Predictability of the Tropical Intraseasonal Oscillation based on Statistical Models: Sensitivity to ISO Indices

The predictability of the tropical intraseasonal oscillation (ISO) is investigated based on empirical models for the period 1979-2013. Two recently developed ISO indices including the bimodal-ISO index and OLR-based MJO index, developed by Kikuchi et al. (2012) and Kiladis et al. (2014), respectively, as well as the popular, real-time multivariate MJO (RMM) index are used to examine the sensitivity of the predictability to the choice of ISO index. Both of the new indices are based on seasonally-varying canonical ISO patterns of band-pass filtered OLR anomalies and, through elaboration, provide real-time information. Preliminary results based on a simple multilinear regression model indicate that the predictive skills of the ISO based on the new indices generally outperform that based on the RMM index throughout the year. The predictive skill based on the RMM index is relatively low of ~15 (12) days during boreal winter (summer), while it can be significantly extended by using band-pass filtered RMM time series as target, with the skill ~24 (25) days for 25-90 day band-pass filtered data. In contrast, the predictive skill of the new indices is ~30 (26) days. The results show that the new indices based solely on OLR are at least as useful as, if not more useful than, the conventional multivariable-based RMM index for predictability studies of the ISO. We plan to discuss whether the predictive skill can be even extended by employing a nonlinear stochastic model.