Google's AI Model Predicts Tropical Cyclone Tracks with Up to One-Day Lead Time Advantage
Google's AI model for forecasting tropical cyclones has been found to have a significant advantage over other models in terms of accuracy and lead time.
A study published in Nature found that WeatherNext Cyclones, Google's probabilistic global AI weather model, was able to predict the tracks, intensity, and wind structure of tropical cyclones with an average lead-time advantage of at least one day compared to other models.
The researchers trained the AI model on nearly 20 terabytes of global atmospheric data and evaluated its performance on tropical cyclones from 2023 to 2025. The study found that WeatherNext Cyclones recorded an average track error of 230 km at five days out, compared to 370 km for the European Centre for Medium-Range Weather Forecasts' ENS and 335 km for Google's GenCast.
The researchers also found that the AI model's probabilistic intensity forecasts reduced the Continuous Ranked Probability Score by more than 50% at many lead times when compared with other models. The study suggests that high-resolution data is not a strict requirement for accurate intensity forecasting, although further research is needed to understand how the model extracts this information from coarser data.