Industries GUIDE
AI in Weather Forecasting
GraphCast and GenCast are machine-learning weather-forecasting systems trained on historical atmospheric data.
Listen to this guideAI-narrated · ~1 min
On this page1 min read
Overview
GraphCast produces deterministic forecasts; GenCast generates probabilistic ensembles.
Deep Dive
GraphCast uses a graph neural network to predict global weather states at six-hour steps, trained on ERA5 reanalysis. The Science paper reports forecast-skill comparisons across variables and lead times against operational systems. GenCast is a probabilistic model that generates ensembles and is evaluated on medium-range forecasts and extreme events. The results are scoped to the paper’s data, baselines and verification metrics; they are not a guarantee for every region, variable or extreme event.
Real-World Implementation
GraphCast predicts atmospheric variables forward in six-hour increments in the published setup.
ERA5 reanalysis supplies historical fields used in GraphCast training.
GenCast samples multiple plausible trajectories to represent forecast uncertainty.
Published evaluations compare model forecasts with specified operational baselines and weather variables.
The Future of AI in Weather Forecasting
Use current operational forecasts for time-sensitive decisions and treat research benchmarks as evidence about tested cases. Compare forecasts by variable, lead time, region and event type. The papers do not establish that data-driven systems universally outperform physics-based forecasting across all conditions.
Putting it into practice
Why it matters
- Context and rules. Industry context determines whether AI ideas survive contact with reality.
- Quality control. Domain constraints influence acceptable error rates and oversight models.
- Build choices. Successful deployments align technical capability with frontline workflows.
Watch out for
- Regulatory requirements can invalidate otherwise strong prototypes.
- Historical data may encode bias that harms specific communities.
- Legacy systems can create integration bottlenecks and hidden costs.
Next steps
- Involve domain experts from problem framing to evaluation.
- Design audit trails and documentation before launch.
- Validate compliance and safety obligations early.
- Roll out in phases with clear stop and rollback criteria.
Keep Exploring
Free newsletter
Keep up with AI in 3 minutes a day
One short email each weekday with the three AI stories that actually matter. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the AI in Weather Forecasting quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Frequently asked questions
What is AI in Weather Forecasting?
GraphCast and GenCast are machine-learning weather-forecasting systems trained on historical atmospheric data. GraphCast produces deterministic forecasts; GenCast generates probabilistic ensembles.
What dataset do leading AI weather models like GraphCast primarily train on?
GraphCast and similar models are trained on ECMWF's ERA5 reanalysis, which provides decades of consistent hourly global atmospheric data.
How does an AI model like GraphCast fundamentally differ from traditional numerical weather prediction?
NWP numerically solves fluid-dynamics equations, while AI models pattern-match learned relationships from historical data.
How does GraphCast produce a 10-day forecast from a 6-hour prediction step?
GraphCast predicts 6 hours ahead, then feeds that prediction back as input, repeating about 40 times to reach 10 days.
Roughly how long does the GraphCast paper report for generating a 10-day global forecast on a single Google Cloud TPU v4?
The GraphCast paper reports a 10-day forecast in under one minute on a single Google Cloud TPU v4 device; this is a paper-specific hardware and model runtime, not a general guarantee for other implementations.
What result did the GraphCast paper report across its 1,380 verification targets?
The GraphCast paper reports stronger performance than the operational deterministic system on 90% of 1,380 verification targets; the figure is specific to the paper’s variables, leads and evaluation setup.
Keep learning
Related guides
More guides picked for this topic