Google releases AI-based weather model aimed at improving hurricane forecasts
Baku, June 14, AZERTAC
Google launched a new weather model Thursday that uses artificial intelligence (AI) in an effort to increase the accuracy of hurricane forecasts, including those developed by the National Hurricane Center, according to foxweather.com.
The model is now available for free on the new interactive Google website Weather Lab, which provides the model’s predictions on the formation, track, size, shape and intensity of hurricanes, officials said.
Also known as tropical cyclones, hurricanes are notoriously difficult to predict due to their sensitivity to the slightest changes in the atmosphere.
However, the tropical cyclone model developed by Google DeepMind, the company’s AI lab, and Google Research, a team that manages a diverse research portfolio, is able to produce 50 possible scenarios for the storms and do so with a lead time of up to 15 days.
This advance notice is significant as current models provide predictions with a lead time of up to 3-5 days.
The new tropical cyclone model from Google also resolves a challenge traditional models face, in that no one model can predict both track and intensity due to the different types of information needed to predict them, according to officials.
On the one hand, predicting the track requires taking into account the vast atmospheric steering currents. On the other hand, predicting the intensity of a tropical cyclone involves focusing on the storm’s compact core. The new tropical cyclone model is able to process both types of information.
Additionally, the AI-based model uses historical data to predict future scenarios. Google said the dataset uses millions of observations to reconstruct past weather over the entire planet, in addition to a specialized database containing key information about the track, intensity, size and wind radii of nearly 5,000 observed cyclones since 1980.
"Modeling the analysis data and cyclone data together greatly improves cyclone prediction capabilities," Google said in a statement.
For example, when researchers used NHC hurricane data from 2023 and 2024 in the new AI-based model, they found that its five-day cyclone track prediction was, on average, more than 85 miles closer to the true cyclone location compared to the prediction from the leading global physics-based ensemble model from the European Centre for Medium-Range Weather Forecasts (ECMWF).
When comparing the AI-based model’s ability to predict cyclone intensity, which has been challenging for traditional physics-based weather models, researchers found that it outperformed the average intensity error of NOAA’s Hurricane Analysis and Forecast System (HAFS), a leading regional, high-resolution physics-based model.
Such accuracy with the new tropical cyclone model has led Google to partner with the NHC, whose expert human forecasters will be able to see predictions from the AI-based model in real time, said DeepMind Research Scientist Ferran Alet. Information from the model will enhance the NHC forecasts by complementing the model prediction with other models, as well as the human expertise of those at the NHC.
As the NHC will be able to use the tropical cyclone model, so will the general public through Weather Lab. There, users will be able to see two other AI weather models in addition to the tropical cyclone model, WeatherNext Graph and WeatherNext Gen – both of which help shape the tropical cyclone model.
WeatherNext Graph, released in 2021, can produce a single weather forecast per time and location with greater accuracy and speed than the industry gold-standard weather simulation system known as the High Resolution Forecast (HRES) from the ECMWF, according to Google.
While WeatherNext Graph hones in on a specific time and location, WeatherNext Gen produces a range of up to 50 likely future weather scenarios to account for weather uncertainties and risks of extreme conditions, Google said. Released earlier this year, WeatherNext Gen, the company said, creates better forecasts of both day-to-day weather and extreme events than the top operational system, ECMWF’s ENS, up to 15 days in advance.
Recognizing the capabilities of WeatherNext Graph and WeatherNext Gen in forecasting a variety of weather events, the Google researchers saw their potential in forecasting tropical cyclones, specifically.
"We decided to focus more on tropical cyclones for two reasons," Alet said. "The first one is that they have a massive impact on communities. Storm surge or strong winds destroy lots of properties and cause massive damage and often hundreds of deaths. And the second thing is that they're very hard to predict because complex interactions make their path and intensity chaotic. So we think AI can provide a solution here."
FOX Weather Hurricane Specialist Bryan Norcross served as an advisor to the Google team developing the new AI-based tropical cyclone model.
"It's the first hurricane season where we're going to really see what AI can do by this development of a tropical storm- and hurricane-specific AI model," Norcross said. "That's what's new, and it's really pretty exciting that we're going to be able to look at this in parallel with all the things we normally look at for the hurricane season."
Hurricane season in the Atlantic basin began on June 1. Leading hurricane researchers have maintained their forecast for an above-average 2025 Atlantic hurricane season.
Other news in this section
