GDMI gave the best intensity forecasts for category 5 Hurricane Erin

NHC found that Google DeepMind's ensemble mean had the lowest mean intensity errors of the guidance for Hurricane Erin from 24 to 120 hours. NHC's own errors ran above average.

  • 2 an organization

    GDMI's mean intensity errors were the lowest of the guidance from 24 to 120 hours, 9.7 knots at 48 hours against 13.7 knots for the official forecast.

  • Not rated society

    NHC does not report that the AI guidance changed warnings or outcomes along the coasts that Erin affected.

Depth runs from 1, a convenience or time saved, to 5, a life or a livelihood preserved.

Status
Reported benefit.
Evidence

foundstudiedreportedclaimed

An independent news report supports the benefit.

Compared with

NHC official intensity forecasts (OFCL) for Erin in the homogeneous comparison. Measured.

Without the system
OFCL errors of 11.9, 13.2, 13.7, 14.5, 15.5, 15.7 and 14.2 kt at 24, 36, 48, 60, 72, 96 and 120 h
With it
GDMI errors of 10.4, 10.6, 9.7, 12.0, 12.5, 13.9 and 13.6 kt
The system’s part
NHC verified GDMI's own intensity and track forecasts against the best track alongside the official forecast and other models, which isolates the model's skill. NHC gives no account of how forecasters used GDMI for Erin. Because NHC uses the model and helped Google DeepMind develop it, NHC is not fully outside the case.
Limits
Not yet checked.
Costs and harms
Not yet checked.
Later
Not yet checked.

Download data Cite Suggest a correction

Benefit type
Improved Task Accuracy
System
Google DeepMind tropical cyclone model, ensemble mean (NHC designation GDMI)
Made by
Google DeepMind
Used by
NOAA National Hurricane Center (experimental guidance), with output supplied by Google DeepMind
Where
Western North Atlantic, with warnings for the Leeward Islands, the Bahamas, Bermuda and the coasts of North Carolina and Virginia, United States
Stage
In a pilot
Date
Kind
Aggregate result
Broad benefit
Estimating the Quantifiable Characteristics of Products, Events, or Information
Collections
Cross-border records
Last revised
31 October 2026

Sources

  1. Report, by the organization that used it.

    Pasch, R. J., and A. D. Mahoney. Hurricane Erin (AL052025), 11–22 August 2025. National Hurricane Center Tropical Cyclone Report, 30 January 2026 (original report dated 16 January 2026), p. 6 and Table 6b.

    A homogeneous comparison of the official intensity errors with selected guidance models is given in Table 6b. For the 24-h through 120-h forecast intervals, the GDMI had the lowest mean errors.

  2. News report, by authors with no tie to the system.

    Michael Lowry, 'New AI Model Shines During Hurricane Erin', Eye on the Tropics (newsletter), 26 August 2025. Lowry publishes verification charts for Erin prepared by James Franklin, former branch chief at the National Hurricane Center.

    Google's DeepMind (GDMI - red line) beat all other models for Erin's intensity through 3 days (72 hours) and was only bested by NOAA's next-generation hurricane model (the Hurricane Analysis and Forecast System or HAFS model) beyond 3 days out.

  3. Press release, by the organization that used it. Context or contrary evidence.

    National Weather Service. AI in Hurricane Forecasting at the National Hurricane Center: A Q&A with NHC's Science Operations Officer. NOAA/NWS, undated.

    NHC has partnered directly with Google DeepMind to develop a new AI hurricane forecast model that was used experimentally during the 2025 season.

Details

AI system

Workflow

NHC included GDMI among the guidance models in its homogeneous comparisons for Erin. NHC measured each model's errors against Erin's best track and intensity.

Follow-up

Conditions for replication

The comparison covers the forecasts for a single long-lived storm. The official forecast remains NHC's own judgment drawn from many models.

Entry details

Sector

Weather forecasting / tropical cyclones.

Purpose

Google DeepMind built its experimental cyclone model to give forecasters ensemble guidance on where storms will go and how strong they will become.

Not yet checked: model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.

Cite this record

Peter H. Vartanian. (2026). GDMI gave the best intensity forecasts for category 5 Hurricane Erin (AIBD-0751). Artificial Intelligence Benefits Database, edition of 31 October 2026. https://petervartanian.xyz/aibd/records/AIBD-0751.html (archived at https://doi.org/10.7910/DVN/AADSII, version 4.0)

BibTeX Data Field definitions