Microsoft Research's Probabilistic Bias Correction method for two-to-six-week weather forecasts

Microsoft Research says its Probabilistic Bias Correction (PBC) method improves weather forecasts two to six weeks ahead, placing first in a global competition, according to a company post on X. The method uses machine learning to correct systematic bias in existing weather models.

A Nature paper reports that PBC is more accurate than ECMWF's physics-based model on 98 percent of precipitation targets and 92 percent of temperature targets, and doubles the accuracy of ECMWF's AI forecasting system in the study. Microsoft Research New England senior principal researcher Lester Mackey says the tool can run on CPUs alone.