Weather Predictions and AI Forecasting

Overview

• Discussion of Hurricane Lee's trajectory and impact (September 11-16, 2023).
• Traditional weather forecasts predicted storm's path, but AI models provided earlier predictions.
• Highlight the importance of AI in meteorology, especially during extreme weather events.

Data Collection for Weather Forecasting

• Snapshot of Earth's Atmosphere: Collecting data from satellites, weather stations, and buoys.
  • Measurements include cloud photos, temperature, pressure, wind speed, and humidity.
• 3D Model Creation: Data processed to form a 3D grid of the atmosphere.
  • Supercomputers are used for complex calculations (150 trillion calculations per forecast).
  • Major supercomputers operated by the European Center for Medium-Range Weather Forecasts and the National Weather Service.

Local Forecasting Adjustments

• Meteorologists refine global models using local expertise (e.g., geographic features).
• Ensemble Forecasting: Producing multiple forecasts to measure uncertainty.
  • If many forecasts are similar, there's higher certainty; if varied, lower certainty.

Limitations and Challenges

• The initial 3D grid never perfectly replicates reality due to data gaps.
• Forecasts become blurrier with time, hence the need for multiple forecasts.

Advances in AI Weather Forecasting

• AI's Role in Meteorology: AI models trained on extensive datasets (e.g., ERA5) outperform traditional models in some areas.
  • Capable of generating predictions quickly (e.g., Huawei's Pangu model can forecast a week in 1.4 seconds).
• Current AI models still rely on observation data from traditional weather centers.

Strengths of AI Models

• AI models have enhanced ensemble forecasting capabilities, predicting 1,000+ outcomes vs. traditional 50.
• Potential for better measurement of uncertainty and rare weather events.
• Can provide more detailed forecasts further into the future, beneficial for emergency management.

Ongoing Challenges and Human Element

• AI models still need improvement for predicting hurricane intensity.
• Despite their advancements, human meteorologists remain crucial for communicating uncertainty and making decisions.
• AI forecasting remains experimental, but some centers are publishing AI forecasts alongside traditional ones.

Conclusion

• The future of weather forecasting may involve a combination of AI and traditional physics-based models, enhancing our preparedness for weather events.