Google is set to enhance its capabilities in predicting tropical cyclones by deploying a new artificial intelligence model, in collaboration with the US National Hurricane Center (NHC).
Today, Google DeepMind and Google Research unveiled a new platform, Weather Lab, dedicated to sharing the AI-driven weather models being developed by the tech giant. This innovative model claims the ability to produce 50 distinct forecasts for a storm’s trajectory, size, and strength, extending predictions up to 15 days ahead. The NHC is currently assessing this model’s performance.
Recent advancements do not diminish the importance of conventional weather models.
This development follows reductions to the National Weather Service’s workforce and its capacity for climate and weather research during the previous presidential administration. Numerous companies and agencies are investigating the potential of AI to enhance forecasting, yet current technological improvements still require the foundational support of traditional models.
Today, Google also published a research paper outlining the mechanics of its tropical cyclone prediction model, which has yet to undergo peer review. Early findings suggest that the model’s accuracy rivals that of established physics-based forecasting systems. The NHC’s evaluation will provide further insights as the Atlantic hurricane season progresses through November.
The primary goal is to enhance the NHC’s forecasting capabilities to ensure timely and precise storm warnings, allowing individuals to prepare adequately. According to Google, its model’s five-day projections for cyclone paths in the North Atlantic and East Pacific were, on average, 87 miles (140 km) closer to the actual storm trajectories compared to forecasts provided by the European Center for Medium-Range Weather Forecasts (ECMWF) in the past two years.
The Weather Lab’s interactive features allow users to compare Google’s AI models with ECMWF’s physics-based predictions. However, Google stresses that this platform is intended for research purposes at this time and should not be considered a definitive forecasting tool for the public.
The cyclone model has been trained using data from Europe’s ERA5 archive, which incorporates extensive observations gathered globally by various weather agencies, alongside predictions from a classical weather model. This approach mirrors Google’s previous AI weather model, GenCast, which reportedly surpassed one of ECMWF’s top physics-based models 97.2 percent of the time, according to a study published in Nature in December 2024.
Animation depicting the predictions from Google’s model for Cyclone Alfred, a Category 3 cyclone in the Coral Sea. The model effectively predicted Cyclone Alfred’s rapid decline to tropical storm status and subsequent landfall near Brisbane, Australia, a week later, indicating a high probability of landfall along the Queensland coast. Credit: Google
The company is also collaborating with the Cooperative Institute for Research in the Atmosphere at Colorado State University, along with researchers in the UK and Japan, to refine its AI weather models further.
The integration of real-world observations and conventional weather models remains crucial in developing these innovative tools, highlighting why AI is currently expected to supplement rather than replace traditional forecasting methods. Adaptation to a shifting climate will depend on the capacity to gather and analyze data concerning increasingly severe and unpredictable weather patterns.
Concerns are mounting about the ability of the United States to maintain momentum in this line of research, especially following significant staff cuts at the National Oceanic and Atmospheric Administration (NOAA), which oversees the NHC and the National Weather Service. The National Weather Service has reduced weather balloon launches in light of workforce reductions, and NOAA is increasingly depending on private companies for weather balloon data. Project 2025 has proposed dismantling NOAA, which leads crucial climatic research in addition to weather forecasting, raising concerns over the potential commercialization of these services. Some advocates express distress over the move to transform weather forecasts into a paid service rather than a public utility accessible to all.
“For a long time, weather has been viewed as a public good, and I think, you know, most of us agree with that … Hopefully we can contribute to that, and that’s why we’re trying to kind of partner with the public sector,” remarked Peter Battaglia, a research scientist at Google DeepMind, in a press discussion when addressing concerns about privatizing weather services.
Interestingly, Google’s announcement omits direct references to the climate crisis, which contrasts with the company’s previous communications regarding similar initiatives. The firm has previously stated, “As climate change drives more extreme weather events, accurate and trustworthy forecasts are more essential than ever,” in relation to its GenCast project.
