes
Google’s Weather Lab is a new interactive website where people can explore how these AI hurricane forecasts compare to traditional models like ECMWF. It’s designed for research and education — not for making personal safety decisions. Official forecasts still come from trusted sources like the NHC.