Machine learning cracks six-week extreme weather forecasts
A machine learning tool improved weather forecasts to six weeks ahead. It boosted severe flood predictions by up to 150 percent.
Scientists published a breakthrough in Nature this week. A machine learning tool fixes errors in weather models and makes forecasts accurate six weeks into the future.
Current forecasts beyond two weeks are unreliable because small errors get worse over time. The new system won a 2025 global competition, beating 34 other teams and models from six major forecasting centers.
When applied to real-world floods from the past decade, the tool improved prediction accuracy by 76 to 92 percent. For severe floods that caused more than 100 deaths, it improved accuracy by 86 to 150 percent.
The system is open-source and runs without a supercomputer, making it accessible even in developing countries.
- Six weeks ahead
- Forecast range
- 86 to 150 per cent
- Severe flood accuracy boost
- 246 deaths, February 2021
- Texas storm deaths
- 34 competing teams
- Teams beaten
Why it mattersEarlier and more accurate warnings of extreme weather give communities time to evacuate, strengthen infrastructure and save lives. Better flood forecasts could prevent deaths and keep people in their homes.
AustraliaAustralia faces increasing droughts, floods and bushfires; more reliable six-week forecasts would help farmers plan planting, water managers allocate supplies, and emergency services prepare for disasters.
✓ Claims checked against the source and corrected before publish. checked 3 h ago
Open this story in InSnip →
