Deepmind's weather AI solves cyclone intensity track tradeoff
google deepmind's wnext cyclones model predicts both path and intensity of storms more accurately than top specialized models, extending forecast range by a day and running 8x faster via a new generative technique.
google deepmind introduced wnext cyclones, a model that resolves a decades-old tradeoff by accurately forecasting both cyclone tracks and intensities using data resolution 100 times coarser than specialized systems. built with the uk met office and noaa, wn-c reduced track error by 140km compared to ecmwf ensembles and beat regional models on intensity metrics while requiring only a single neural network pass thanks to functional generative networks. the system went live in june 2025 and successfully predicted rapid intensification for hurricane melissa, suggesting that coarse atmospheric data contains more signal than previously understood.