AI has storms in its sights
Neural networks are cheap and fast at predicting storm tracks, and are becoming more accurate too. But don't dismiss traditional weather models yet.
AI is getting better at predicting hurricane tracks than traditional computer models, an advance that could give people more time to get out of the way of deadly storms.
Why it matters: Over time, cheaper AI models could more accurately predict catastrophic weather and help reduce the human and financial costs of disasters.
The details:
- Weather is difficult to predict because so many dynamic factors shape a storm’s direction and intensity. Running those calculations also requires enormous computing power.
- AI can quickly scan massive databases of weather observations, identify patterns and produce forecasts with speed and accuracy.
- Google DeepMind’s GraphCast predicted that Hurricane Beryl would strike the Houston area in July, outperforming established models that had projected landfall in Mexico.
- GraphCast beat the European Centre for Medium-Range Weather Forecasts’ leading model more than 90% of the time, DeepMind scientist Rémi Lam told The New York Times 🔒.
- Don’t dismiss traditional forecasting models just yet. For now, they’ll work alongside AI.
The local angle: LSU has been using AI to provide real-time storm surge forecasts. Its Coastal Emergency Risks Assessment tool is used by more than 2,000 U.S. emergency management agencies.
The bottom line: AI can make cat memes. Its real promise is helping science solve problems that matter.
For hurricane forecasting, AI and traditional models will work side by side for now, giving people in South Louisiana better information about storm tracks, wind gusts and other threats—the kind of information that helps residents decide whether to get out of the way or buy extra water and snacks and ride it out at home.