MIT and Microsoft researchers improve weather forecasts two to six weeks out

Researchers including MIT's Judah Cohen developed a machine-learning layer that corrects errors in existing forecasting models for the two-to-six-week range.

When applied to ECMWF's model, it doubled forecasting accuracy in that window and was the first to beat climatological averages there, according to a Nature paper.

The article says the method also improved extreme-weather prediction and works as an open-source add-on to any forecasting model.