Overview
- A peer-reviewed study by the University of Geneva and Karlsruhe Institute of Technology finds AI weather models beat traditional tools in routine forecasts but make larger errors on record-breaking heat and wind.
- They say AI is bound by its 1979–2017 training data, which keeps forecasts near past extremes, while physics-based models can represent never-before-seen states.
- The study advises against using current AI systems on their own for early warnings or disaster response until they improve.
- Operational forecasters in Colorado report AI is speeding their work through faster runs and much larger ensembles, and some now consult AI first for day-to-day guidance.
- Event tools are emerging as well, with OpenSnow citing a two-minute update system trained on decades of tornado reports and live radar and lightning to flag near-term thunderstorm risks.