Earlier this year's hurricane season, the US National Hurricane Center (NHC) announced a partnership with Google DeepMind, Google's AI research department, to test the latest AI weather model, which is included in the hurricane forecast process.
"The partnership between NOAA and Google helps the NHC quickly assess new typhoon forecasting technologies as soon as they appear," said NHC Director Michael Brennan.
According to the US National Oceanic and Atmospheric Administration (NOAA), after being added to the forecast source, DeepMind's model has outperformed some traditional models that are bulky and time-consuming.
A typical example is the super typhoon Melissa forecast.
Both DeepMind's model and Europe's AI model provide NHC experts with unusual reliability: Melissa will rapidly intensify, becoming a Category 5 super typhoon before causing heavy damage to Jamaica.
I am really impressed with the ability to handle the phenomenon of sudden increase in level, the weakness of many models. What this model shows in Shelter is clearly key in issuing extremely high risk warnings, said Matt Lanza, manager of The Eyewall.
NOAA said DeepMind's model is the most accurate model of the reporting path and intensity, behind only the NHC's official forecast.
Traditional forecast models, such as the European model or the US GFS model, rely on complex physical and atmospheric equations, requiring huge amounts of calculation and many hours of machine operation.
Meanwhile, the AI model like DeepMind is faster and consumes less resources. These models predict by analyzing decades of weather data, identifying patterns from previous storms, and making forecasts in just a few seconds.
"AI/technology uses information from more than 40 years of atmospheric data, with a 6-hour cycle, to understand how the atmosphere changes over time," said Professor Ryan torn, Albany University, USA, weather model expert.
He further explained that once learning how the atmosphere moves, the AI model only needs the current state of the atmosphere to make a forecast.
In the case of DeepMind, Google said that this model can create hundreds of different weather scenarios from a starting point in just a few minutes, while the traditional model can take hours.
Despite the benefits of the 2025 typhoon season, experts say more time to verify before the AI model can completely replace the traditional model.
Mr. Lanza noted that DeepMind has not yet proven its ability to forecast storms in the Gulf of Mexico, especially as this year's storm season is unusually quiet.
"In terms of extreme weather and climate change, you have to take into account unpredictable scenarios, and AI models can ignore those risks. Thats why physics-based models are so important, he said.
