AI system
- System
-
GraphCast (Google DeepMind graph neural network weather model), experimental live run on ECMWF public charts
- Model
-
GraphCast.
- Release
-
Exact release not established.
- Application developer
-
DeepMind - Role
-
GraphCast generated weather forecasts, including the projected path of Hurricane Lee.
- Configuration
-
Researchers used the live GraphCast forecasts available through ECMWF.
- Inputs and outputs
-
Not assessed
- Autonomy
-
Not assessed
- Workflow
-
GraphCast trained on decades of ERA5 reanalysis and run live from ECMWF initial conditions on ECMWF's experimental charts. A long-lived Cabo Verde hurricane whose track was, per NHC, easier than normal to forecast (low climatology-persistence errors). Vendor retrospective comparison of its own run against conventional models.
- Human oversight
-
Not assessed
- Relation to AI
-
The only sources for the nine-day Lee forecast are the maker's own blog and a news report quoting the maker's scientist. ECMWF's editorial confirms only that GraphCast ran experimentally on its charts. NHC's Lee report records exceptionally low official track errors without mentioning AI guidance. The AI output is therefore documented as a forecast, not as an input to any decision.
- AI contribution
-
The only sources for the nine-day Lee forecast are the maker's own blog and a news report quoting the maker's scientist. ECMWF's editorial confirms only that GraphCast ran experimentally on its charts. NHC's Lee report records exceptionally low official track errors without mentioning AI guidance. The AI output is therefore documented as a forecast, not as an input to any decision.
Benefits
- Outcome identifiers
-
Not assessed
- Benefit
-
forecast Lee's Nova Scotia landfall about nine days ahead
- Category
-
Improved Task Accuracy
- Status
-
Not assessed
- Beneficiary groups
-
People. Organizations. Society.
- Beneficiaries
-
Not assessed
- Reach
-
Not assessed
- Distribution and access
-
Not assessed
- Duration
-
Not assessed
- Expected?
-
Yes
- Costs and harms
-
Not assessed
Measurement
- Studies
-
Not assessed
- Study design
-
Not assessed
- Measurement
-
Not assessed
- Setting and sample
-
Not assessed
- Measures
-
forecast Lee's Nova Scotia landfall about nine days ahead
- Comparison
-
Assumed alternative: traditional physics-based forecasts of Hurricane Lee's landfall (DeepMind's own account)
- Results
-
Assumed comparison. Reference: only locked in on Nova Scotia about six days in advance. Result: about nine days in advance.
- Additional value from AI
-
Not assessed
- Limitations
-
Not assessed
- Denominators
-
Not assessed
- Materials for replication
-
Not assessed
Money
- Kind of value
-
Not assessed
- Amount
-
Not assessed
- Currency and prices
-
Not assessed
- Period and unit
-
Not assessed
- Realization
-
Not assessed
- Recipients and payers
-
Not assessed
- Gross or net
-
Not assessed
- Included costs
-
Not assessed
- Comparison
-
Not assessed
- Method
-
Not assessed
- Return and payback
-
Not assessed
People and places
- Participants
-
Developer:
DeepMind. Operator: 
ECMWF (experimental data-driven forecasts on its public charts pages). - Jurisdictions
-
Benefit: Long Island, southwestern Nova Scotia,
Canada (landfall). Developer: 
United Kingdom. Operator: United Kingdom. - Ownership and control
-
Not assessed
- Dependencies
-
Not assessed
- Cross-border connections
-
Not assessed
- Geopolitical implications
-
Not assessed
Follow-up
- Conditions for replication
-
Independent verification of the specific nine-day run against the official track record (not found in any ECMWF or NHC document among the sources). Access to ECMWF initial conditions for the live run. A forecaster or authority actually consulting the AI run for it to become an attributable benefit.
- Replication attempts
-
1 follow-up entries. Benefit recurred: 1.
- Later findings
-
7 November 2024. An independent Uppsala University evaluation published 7 November 2024 confirmed GraphCast's general advantage over ECMWF's IFS HRES, including on near-surface temperature and wind extremes; it does not examine the Hurricane Lee forecast, and the NHC's March 2024 report on Lee names no AI model.
