Generate a medium-range forecast
Strong benchmark performance across many forecast targets; this score covers forecasting, not all atmospheric science.
SCIENCE UNDER AUTOMATION
AI in Weather forecasting: capabilities and evidence
12–24 hours of medium-range skill advantage
What has AI demonstrated?
12–24 hours of medium-range skill advantage
- Test
- AIFS Single 1.1.0; 730 twice-daily 2023 initializations; Northern Hemisphere 3–10-day anomaly correlation.
- Comparison
- Physics-based IFS; shared reference analyses and initialization dependencies.
- System and version
- AIFS Single 1.1.0 · 2026-06-01
- Evidence checked
- 2026-10-20
Gains compare versions and variables. Observation collection and physics-based initialization remain external dependencies.
What does the field require?
Routine medium-range forecasting
- Use supplied atmospheric analyses
Operational inputs are supplied by a separate observing and assimilation system.
AIFS Single 1.1.0 - Produce a global forecast
AIFS runs operational forecasts; evaluation covers 2023.
AIFS Single 1.1.0 - Evaluate forecast skill
Standard verification is available; specialist monitoring remains.
AIFS Single 1.1.0
Develop forecasts for difficult local events
- Resolve local extremes
Global skill does not establish reliable local extreme precipitation.
AIFS Single 1.1.0 - Adapt to new observing conditions
No matched autonomy result in this evidence set.
AIFS Single 1.1.0 - Validate and issue a consequential warning
Forecast skill does not test independent warning decisions.
AIFS Single 1.1.0
Research tasks
Can AI replace the whole workflow?
The results above measure specific tasks. The field assessment also requires completion of the connected workflow, including the requirements listed here. A field-wide replacement percentage has not been established.
Read the assessment method