NHC leans on Google DeepMind's ensemble and calls Isaias's landfall

Peter H. Vartanian

NHC leans on Google DeepMind’s ensemble and calls Isaias’s landfall

The NHC leaned its Hurricane Isaias forecast on Google DeepMind’s ensemble, which called a Category 2 Florida Panhandle landfall from the first advisory while physics-based models pointed west.

2 for an organization

The NHC’s official forecast leaned on the ensemble, and the track it gave, a landfall on the Florida Panhandle, verified while physics-based models pointed to Louisiana.
Not rated for society

No source documents that the forecast changed what people in the storm’s path did.

Depth runs from 1, a small gain, to 5, survival.

What happened

Hurricane Isaias formed in the southwestern Gulf of Mexico in October 2026 and made landfall near Destin, Florida, as a Category 2 storm on 9 October. The National Hurricane Center’s forecast discussions record its reliance on the Google DeepMind ensemble, which it designates GDMI. On the morning of 7 October the forecasters nudged their track eastward, though not as far as the DeepMind ensemble mean, which stood with the AI models on the right side of the guidance. That evening they wrote that the forecast leaned more heavily on the corrected consensus and the DeepMind aids, which were performing better, on the faster and right side of the guidance. On 8 October they raised the forecast peak to major hurricane strength because a majority of DeepMind ensemble members showed it. CNN reported on 8 October that the NHC had shaped its predictions to match the DeepMind model rather than the European and GFS models and had forecast a Category 2 storm from the first advisory. NPR reported on 26 October that physics-based models had initially sent the storm toward southeastern Louisiana, that the DeepMind model had called the Florida Panhandle from the start, and that it beat the physics-based models on the path. The NHC’s director told NPR that the model did less well than the NHC’s blended forecast on intensity. Michael Lowry, a hurricane expert who has advised the NHC, called it the best performing model that forecasters have seen.

Status

Realized benefit.

Evidence

foundstudiedreported, the grade of this benefitclaimed

Two independent news reports support the benefit.

Size

No magnitude yet. The distribution is not documented.

Compared with

The physics-based models, the European and GFS. Measured.

The system’s part

The NHC’s own discussions say that its forecast leaned on the DeepMind aids for track and raised its intensity forecast on the ensemble’s members. Human forecasters blended the guidance and issued the forecast, and the NHC’s director says that the model did less well than the blend on intensity.

Limits

The model led on the storm’s track but trailed the NHC’s blended forecast on intensity, and the comparison rests on one storm. No source documents what people in the storm’s path did with the forecast.

Trade-offs

No costs or harms are reported in the sources.

Learns from data

Shown. The developers’ Nature paper says that the cyclone model is trained on ECMWF analyses and historical cyclone tracks.

“It is jointly trained on a combination of two data sources (a process we refer to as co-training): decades of global atmospheric analysis data from the ECMWF” Source

Later outcome

Persisted, 26 October 2026. NPR reported on 26 October 2026 that the model had beaten the physics-based models on the storm’s path.

Download data Cite Suggest a correction

System: Google DeepMind tropical cyclone model, ensemble mean (NHC designation GDMI)
Developer: Google DeepMind
Deployer: NOAA National Hurricane Center
Where: Northern Gulf Coast, United States, with landfall near Destin, Florida
Stage: In routine use
Date: 9 October 2026
Kind: Benefit created
Broad benefit: Positive environmental impact
I want to: map where AI was used

Sources

Report, by the organization that used it.

National Hurricane Center, Hurricane Isaias Discussion Number 6, 10 PM CDT, 7 October 2026.

The NHC forecast continues to lean more heavily on the better-performing corrected consensus (HCCA) and Google DeepMind (GDMI) aids, remaining on the faster and right side of the guidance envelope.
Report, by the organization that used it.

National Hurricane Center, Tropical Storm Isaias Discussion Number 4, 10 AM CDT, 7 October 2026.

The new NHC forecast is nudged a bit eastward, but isn't as far east as the latest HFIP Corrected Consensus Approach (HCCA) or Google DeepMind ensemble mean (GDMI).
Report, by the organization that used it.

National Hurricane Center, Hurricane Isaias Discussion Number 10, 10 PM CDT, 8 October 2026.

A majority of Google DeepMind ensemble members show Isaias briefly reaching major hurricane status early Friday (along with some of the hurricane regional models), and this is now reflected in the NHC forecast.
News report, by authors with no tie to the system.

Meszaros, J., How AI is transforming hurricane forecasts, WUSF and NPR, 26 October 2026.

In early October, Hurricane Isaias made landfall near Destin, Fla., as a Category 2 storm. Conventional forecasting models initially predicted it would head further west, toward southeastern Louisiana. But Google DeepMind's artificial intelligence model called it right out of the gate: the hurricane was going to hit the western Florida Panhandle.
An AI model from Google Deepmind beat out traditional physics-based models at predicting the path of the storm, though it didn't do as well in forecasting the storm's strength.
While it topped other models in tracking the path of Isaias, it didn't do as well as the NHC's blended forecast in determining the intensity, or strength, of the storm.
News report, by authors with no tie to the system.

Freedman, A., Why the National Hurricane Center bet big on AI to forecast Isaias, CNN, 8 October 2026 (as carried by KSL).

They shaped their predictions to closely match an AI hurricane model developed by Google DeepMind, rather than leaning on the classic physics-based models many Americans are more familiar with, such as the European and GFS.
Relying heavily on DeepMind meant NHC forecast Isaias would become a Category 2 from the first advisory on Tuesday, when it didn't even have a name yet.

Details

AI system

Workflow: A cyclone ensemble trained on decades of atmospheric analyses and historical tracks, run for each forecast cycle. Human forecasters who blend the AI guidance with physics-based models and corrected consensus aids.
Human oversight: NHC forecasters blended the DeepMind aids with other guidance and issued the forecast themselves, leaning on the ensemble for track and raising the peak intensity when a majority of its members showed it.

Follow-up

Conditions for replication: Forecasters must have the ensemble’s output in time for each advisory. The ensemble’s track skill must hold for the storm at hand, since its intensity guidance trailed the blend.

The entry

Sector: Weather forecasting / tropical cyclones.
Purpose: Google DeepMind’s cyclone ensemble forecasts the track and intensity of tropical cyclones for the NHC’s use alongside physics-based models.

Cite this benefit

Peter H. Vartanian. (2026). NHC leans on Google DeepMind's ensemble and calls Isaias's landfall (AIBD-1238). Artificial Intelligence Benefits Database. https://w3id.org/aibd/AIBD-1238 (archived at https://doi.org/10.57967/hf/11470)

BibTeX Data What the fields mean