Weather AI: Not a Revolution, Just Faster Machine Learning
Machine learning models forecast faster but still face old limitations.
Deep Dive
AI in weather and climate science refers to machine learning (ML), not generative models. Companies including Google, Nvidia, Huawei, and Microsoft have developed initial models that could compare favorably to current forecast models. The European Centre for Medium-Range Weather Forecasts (ECMWF) put its first machine-learning-based model into service in February 202
Key Points
- AI in weather modeling uses machine learning (e.g., neural networks), not generative LLMs, trained on reanalysis data.
- ECMWF's AIFS model went operational in February 2025, running alongside the traditional IFS for faster forecasts.
- ML models are computationally efficient but lack interpretability and can fail on out-of-distribution weather events.
Why It Matters
Professionals should see AI in weather as an efficiency boost, not a replacement for physics-based forecasting.