AI Weather Forecasting Pre-Training

1. Which of the following makes weather forecasting challenging? (Choose all that apply) The atmosphere is chaotic Not enough observational coverage across the globe Measurement errors in data Complex processes like cloud formation

2. Match each forecast type to its lead time:

Nowcasting

Short-range

Medium-range

Subseasonal (S2S)

Seasonal

3. In an agricultural advisory, why is impact-based forecasting more useful than just reporting raw weather data? It’s faster to produce It focuses on what the weather means for people and decisions It removes uncertainty from forecasts It’s more accurate for all timescales

4. Give one example of how forecast lead time can change the possible actions a farmer can take. (Short answer)

5. Which of the following is NOT true about AI weather models? Requires high-quality, well-prepared data May need local validation Can work without any observational data Needs expert interpretation

6. Name one potential use of AI in weather forecasting besides generating a forecast. (Short answer)

7. Which is an example of a hybrid AI model that combines a dynamical core with machine learning? Pangu-Weather Graphcast NeuralGCM AIFS

8. Arrange these in the correct historical order of forecasting methods (1 = earliest):

Enter the number (1–4) next to each method to indicate order.

Numerical Weather Prediction (NWP): Ensemble forecasting: AI-based models:

9. What is the main reason ensemble forecasting became an important step in forecast evolution? It increases computational cost It removes the need for satellites It captures uncertainty by running multiple simulations It eliminates the need for human interpretation

10. True or False: AI models today are trained only on real-time observational data, not on reanalysis datasets like ERA5. True False

11. Which timescale is typically the hardest to predict because atmospheric memory fades before oceanic memory takes over? Short-range Medium-range Subseasonal Seasonal

12. Why do probabilistic forecasts become more useful than deterministic forecasts at longer lead times? They are faster to compute They provide a range of possible outcomes and their probabilities They guarantee accuracy They remove the need for data assimilation

13. Which of the following is NOT required to run an AI weather model effectively? High-quality data Understanding of model outputs GPUs or computing resources Eliminating all ensemble members

14. Before using an AI model in forecasting, which question is part of good benchmarking practice? “Do I have enough clean, relevant data?” “Can I run the model without electricity?” “Can I skip model validation?” “Will the model give me 100% certainty?”

15. Why is regional validation important before using AI forecasts for agricultural decisions? (Short answer)

16. What does ConOps stand for, and what is its main purpose in forecasting systems? (Short answer)

17. Why is post-processing (bias correction) important for AI model outputs? It makes it faster to produce forecasts It improves the alignment of outputs with local conditions It replaces ensemble forecasts entirely It removes the need for observational data

Pre-workshop Preparation Quiz