WeatherNext Predicted Hurricane Melissa's Rapid Intensification Five Days Before Jamaica Landfall

Google DeepMind’s experimental model WeatherNext forecast, roughly five days before landfall in October, that Hurricane Melissa would rapidly intensify from Category 1 to Category 5 and strike Jamaica with about 80% confidence. That early, high‑confidence rapid‑intensification signal arrived while conventional forecast models were still undecided, and Google provided the scenario outputs to the U.S. National Hurricane Center (NHC), where the AI guidance was incorporated into the forecasting process as the storm approached.

WeatherNext generated probabilistic scenario ensembles by running dozens of simulations; during Melissa it produced sets of 50 possible futures every six hours. DeepMind and the NHC plan to increase both cadence and ensemble size to 1,000 possible futures every six hours for the next season, a move NHC senior specialist Philippe Papin said “should provide more stable and consistent guidance,” which could improve decision support for emergency managers and the public.

The model was reported to be more accurate than other models the NHC used during Melissa and explicitly issued the early rapid‑intensification alert. Officials and researchers promoting AI‑based forecasting highlight two practical benefits: earlier, more consistent guidance during fast‑changing storms and a scalable mechanism to donate compute and modeling capacity to public agencies that lack those resources.

WeatherNext’s work on Melissa sits amid a broader push by technology firms and startups to apply machine learning to weather prediction. Large companies including Microsoft, NVIDIA, and Huawei and startups such as Atmo, Tomorrow.io, and WindBorne are developing ML approaches and new observation chains. Several teams are also deploying lower‑cost satellites and redesigned weather balloons to improve input data, and AI systems are being touted as faster and cheaper to run than many full‑physics models while showing higher short‑term skill in extreme events in multiple cases.

Local authorities said the early forecast materially affected preparations. Evan Thompson, principal director of the Meteorological Service of Jamaica, called the early forecast “critical” and said consistent messaging helped prompt public preparations; a Category 5 had never previously made landfall on Jamaica. When Melissa struck, peak winds exceeded 131 mph; roofs were torn off more than 120,000 buildings, tens of thousands of homes were destroyed leaving many homeless, and 45 people were killed. Authorities stated that the early AI‑informed warnings likely reduced additional casualties.

The NHC confirmed it will continue working with Google as the June 1 hurricane season begins, aiming to expand the model’s ensemble capability and cadence for future storms. The planned scaling is intended to deliver steadier probabilistic guidance to forecasters and emergency managers facing rapidly intensifying tropical cyclones.