Google DeepMind launches Weather Lab, an AI-driven platform offering experimental tropical cyclone predictions. Partnering with the U.S. National Hurricane Center (NHC), the tool aims to enhance disaster preparedness by forecasting storm paths and intensity up to 15 days in advance. The model, leveraging stochastic neural networks, generates 50 possible cyclone scenarios, improving accuracy over traditional physics-based methods. Internal tests show its 5-day track predictions outperform ECMWF’s ENS model by 140 km on average—a 1.5-day leap in precision. NHC forecasters now integrate the AI’s outputs alongside existing models, hoping to refine warnings for communities facing cyclones like Honde, Garance, and Alfred.

Weather Lab’s real-time and historical data lets researchers compare AI predictions with ECMWF’s physics-based models. For instance, the experimental cyclone model accurately tracked Cyclone Alfred’s weakening trend seven days before landfall in Australia. NHC and Colorado State University’s CIRA validated the tool, noting its “comparable or greater skill” in track and intensity forecasts. The platform also provides a historical cyclone archive for external evaluations, fostering collaboration between AI developers and meteorological agencies.

Unlike physics-based models, which struggle to balance track and intensity predictions, Google’s single-system approach models both using reanalysis data and 45 years of cyclone observations. Preliminary results show it outperforms NOAA’s HAFS in intensity accuracy, a critical advancement for regional forecasting. By unifying these capabilities, Weather Lab addresses a longstanding challenge in meteorology: predicting both where a storm will strike and how powerful it will become.

Weather Lab is a research tool, not a replacement for official forecasts. Its live predictions remain experimental, and users are urged to consult local agencies for warnings. Nonetheless, the project marks a milestone in AI’s role in climate science, offering decision-makers new insights to mitigate $1.4 trillion in annual cyclone-related economic losses. Collaboration with global experts ensures the model’s evolution aligns with real-world needs.