Subscribe
ResearchOct 11, 20269 sources

As Category 4 Hurricane Simon skirts Mexico, forecasters weigh Google DeepMind's AI model alongside physics-based guidance

Hurricane Simon, the only tropical cyclone active in the Atlantic, East Pacific and Central Pacific on October 11, is brushing west-central Mexico. In its latest discussion the National Hurricane Center cited Google DeepMind's AI model alongside the European ensemble as supporting its track forecast.

GOES-19 satellite image of Hurricane Simon off the coast of west-central Mexico on October 11, 2026

Image: NOAA/NESDIS/STAR GOES-19 GeoColor, Oct 11, 2026, 23:00 UTC (public domain)

Hurricane Simon was a Category 4 storm with maximum sustained winds of 130 mph on Sunday afternoon, centered about 105 miles west-northwest of Manzanillo, Mexico, and moving north-northwest at 6 mph, according to the National Hurricane Center's 3 p.m. CST advisory. A hurricane warning was in effect from Punta San Telmo to El Roblito and for Las Islas Marías, and Mexico issued a new hurricane watch farther up the coast (NHC advisory).

Simon is the only active tropical cyclone in the basins the NHC and the Central Pacific Hurricane Center cover. Both centers' outlooks said there were no tropical cyclones in the Atlantic or Central North Pacific at the time (NHC; CPHC).

AI guidance in the forecast

In its forecast discussion, the NHC said some outlier models, such as the American GFS, turn Simon southwestward after 48 hours, but that "the majority of the ensemble guidance (European-EPS, Google DeepMind) do show this slow recurving solution back into Mexico." The center forecast Simon to move inland and dissipate in the day 3–4 period and called the forecast "low confidence" (NHC discussion).

Google DeepMind began publishing live cyclone forecasts on its Weather Lab site in June 2025 and partnered with the NHC to support its forecasts. The company said internal testing on 2023 and 2024 storms showed its model's 5-day track predictions were on average 140 km closer to the true position than ECMWF's physics-based ensemble (Google DeepMind).

How it performed last year

The NHC's verification report for the 2025 Atlantic season said official forecasts generally outperformed most models, but that the Google DeepMind ensemble mean, labeled GDMI, "slightly outperformed NHC at the shorter lead times from 12 to 72 h." The report noted GDMI was not available for the first two storms of the season. It also said intensity forecasts were harder in 2025, with errors above the five-year average during many rapid-intensification events (NHC 2025 Verification Report).

An earlier NHC preview said 2025 was the first season the center used AI-based models in real-time operations and called the DeepMind model "very useful," while cautioning that some AI systems "were not consistently available on a timely basis" (NHC preview).

Europe's AI models

The European Centre for Medium-Range Weather Forecasts runs its own AI system, the AIFS. Its deterministic version has been operational since February 25, 2025, and its ensemble since July 1, 2025. Both were upgraded to version 2 on May 12, 2026, when the ensemble began producing tropical cyclone tracks out to 15 days from 52 members (ECMWF).

ECMWF has also described a prototype, built with the University of Cambridge, that uses a second machine-learning system to correct AIFS's storm intensity errors. ECMWF says the next steps are to run it in real time and evaluate it with forecasters (ECMWF blog).

Sources (9)

  1. NHC — Hurricane Simon Public Advisory 18, Oct 11, 2026
  2. NHC — Hurricane Simon Forecast Discussion 18, Oct 11, 2026
  3. NHC — Atlantic Tropical Weather Outlook
  4. CPHC — Central North Pacific Tropical Weather Outlook
  5. Google DeepMind — How we're supporting better tropical cyclone prediction with AI
  6. NHC — 2025 Forecast Verification Report
  7. NHC — 2025 Verification Report Preview
  8. ECMWF — AIFS Machine Learning data
  9. ECMWF — Improving tropical cyclone intensity forecasts with a simple AI correction
← Back to AI World News