- Availability
-
Not assessed
- Institutional decision
-
Not assessed
- Grounds
-
Not assessed
- Action and result
-
Not assessed
- Success criterion
-
Not assessed
Evidence
- Source numbers
-
AIBD-0341-source-0001–0008
- Citations
-
Numbered references below.
- Language
-
Not assessed
- Source relationships
-
Developer, Independent, Operator
- Original evidence
-
Not assessed
- Access and versions
-
Not assessed
- Supported findings
-
Not assessed
- Evidence links
-
Not assessed
- Evidence strength
-
Not assessed
- Missing evidence
-
Not assessed
Grades
- Assessed benefits
-
People. Organizations. Society.
- Depth
-
Not assessed
- Reasons
-
People: No individual outcome is documented. Organizations: General benefit. No cited source documents a forecast office using the GraphCast Lee run in its products. Society: General benefit. No public-safety decision is linked to the run in any cited source.
- Observed or intended
-
Not assessed
- Rubric
-
Not assessed
- Assessor and date
-
Not assessed
- Alternative assessments
-
Not assessed
- Fill
-
Independently studied
Entry details
- Kind
-
Not assessed
- Related entries
-
Not assessed
- Field
-
Weather forecasting / tropical cyclones. Public-sector meteorological service; AI research laboratory.
- Purpose
-
Machine-learning global medium-range weather forecasting (10 days, 0.25 degree), including severe-event prediction such as tropical cyclone tracks. This forecast was that purpose.
- Dates
-
Date basis unassessed: 2023 (precision unassessed). Updated: 22 October 2026.
- Stage
-
Not assessed
- Version
-
SHA-256: 9911fcda891b…
References
- Other identifiers
-
Not assessed
- Category mappings
-
Not assessed
- Brief
-
Not assessed
- Revisions
-
9 revisions. Latest: 22 October 2026.
Sources
-
Peer-reviewed study Independent author
Olivetti L, Messori G (2024). Do data-driven models beat numerical models in forecasting weather extremes? A comparison of IFS HRES, Pangu-Weather, and GraphCast. Geoscientific Model Development 17:7915-7962, doi:10.5194/gmd-17-7915-2024.
-
Report Independent author
Blake, E. and Nepaul, H. (2024). National Hurricane Center Tropical Cyclone Report: Hurricane Lee (AL132023), 5-16 September 2023. NOAA/NWS, 21 Mar 2024.
making landfall on Long Island, in southwestern Nova Scotia, later that day around 2000 UTC with maximum sustained winds of about 55 kt well east of the center
Official track forecast errors were exceptionally low through 72 h, and below the previous 5-year period means through 120 h.
-
Preprint Developer
Lam, R., Sanchez-Gonzalez, A., Willson, M., et al. (2022/2023). GraphCast: Learning skillful medium-range global weather forecasting. arXiv:2212.12794 (published in Science, Nov 2023).
its forecasts support better severe event prediction, including tropical cyclones, atmospheric rivers, and extreme temperatures
-
News report Independent author
Heikkilä, M. (2023). Google DeepMind's weather AI can forecast extreme weather faster and more accurately. MIT Technology Review, 14 Nov 2023.
In September, GraphCast accurately predicted that Hurricane Lee would make landfall in Nova Scotia nine days in advance, says Rémi Lam, a staff research scientist at Google DeepMind.
Traditional weather forecasting models pinpointed the hurricane to Nova Scotia only six days in advance.
-
Press release Developer
Google DeepMind (2023). GraphCast: AI model for faster and more accurate global weather forecasting. DeepMind blog, 14 Nov 2023.
a live version of our publicly available GraphCast model, deployed on the ECMWF website, accurately predicted about nine days in advance that Hurricane Lee would make landfall in Nova Scotia
-
Press release Deployer
Rabier, F. (2023). The AIFS is launched (editorial). ECMWF Newsletter No. 177, Autumn 2023.
We have made these systems available on ECMWF's public charts pages, based on our own initial conditions.