Researchers from the APEC Climate Center (APCC) and the Korea Institute of Science and Technology (KIST) have announced a breakthrough in predicting the Madden-Julian Oscillation (MJO), a major atmospheric phenomenon that exerts a profound influence on global weather and climate.

By leveraging advanced deep learning technology, the research team successfully developed a new framework that significantly improves MJO forecasting accuracy.

The study, led by Dr. Miae Kim of the APCC, was recently published in the prestigious journal Geophysical Research Letters under the title, "Multi-Scale Decomposition for Skillful All-Season MJO Prediction With Deep Learning".

The MJO is a massive convective system that propagates through the tropics, often triggering extreme weather events such as hurricanes and floods across the globe. While existing AI models primarily focused on isolated MJO signals, this new research introduces a "multi-scale decomposition" approach. By teaching the AI to analyze both the MJO signals and the underlying "background climate conditions"—such as seasonal and interannual variations—the model provides a more comprehensive view of the atmosphere.

Key findings of the study include:

Extended Predictability: The model achieves skillful MJO predictions up to 26 days in the boreal winter and 29 days in the boreal summer.

Enhanced Long-Range Accuracy: The research reveals that background climate data becomes increasingly crucial for the accuracy of long-term forecasts.

All-Season Applicability: Unlike previous models limited to specific seasons, this unified framework remains stable and effective throughout the entire year.

This achievement is expected to significantly advance global weather predictability and disaster preparedness. The study was conducted as part of the Research and Development Program supported by the Korea Meteorological Administration.

- Original Paper: https://doi.org/10.1029/2025GL117981