AI postprocessing model doubles subseasonal forecast accuracy

MIT climatologists Judah Cohen and Lester Mackey developed an AI system with postprocessing that doubled the accuracy of subseasonal weather forecasts for the two-to-six-week window. Three versions of the model were submitted to ECMWF, with results published in Nature.

• Forecast accuracy for the 2-6 week window doubled
• Postprocessing lets the model learn from its own errors
• The model also improves prediction of extreme weather events
• Results from three model versions published in Nature