Low-Frequency Variability in the Real-Time Multivariate MJO Index: Real or Artificial?

The real-time multivariate Madden-Julian oscillation (MJO) (RMM) index has now been widely applied as a standard in operational subseasonal prediction and monitoring. Its calculation procedures involve the extraction of major intraseasonal variability (ISV) by subtracting the prior 120-day mean. However, this study uncovers that such a real-time strategy artificially creates unwanted low-frequency variability (LFVartificial) that might cause nonnegligible influences on the RMM amplitude and phase. Compared to the real LFV, the LFVartificial explains more (-70% in boreal summer) of the residual LFV (LFVresidual) in the RMM index. It occupies 33% of all days that the LFVresidual explains more than one-half of total RMM amplitude, 19% that the LFV contribution exceeds ISV, and 10% that the LFVartificial-associated RMM am-plitude surpasses 0.8. The RMM-defined MJO is obscured by the LFVresidual in such a way that the eastward-propagating mode is stronger and bigger with a slower phase speed, as compared with the true MJO derived from the 20-100-day filtered data. The interference effects of LFVresidual on the MJO might be particularly strong when the background state is changing rap-idly with time. However, these issues can be well avoided when one chooses to remove the centered 120-day mean, as evidenced by the largely reduced three percentages (17%, 8%, and 1%) mentioned above in the so-derived index. These results give us a reminder that more attention should be paid to monitoring or predicting an MJO using the RMM index in a rapidly changing low-frequency background or in the boreal summer. SIGNIFICANCE STATEMENT: The real-time multivariate MJO (RMM) index has been widely applied in the mon-itoring and prediction of the MJO, the major tropical intraseasonal variability influencing global weather and climate. Using observational analysis, we reveal that there exist such scenarios (-16%) when large-amplitude RMM indices do not represent a strong MJO, mainly due to the obscuring effect of residual, while largely artificial, low-frequency vari-ability introduced by the RMM calculation procedures. This finding is of great significance as it informs the research community that serious caution should be given when relating large RMM amplitude to the MJO, especially in a condi-tion when the low-frequency background state is rapidly changing with time or in the boreal summer.

Automated summary

The real-time multivariate Madden-Julian oscillation (MJO) (RMM) index has been widely used in operational subseasonal prediction and monitoring. However, this study reveals that the real-time strategy used to calculate this index artificially introduces unwanted low-frequency variability (LFVartificial), which can have significant impacts on the amplitude and phase of the MJO. These interference effects are particularly strong when the background state is changing rapidly with time or during the boreal summer.