Google DeepMind helped forecasters give earlier Hurricane Melissa warnings (theconversation.com)

Google DeepMind’s machine‑learning guidance helped forecasters issue earlier, more accurate warnings for Hurricane Melissa, which rapidly intensified into a Category 5 and struck western Jamaica with 185 mph winds on Oct. 28, 2025. By outperforming many traditional physics‑based models this season, DeepMind contributed to correctly predicting Melissa’s abrupt northeast turn — a forecast that gave residents across Jamaica, Cuba and the eastern Bahamas extra time to evacuate. The outcome highlights how AI can materially improve life‑saving lead time during fast‑evolving storms amid a season in which the subtropical ridge (the Bermuda/Azores high) and an active jet stream steered many powerful systems out to sea. Technically, DeepMind’s approach infers storm tracks from statistical patterns in historical cyclone behavior rather than solving full atmospheric equations, and it showed tangible gains when combined with improved in‑storm observations. Meteorologists note that Melissa’s rapid intensification (roughly 70 → 140 mph in 24 hours) resulted from warm Caribbean sea surface temperatures and better vertical alignment, while subtle changes in steering currents — influenced by highs, troughs and jet‑stream ripples — made track forecasting difficult. Integrating NOAA Hurricane Hunter reconnaissance and targeted drone flight data with ML models promises further improvements, though challenges remain in reliably predicting steering‑current shifts and intensity changes.