Microsoft open-sources PBC, an ML tool to sharpen 2–6 week forecasts
Microsoft researchers published a method called Probabilistic Bias Correction (PBC) in the journal Nature, using machine learning to spot and correct systematic bias in existing weather models rather than building a new standalone model; the tool is now open-source. The study says PBC improved a leading physics-based model from the European Centre for Medium-Range Weather Forecasts (ECMWF) for 98 percent of precipitation and 92 percent of temperature targets, and doubled the accuracy of ECMWF's AI forecasting system in the paper's analysis. The researchers say PBC is lightweight and can run on CPUs, aiming to make AI-enhanced forecasting more accessible to resource-strapped agencies.
If lightweight bias correction holds up in operational use, under-resourced weather services could get more reliable two-to-six-week forecasts for farming and disaster planning at lower cost.
Microsoft has a new machine learning tool to improve weather forecasts
New open-source software could make AI-enhanced weather forecasting more affordable.
New open-source software could make AI-enhanced weather forecasting more affordable.
by Justine Calma
Oct 14, 2026, 4:05 PM UTC
#Microsoft#Microsoft Research#ECMWF#Nature#MIT#Rhiza Research#Kenya Meteorological Service Authority#Gates Foundation#Lester Mackey#Genevieve Flaspohler