Exploring the long-term changes in the Madden Julian Oscillation using machine learning.

Area of Science:

• Atmospheric Science
• Climate Science
• Data Science

Background:

• The Madden-Julian Oscillation (MJO) is the primary driver of subseasonal tropical variability.
• The Real-time Multivariate MJO (RMM) index, commonly used to represent the MJO, is restricted to the satellite era (post-1974).
• Previous extensions of the RMM index to the twentieth century achieved an 82.5% correspondence with satellite-era data.

Purpose of the Study:

• To reconstruct and improve the historical MJO index using machine learning techniques.
• To analyze long-term changes in MJO intensity, phase occurrences, and frequency from 1905-2015.
• To investigate the influence of climate variability and anthropogenic warming on MJO characteristics.

Main Methods:

• Utilized machine learning algorithms, specifically Support Vector Regressor (SVR) and Convolutional Neural Network (CNN).
• Applied these methods to sea level pressure (SLP) data from the NOAA twentieth century reanalysis, mirroring predictors used in prior studies.
• Analyzed reconstructed RMM indices to assess trends in MJO intensity, phase, and frequency.

Main Results:

• Achieved a significant improvement of up to 4% in historical MJO index reconstruction compared to previous methods.
• Identified a notable increasing trend in MJO intensity (22-27%) between 1905 and 2015.
• Observed multidecadal variations in MJO phase occurrence and periodicity linked to the Pacific Decadal Oscillation (PDO).

Conclusions:

• Machine learning offers a superior method for reconstructing historical MJO indices.
• The MJO has intensified over the past century, with changes in its behavior influenced by both natural variability (PDO) and anthropogenic climate change.