Particle.news

Download on the App Store

DeepMind Weather Lab Tops Short-Range Hurricane Erin Forecasts

An independent review found the model led short-range Erin forecasts, with operational trust now hinging on three- to five-day performance.

Overview

  • Post-storm charts compiled by former NHC hurricane specialist unit chief James Franklin show Google's GDMI had the lowest track and intensity errors through 72 hours, with standout 48-hour intensity accuracy.
  • The AI outperformed the NHC official forecast and the consensus TVCN/IVCN models in that window and bested leading American and European physics-based guidance.
  • Franklin told CBS Miami the model captured Erin’s intensity and life-cycle shape with very little error, while noting it remains developmental and not for public use.
  • Google launched the cyclone-focused model in June, trained on global reanalysis and nearly 5,000 historical storms, and is evaluating it in partnership with the National Hurricane Center.
  • With few real-time tests before Erin, experts emphasize the need to demonstrate consistent gains at three to five days, the timeframe critical for evacuation and preparedness decisions.