Statistical-dynamical prediction of the Madden–Julian oscillation using NCEP Climate Forecast System (CFS)


  • Kyong-Hwan Seo

    Corresponding author
    1. Division of Earth Environmental System, Pusan National University, Busan, Korea
    • Division of Earth Environmental System, Pusan National University, Busan 609-735, Korea.
    Search for more papers by this author


The predictive performance of the Madden–Julian Oscillation (MJO) in the National Centers for Environmental Prediction's (NCEP's) operational coupled, Climate Forecast System (CFS) model is assessed and statistical-dynamical hybrid models are developed based on this numerical model to improve forecasting skill. By projecting ENSO-removed variables onto the principal patterns of the MJO convection and upper- and lower-level circulations, MJO-related signals in the dynamical model forecasts are extracted. It is found that the coupled model exhibits useful predictability out to 2 and 3 pentads when the initial MJO convection is located over the Maritime Continent and the Indian Ocean, respectively. The first hybrid model (HYBRID1) developed is a lagged multiple linear regression scheme, where the previous values of the first two leading principal components (PCs) from the coupled model are used as predictors along with the latest observed PC values. The other hybrid model (HYBRID2) is constructed to fit the predicted PC time series from the coupled model to the observed PCs at each lead time using 13 years of training data fields. HYBRID2 does not produce persistent improvement in forecast skill compared to CFS, whereas for HYBRID1, the resulting correlation coefficient and root mean square (RMS) skills are considerably enhanced due to the incorporation of lagged correlation information between the two PCs. The skill improvement of HYBRID1 ranges from 15 to 50% relative to the CFS forecast and the improvement in the correlation skill is greater than the RMS error correction. When the MJO convection is located initially over the Maritime Continent during the northern summer, the CFS forecast skill is as small as the persistence forecast, whereas HYBRID1 demonstrates the greatest skill improvement. During ENSO winters, the forecast skill is enhanced by ∼30% by HYBRID1. In particular, the skill improvement in La Niña years is noticeably greater than in El Niño years. Copyright © 2009 Royal Meteorological Society