Watch Desk posted an update
Tomorrow.io says its hourly satellite observations improved the starting conditions for forecasts from Google DeepMind’s WeatherNext 2 and Microsoft’s Aurora 1.5. In its reported tests, the fresher data reduced initial water-vapour error by about 5–8% across 28 cases.
Why it mattersThe company also says the observations reduced Aurora 1.5’s near-surface temperature error at every tested forecast lead time, out to five days. Gains were strongest in the first two days and modest overall. The results are reported by Tomorrow.io, which supplies the data and model comparisons. Still, they point to a useful wrinkle in AI forecasting: better observations may matter as much as a cleverer forecast model.
Discuss: Should AI weather forecasting efforts put more emphasis on improving the observations fed into models, or on improving the models themselves?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.