The 2025 hurricane season marks the beginning of a “new era” in meteorology, with official agencies now using artificial intelligence (AI) models to determine the formation, path and intensity of storms, providing the insurance industry with more comprehensive and accurate near-term weather forecasts, according to Gallagher Re.
The reinsurer’s 2025 Natural Hazards and Climate Report notes that while experts warn that one season does not guarantee long-term success, the technology has shown great promise, suggesting that artificial intelligence will only become more important for global weather forecasting.
The U.S. National Hurricane Center (NHC) is a notable user of data-driven artificial intelligence models that have increased confidence in Atlantic hurricane track forecasts by multiple days.
However, the company notes that these models are less good at predicting other storm factors, such as intensity and precipitation.
Google’s DeepMind Tropical Cyclone (TC) model, the most visible AI model in use in 2025, debuted in June through Google Weather Labs and can generate ensemble forecasts up to 15 days.
During the 2025 season, the DeepMind TC model consistently outperformed traditional numerical weather prediction (NWP) models, such as the US GFS, in Atlantic track predictions.
Its performance has been mixed in other basins, especially the Western Pacific, where tracking technology lags behind physics-based systems such as Europe’s ECMWF. Analysts noted that intensity forecasts, especially those for rapid intensification (RI) cycles, also indicate lower technical levels.
“While one season is not enough to make a definitive pronouncement on the success or failure of any model, the performance of AI is promising enough for us to recognize that this technology will only become more entrenched in weather forecasting,” Gallagher Re analysts said.
Adding: “The main advantages of AI models are fast running speed, minimal computing power required, and stability for continuous operation. However, the quality of AI outputs depends entirely on the depth and quality of the historical and reanalysis data on which they are trained.”
Models need to be regularly retrained and recalibrated to maintain basic baseline accuracy, especially in the face of moderate to rapid atmospheric or oceanic changes, including those caused by climate change.
Even reliance on seemingly robust datasets carries risks because many remain limited, especially when applied across a watershed. This limitation may lead to output gaps, particularly the under-representation of rare (tail) events, the report said.
Gallagher Re analysts said that despite the attention on AI forecast models, “scientists and other public and private sector users will continue to rely heavily on traditional numerical weather prediction and tropical cyclone models”.
The integration of AI models is expected to further educate disaster modelers and risk managers, enhance their ability to grasp the range of potential outcomes of an event, and ultimately improve the timeliness and accuracy of real-time communications.
The analysts concluded: “While AI models are not expected to completely replace physics-based models, we are still seeing record investment in this area.
“This includes the Artificial Intelligence/Integrated Forecasting System (AIFS) run by ECMWF and NOAA’s EAGLE project. Private companies such as Microsoft, IBM and NVIDIA are also investing heavily in AI weather forecasting.”
Adding: “More than forecast, the rapid growth of AI and its data center footprint raises broader energy/power concerns. How the world invests in green energy to meet this demand will be critical in determining whether greenhouse gas emissions can be reduced.”