AI in Weather Forecasting Quiz

Answers and explanations appear as you take the quiz.

1. What dataset do leading AI weather models like GraphCast primarily train on?
   • The ERA5 reanalysis of roughly 40 years of global weather
   • Live satellite video feeds only
   • A single year of US radar data
   • Hand-labeled storm photographs
2. How does an AI model like GraphCast fundamentally differ from traditional numerical weather prediction?
   • It solves the physics equations faster on the same supercomputer
   • It learns patterns from past data instead of simulating atmospheric physics
   • It only forecasts temperature, nothing else
   • It requires no input data about the current atmosphere
3. How does GraphCast produce a 10-day forecast from a 6-hour prediction step?
   • It runs 40 separate independent models
   • It interpolates linearly between today and day 10
   • It feeds its own output back in repeatedly (autoregressively)
   • It averages ten different satellite images
4. Roughly how long does the GraphCast paper report for generating a 10-day global forecast on a single Google Cloud TPU v4?
   • About 6 hours
   • Two full days
   • One week
   • Under a minute
5. What result did the GraphCast paper report across its 1,380 verification targets?
   • It outperformed the operational deterministic baseline on 90% of the evaluated targets
   • It outperformed on every target in every weather situation
   • It matched the baseline on exactly half of all targets
   • It was evaluated only on one tropical cyclone
6. What does the GenCast model add compared to a single deterministic forecast?
   • Higher-resolution satellite imagery
   • Probabilistic ensemble forecasts that quantify uncertainty
   • Automatic news reports
   • A replacement for all radar systems
7. What spatial representation does GraphCast use for atmospheric variables?
   • A sequence of text-only tokens
   • A single unconnected pixel at each latitude
   • A high-resolution multi-scale mesh processed with graph neural networks
   • A fixed lookup table with no learned graph
8. What forecast horizon and time interval did GenCast report in its paper?
   • A 1-day regional forecast at 1-minute steps
   • A 6-hour forecast at 24-hour steps
   • A 30-day climate average with no time steps
   • A 15-day global forecast at 12-hour steps