In October 2025, Hurricane Melissa made landfall in Jamaica as the most powerful cyclone ever recorded on the island — and one of the strongest in Atlantic history. What would have seemed unlikely a few years ago: an AI model accurately predicting this trajectory five days before impact.
What WeatherNext did
WeatherNext, Google DeepMind's AI weather forecasting system, predicted Melissa's Category 5 landfall on Jamaica's southern shore with 80% confidence five days in advance. Two days later, confidence was near 100%. According to DeepMind, it was the first time a model successfully predicted such rapid intensification from such a low initial wind speed.
The National Hurricane Center confirmed in its annual verification report that WeatherNext was the top-performing individual model of the 2025 season for both track and intensity.
Operational consequences
The early prediction allowed the NHC to give the Meteorological Service Jamaica unprecedented lead time. Local authorities were able to coordinate evacuations and mobilize resources with a margin that traditional models could not guarantee.
The difference is concrete: going from 48 to 120 hours of reliable lead time is not a matter of better dashboards — it is a matter of lives that can or cannot be saved.
Why the DeepMind blog came out at I/O 2026
Google chose I/O 2026 to publish the full case study, a few months after the event, as a demonstration of WeatherNext's real-world capabilities under operational conditions. This is not speculative marketing: it is an independent assessment by the NHC, already documented in NOAA's official report on Hurricane Melissa.
For anyone working with climate data or managing environmental risk, WeatherNext now sets the public benchmark for AI atmospheric forecasting models.