The 3–4-Week MJO Prediction Skill in a GFDL Coupled Model

Abstract Based on a new version of the Geophysical Fluid Dynamics Laboratory (GFDL) coupled model, the Madden–Julian oscillation (MJO) prediction skill in boreal wintertime (November–April) is evaluated by analyzing 11 years (2003–13) of hindcast experiments. The initial conditions are obtained by applying a simple nudging technique toward observations. Using the real-time multivariate MJO (RMM) index as a predictand, it is demonstrated that the MJO prediction skill can reach out to 27 days before the anomaly correlation coefficient (ACC) decreases to 0.5. The MJO forecast skill also shows relatively larger contrasts between target strong and weak cases (32 versus 7 days) than between initially strong and weak cases (29 versus 24 days). Meanwhile, a strong dependence on target phases is found, as opposed to relative skill independence from different initial phases. The MJO prediction skill is also shown to be about 29 days during the Dynamics of the MJO/Cooperative Indian Ocean Experiment on Intraseasonal Variability in Year 2011 (DYNAMO/CINDY) field campaign period. This model’s potential predictability, the upper bound of prediction skill, extends out to 42 days, revealing a considerable unutilized predictability and a great potential for improving current MJO prediction.