Google DeepMind and Google Research have partnered with the National Hurricane Center to deploy an experimental AI-powered cyclone prediction model that forecasts storm path, size, and intensity at unprecedented speeds. The collaboration, announced August 4, includes sharing forecasts with the NHC and launching Weather Lab, a new data and visualisation platform hosting real-time and historical predictions.
The AI model differs from traditional forecasting approaches that simulate atmospheric physics on supercomputers, instead using probabilistic modelling to produce 50 possible storm outcomes through random perturbations during prediction. According to preliminary internal evaluations, the experimental model demonstrates state-of-the-art accuracy for cyclone track, intensity, and size predictions.
Google DeepMind researchers Ferran Alet and Tom Andersson presented the model to NHC specialists in March at the Miami headquarters. A leading hurricane specialist who had previously suggested cyclone path prediction might have reached its limits told the team the model's performance was "potentially revolutionary" following their presentation.
The model addresses limitations in previous Google weather models like GenCast, GraphCast, and NeuralGCM, which were designed for general weather using low-resolution historical data and provided poor intensity predictions. The new system trains on both general weather and sparse cyclone-specific data to handle the extreme conditions and chaotic nature of cyclones.
Trusted testers have accessed the model's forecasts for approximately two months, providing feedback on accuracy and interface usability. Google Research product manager Olivia Graham worked directly with NHC forecasters to develop an "expert mode" that explores potential cyclones before formation, showing small circles representing approximately 2% cyclone formation probability with potential track and strength predictions.
The platform displays track and intensity predictions for current storms on a global map alongside other Google weather models and official forecasting models for comparison. The collaboration focuses on co-developing technology with daily NHC users to ensure relevant information delivery for life-saving operations.
The partnership demonstrates enterprise-scale AI deployment in critical government operations requiring 24/7 availability during extreme weather events. Organisations in weather-dependent industries could benefit from improved forecasting accuracy and speed for operational planning and risk management during cyclone seasons.
This collaboration represents significant advancement in AI-powered weather prediction for enterprise and government applications. Organisations requiring accurate storm forecasting for supply chain management, emergency preparedness, and operational continuity can leverage improved prediction capabilities. The real-time data platform approach may establish new standards for AI-assisted decision-making in critical infrastructure and public safety applications where traditional physics-based modelling has reached practical limitations.