AI Weather Forecasting
summary
A data-driven nowcast system uses UNET encoder-decoder models trained on a petabyte multimodal dataset—geostationary satellite imagery at 1 km per pixel every minute, radar at 1 km/5 min, NWP context at 2–7 km, and ground stations every 10 min—to forecast precipitation, cloud evolution and solar radiation. Training runs on distributed GPU HPC, using PyTorch Lightning and MLflow, achieving 1–4 km spatial and 1–2 min temporal resolution and claims actionable outputs such as predicting a 30% PV drop 50 minutes ahead; project funded by EuroHPC Fortissimo Plus.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:26Nowcasting importance
- 2:30Physics-based model limits
- 3:56How AI weather works
- 6:28AI model data sources
- 9:47High-performance computing and MLOps
- 11:14Energy sector impact
- 13:25Grid stability with renewables
- 14:35AI evolutionary step
- 15:40Future of AI weather
- 16:45Conclusion
related talks
25mWeather Forecasting: An Application of Artificial Intelligence, Cloud, IOT &...
09mGen AI Workflows for Climate Risk Modeling in Food Systems
49mModeling our future: Advancing climate research with AI
07mFine-Grained Air Pollution Inference with Mobile Sensing Systems: A Weather-Related Deep ...
01h 16mPredicting Weather Patterns using Time Series ML
32mPredicting Grid Faults Before They Happen: AI for Weather-Driven Grid Resilience
10mWhy Your Forecasting Transformer Isn’t Working (And How To Fix It...
28mKubernetes Meets Climate Science: Building Large-scale Feature Detection From Climate Data Records