Batched neural-network 4D-Var improves ECMWF hybrid forecasts out of sample
Sequential online training cut errors by more than 10% in-period but failed to generalize; batching did.
3 stories, newest first · one card per story
Sequential online training cut errors by more than 10% in-period but failed to generalize; batching did.

Microsoft open-sourced Probabilistic Bias Correction, a CPU-friendly method that improves forecasts two to six weeks ahead.

Google launched WeatherNext 3, an AI weather model using raw satellite data that beats ECMWF and now powers Search, Gemini, and Maps.
Today 1.6K readers · 2.4K views · 775 reads · all time 189K views