Limit of Predictability of the Madden-Julian Oscillation as a Stochastically-Driven Chaotic Oscillator

Phenomenologically the tropical 20-90-day Madden-Julian oscillation (MJO) is known to exhibit multiscale characters in anomalous circulation and the coupled convective systems in space and time. It also interacts with, or has shown to be correlated with, a wide range of weather and climate regimes. In previous studies the dynamical predictability of the MJO has been estimated to be around ~10-30 days. In this work, a recently developed adaptive detrending algorithm was used to bandpass the MJO indices to the range of 27.5-82.5 days. A complexity measure called the scale-dependent Lyapunov exponent and a pseudo-ensemble approach were applied to characterize the dynamics underlying the bandpassed indices. Two distinct scaling behaviors were found. One suggested stochastic forcing, where predictability was rapidly lost in a time span of about 25 days. The other one suggesting a weakly chaotic behavior, also with a prediction time scale of about 25 days. These findings impose fundamental constraints to the simulation and prediction of the MJO.