Google WeatherNext AI beats standard models at cyclone forecasts, adds a day of warning
nordicinst · x · 2026-09-10
A Nature paper shows Google's WeatherNext AI model outperforms existing systems at cyclone forecasting, delivering three-day hurricane/typhoon forecasts as accurate as previous two-day predictions — effectively giving residents an extra day of warning for extreme storms.
AI sidesteps the compute bottleneck of numerical weather prediction: instead of detailed physics-based calculations (dating back to Lewis Fry Richardson, whose hand-computed daily forecast once took six weeks), it learns patterns from training data. AI systems now rival conventional models at several forecast scales, especially longer ranges, part of a gradual shift gaining momentum over the past year.
More from Models
- User mocks OpenAI's 'math genius' models after apparent blunder — willcb · 2026-09-10
- DeepSeek releases Harness v0.1.5 with Agent Teams, co-trained with V4.1 Flash — teortaxesTex · 2026-09-10
- iCoder-27B: An AI Agent Autonomously Built a 27B Coding Model at SJTU, DP Technology & NUS — jiqizhixin · 2026-09-10
- Next Gemini Pro model reportedly gets 262k single output token limit, up from 65k — Able-Line2683 · 2026-09-10
- Switched from Claude to Codex: higher 5-hour limits, but weekly cap drains in 2 days — Additional-You9968 · 2026-09-10
- API relay silently routes Pro users to Flash, breaking months-old workflows — teortaxesTex · 2026-09-10