Madden-Julian Oscillation Prediction and Climate Variability

The Madden-Julian Oscillation (MJO) is the leading mode of tropical intraseasonal variability, characterised by large-scale eastward-propagating convective anomalies with a period of 30–60 days. Its interaction with monsoon systems, mid-latitude weather patterns and tropical cyclones confers a pivotal role in modulating rainfall, temperature extremes and atmospheric circulation across the globe. Despite decades of dynamical modelling, forecast skill often falls short of the theoretical limit owing to systematic model biases, inadequate representation of moisture processes and imperfect initialisation of three-dimensional humidity structures. Recent advances have focused on blending dynamical and statistical approaches, employing machine-learning techniques to identify key predictability sources, correct model error and extract quasi-periodic signals. Improved ensemble strategies, enhanced coupling with ocean and land surfaces, and deeper understanding of MJO teleconnections with large-scale modes such as ENSO are gradually extending useful skill in the sub-seasonal to seasonal (S2S) window. This progress holds promise for more reliable rainfall and temperature forecasts, offering tangible benefits for agriculture, water resource management and disaster preparedness in vulnerable regions.

Research from Nature Portfolio

Innovative use of deep learning has enabled the extraction of quasi-periodic signals beyond the MJO itself to enhance S2S forecasts of surface temperature over Asia. By incorporating information from subtropical and polar jet streams into a long short-term memory (LSTM) network, researchers have optimised ensemble weights and improved operational forecast skill for minimum 2 m air temperature. Meanwhile, a bias-correction framework has been developed to blend dynamical MJO forecasts with observations via a neural network, reducing amplitude and phase errors by up to 90 % and 77 % over four-week leads. This approach demonstrates that deep-learning correction can substantially elevate multi-model MJO prediction skill, particularly for events originating in the Indian Ocean and traversing the Maritime Continent.

Madden-Julian Oscillation Prediction and Climate Variability publication trend

The graph below shows the total number of articles in madden-julian oscillation prediction and climate variability across all publications each year (not limited to Nature Index journals).

Technical terms

Madden-Julian Oscillation (MJO): A planetary-scale mode of tropical variability characterised by eastward-propagating convective and circulation anomalies with a 30–60 day period.

Sub-seasonal to seasonal (S2S): The forecast lead time ranging from two weeks to several months, bridging weather prediction and climate projection.

Real-time Multivariate MJO (RMM) index: A diagnostic index combining principal components of zonal wind and outgoing longwave radiation to quantify MJO amplitude and phase.

Convolutional Neural Network (CNN): A machine-learning architecture that uses convolutional layers to extract spatial features from gridded climate data for prediction tasks.

Explainable Artificial Intelligence (XAI): Techniques that interpret and quantify the contribution of input variables in complex machine-learning models, improving transparency and physical insight.

References

1. Taking advantage of quasi-periodic signals for S2S operational forecast from a perspective of deep learning. Scientific Reports (2023).
2. Deep learning for bias correction of MJO prediction. Nature Communications (2021).
3. Deep learning reveals moisture as the primary predictability source of MJO. npj Climate and Atmospheric Science (2024).
4. Impacts of humidity initialization on MJO prediction: A study in an operational sub-seasonal to seasonal system. Atmospheric Research (2023).
5. Using Simple, Explainable Neural Networks to Predict the Madden‐Julian Oscillation. Journal of Advances in Modeling Earth Systems (2022).