Microsoft and MIT researchers develop AI to improve long-term weather forecasting

Researchers from MIT and Microsoft have developed a new artificial intelligence system that doubles the accuracy of weather forecasts in the two-to-six-week range. This period, often referred to as the “predictability desert,” has historically been difficult to forecast due to atmospheric chaos and the influence of slow-moving variables like El Niño.

Published in the journal Nature, the methodology—known as Probabilistic Bias Correction (PBC)—does not attempt to build entirely new weather models from scratch. Instead, it uses machine learning to identify and correct systematic errors in existing forecasting systems. By comparing previous errors with historical data, the AI adjusts subsequent predictions to account for biases, such as those caused by global warming.

This open-source approach is designed to be computationally efficient, allowing meteorological organizations with limited resources to access more reliable long-term forecasts using conventional processors.